Financial leverage is much easier to understand than many people believe it to be.

Basic balance sheet math

Every business has a balance sheet, which contrasts its assets (valuable things it owns) against liabilities (valuable things it owes to other people). The difference between assets and liabilities is equity.

Financial businesses will frequently have non-financial assets and liabilities. Ignore those for the sake of simplicity. Ignore the nice building, the computers, the payroll due on Friday for work which has already been completed. Focus just on the financial assets and liabilities, things like “mortgages our bank owns” (asset) and “deposits from customers” (liability).

Leverage is the ratio of your liabilities to your equity. Simple division. Fourth grade math. If you have $110 million in assets and $100 million in liabilities you, by subtraction, have $10 million in equity against your $100 million in liabilities. You are said to be levered 10:1.

Why leverage as a concept is useful

Finance is in the business of teleporting value over time and space. This is an enormously useful business to be in, but it requires taking various risks. Leverage provides a quick way to quantify how much risk a financial business is taking on.

This is easiest to understand in the banking example. Imagine the simplest possible toy version of a bank, which offers only two products: checking accounts and mortgages. The bank’s primary source of value as a business is maturity transformation: it teleports value over time to allow someone whose checking account currently has much less in it than the cost of a house to buy a house anyway and gain the benefits of living in it. It does this by putting a bit of the bank’s own capital at risk and levering that capital using the deposits in the checking accounts.

This bank is exposed to a variety of risks. One is that, because people can ask for money from their checking accounts at any time, impecunious operation of the bank could cause the value of the mortgages to be impaired just a tiny little bit at a time when people need most of the money in their checking accounts. This has killed many, many banks.

(Values of mortgages sometimes decline both for easy-to-understand reasons like the homeowner suddenly not wanting to pay it back and for reasons which are striking to many non-specialists, like prevailing interest rates increasing. This is a long story, which we will not tell today. Back to leverage.)

How much impairment to assets can the bank take before its equity is exhausted? If you know the leverage ratio, you can mental math out an approximation trivially. Add one to the leverage ratio. Take the reciprocal. Bam, that’s how much of a decline in the value of the assets wipes out the equity.

A 10:1 levered firm is wiped out if assets lose 1/11th (~9.09%) of their value.

Managing leverage of one’s customers

Many financial firms are levered as a matter of course; it is required to do the work. Interestingly, their customers are also often levered. A homeowner who has just made their 20% down payment is levered 5:1.

Financial firms frequently have to manage the leverage of their customers, because just like a financial firm gets riskier as the leverage ratio goes up, so do their customer accounts. Past a certain point, the financial firm will take painful actions to delever you.

Consider the example of margin accounts at stock brokerages. Google costs about $100 a share these days, so you could buy about 10 shares for $1,000 at your brokerage of choice.

Your stock brokerage, though, is willing to offer you leverage on your assets. In return for a fee (and to gain your business, because this is considered a high-saliency feature for the customers of brokerages), they will lend you money against your assets, allowing you to buy more Google than you had cash for. They might allow you 2:1 leverage when you buy stock: your $1,000 buys 20 shares now.

Why would you want that? Because you like Google, and want more exposure to it than you can afford. In future worlds where Google successfully sells ads and cancels chat apps, you'd make even more money that way.

So: The brokerage lent you money (where it comes from is a wonderful topic for another day) to match your money, and then you sent all that money to the person who previously owned those 20 shares, and now they are all yours.

But you now owe $1,000 to your brokerage, and are 1:1 levered. And this is fine… for now.

In some unfortunate future universes, perhaps one in which Google runs out of chat apps to cancel, the price of Google drops by more than 50%. In this case, the value of your stock is now below your debt to the brokerage. Your equity has been vaporized.

But let's not go immediately to that calamitous case. Googlers are, even now, working on more chat apps to refill the cancelable chat app inventory. Perf is just around the corner.

Say the price of Google is now $80. It only took a moderate hit.

Now your brokerage has a few choices, and none of them are fun, because someone has lost money and that can never be made fun. The value destruction has already happened, principally at Google. The results of that destruction are being communicated to you, because you chose to be a part-owner of their business, partially using other people's money.

One thing your brokerage could do is to ask you nicely to de-lever. You’re not 1:1 levered, which is relatively safe. You’ve got $1,600 in assets against $1,000 in debt, so $600 in equity, so ~1.67:1 levered.

To bring that leverage ratio down, the brokerage could potentially issue a margin call. This is anachronistic lingo (you, if you are reading this and do not work in finance, are unlikely to actually get a telephone call) to require you to either deposit more cash or sell stock. Either of these will decrease your leverage ratio.

If you sell $400 worth of stock, you now have $1,200 in assets against $600 in debt. This is 1:1 leverage.

If you deposit $400 of shares, you now have $2,000 in assets against $1,000 in debt. You’re back to 1:1 leverage.

If you deposit $400 and pay back the loan, you have $1,600 in assets against 600 in debt. You’re now 0.6:1 levered; even safer than 1:1.

Why the discrepancy in the impact of $400? Because leverage measures risk, and depositing $400 in shares dilutes your risk by adding assets that are themselves an incremental source of risk. Repaying the loan, on the other hand, strictly decreases your risk.

Mismanaging customer leverage

Why do brokerages care about customer leverage? Because if they do not promptly de-leverage customers, the margin loans are unlikely to get repaid. The risk of the stock decline has, effectively, been transferred from the customer to the broker. The losses it is incurring are eating into the equity of the brokerage now.

In principle, brokerages can pursue customers for bad margin debt. In practice, this is rarely going to restore them to whole, particularly not for retail customers, because the financial picture of someone whose brokerage account has lost all value is rarely rosy elsewhere. Brokerages take substantial efforts to avoid losing money on margin accounts.

In addition to the capitalistic reason to do this, their regulators are acutely aware of the risk that overextended customers pose to the brokerage. You can sort of think of society as being the broker’s broker: if we do not manage your leverage, because you do not manage your customer’s leverage, your losses eat into our equity. So we are supportive of you taking a carefully measured amount of risk, because risk makes your business possible, but will be cross if you take an impecunious amount and then stick us with the losses.

Brokerages really do lose money on margin accounts sometimes! It’s relatively rare. Large losses usually have a complicated story. But since they happen, brokerages spend a lot of effort carefully managing their balance sheets such that the hit does not impair their solvency.

Decentralized finance innovations in leverage management

I was recently playing around with a decentralized finance (DeFi) protocol. I am notoriously a crypto skeptic and do not perceive there to be much value in DeFi, but a friend won me over with this argument:

You believe Tether is a fraud. You want to short them, but do not trust cryptocurrency exchanges. So short them using DeFi. In worlds where you are right about Tether, you make money and will feel vindicated. In worlds where you are wrong about Tether, you will feel appropriately chastened, but you think basically no worlds like that exist so what do you care. And in worlds where the DeFi protocol blows up due to any of a long list of causes, you will get a great story out of it.

So let me tell you a story about Solend this week.

Solend would be described by many people as operating a decentralized protocol. I would describe it as an API with an ops team attached, but whatever, not germane to the following.

Solend lets you deposit assets that are on the Solana slow database, earn interest, and withdraw assets that other people have deposited back to the Solana slow database, paying interest for the privilege.

This allows you to do things like e.g. lever up on buying the Solana token.

In my case, the specific use was “Deposit USDC, a stablecoin that I have some degree of trust for. Borrow Tether. Use a separate DeFi program to sell those Tether for USDC. Wait for the price of Tether to decline, buy the now cheaper Tether with the USDC, repay the Tether-denominated loan to Solend, profit according to the degree of decline less the net interest paid over the interval.”

Now a funny thing happened while I was patiently waiting for Tether to blow up: global systemic crypto calamity. Somehow, this was not caused by Tether. Dang it. They can’t get anything right.

That quickly became relevant to my interests, because there was another borrower on Solend. Let’s call him the whale. (His identity is knowable but not germane to this story.)

The whale had, previously, deposited a lot of Solana tokens (SOL) to Solend and withdrawn a lot of USDC stablecoins, back when the price of Solana was much higher. At that point, he was conservatively leveraged. What did he do with the stablecoins? Who cares; they’re money.

The price of SOL declined precipitously recently. Solend, the computer program, and Solend, the team, had planned for this eventuality, and the program started doing what it was designed to do.

The design was “Make some of the whale’s debts assumable by third parties, if and only if they ask for that to happen, in return for an incentive payment of the whale’s collateral.” The idea was that fast-acting bots, operating in a tight loop and incentivized by the juicy 5% discount to the market price of SOL, would quickly eat the collateral, deleveraging the account and bringing it back to a healthy state. (Small detail: a portion of that 5% would get paid by the bot back to the Solend team, partly as compensation for their efforts and partially to help build a buffer against bad debt.)

That… didn’t happen.

Why not? Well, the whale is much, much bigger than the bots, and it was out of their price-range to liquidate in one go. Instead, they’d have to chip away at it, buying up to the USDC they were willing to spend with their little bot brains, getting attractively-priced SOL, selling the SOL for even more USDC, and repeating.

And with the price of SOL in freefall, some of the bots (or operators of the bots) couldn’t find liquidity at attractive prices for people wanting to buy their SOL. This let them unable to loop.

So the price of SOL kept falling, and the whale went from conservatively leveraged to dangerously leveraged to insolvent.

Many people in DeFi believe that, if this happens, you can basically just walk away from your collateral and that is the end of it. I take no position on that point of view or the righteousness of it; it’s just very different from the default expectation for traditional finance.

Given that the whale was assumed likely to just walk away, the hole in the balance sheet is now someone else’s problem, and it grew as SOL declined.

How did this present to users of the protocol? Well, Solend was owed USDC and it had SOL. But one thing people like doing when the price of an asset is in freefall is borrowing it to short it, so other users borrowed alllll the whale’s SOL and sold it… somewhere. Now the protocol had no SOL, and claim against USDC from a whale that was in another ocean.

And so if you liquidated the whale now, you would not get SOL for your USDC. You would get a wrapped token—an extremely technologically sophisticated IOU—instead, entitling you to some future SOL from Solend whenever they had the SOL to give you and you asked for it.

So Solend (the people, not the computer program) did some active liquidity management, by tweaking parameters to their program, causing the offered interest rate for SOL deposits to be 2,400% to 2,600%.

In the midst of the market contagion, even 2,600% was insufficient to attract lots of SOL, for a long list of market microstructure reasons. Many of the bots were off and their creators went to bed, worrying that Solend would have an even larger hole in it when they woke up.

Who pays for that hole? The default answer is “It’s the people who deposited other assets into the same pool but will not be able to withdrawal them, even if they repaid their loans, because their collateral are… swimming with a whale, in another ocean.” A frequent addendum to this answer in DeFi is “And since people would be really annoyed in that circumstance, there exists a side pocket of money—the project treasury—which would step in to make depositors less-damaged in this circumstance. Of course, if the treasury is exhausted... well, that would be bad.” (Ask Solend for their actual resolution here.)

But as the price of SOL kept going down, the hole kept getting deeper, and some combination of depositors and the project treasury was hemorrhaging money. So they kept doing things to entice liquidators. (And they banned people from further borrowings of any assets, because... you can predict the logic, right.)

A funny thing happens when you create incentive systems. For example, you might induce a technically proficient financial columnist to spend a few hours playing a very boring video game for money. Perhaps he knows enough about slow databases to understand how the words "epoch" and "unstaking" relate to each other and to know what this implied for the difficulty level of the video game in the early Tokyo evening.

That video game, in lieu of being able to quickly program a bot, was taking a trivial amount of USDC, waiting for more SOL to be deposited in response to the 2,600% interest rate offered, and quickly liquidate as much of the whale’s debt as the program would let me. The video game crashed about 70% of the time I tried to do this, but I pressed on.

Then, I’d use my 5%-cheaper-than-market-price-recently SOL and use a different video game to convert them into USDC, tolerating both a high chance of the video game crashing and a very real risk of slippage in the price of SOL between when the slow database first detected me buying it and the slow database would allow me to sell it. (The Solana slow database is a fast database as far as slow databases go, but due to market conditions, both it and some other necessary technological infrastructure were having a very bad day. What sounds like "click two buttons in a web app with a barely perceptible lag between them" took between 3 and 12 minutes per iteration.)

And so I round-tripped that transaction, twenty two times, each time with more money used to liquidate. Then people in other time zones woke up, looked at market conditions, and turned their bots back on. Once that happened, AI stole my button-pushing job.

Now you might naively think that means I roughly tripled my starting capital (1.05 ^ 22), but the combination of the slippage and the fees I was paying to the designers of the video games meant it was actually more like a 50% gain. I made more money than I typically would playing poker for 2 hours while probably at worse risk, and would probably choose to play Dyson Sphere Program in the future, but Outsourced APAC DeFi Defaulting Accounts Risk Manager did push my buttons for a few hours.

Well, I pushed its buttons, at any rate.

What can be learned from this experience?

The source of the payment for my button pushing was the built-in incentive to deleverage (via forced liquidation) the defaulting account. The programmers chose to build in that incentive so that someone, and they had (sensibly) no idea it would be me or operate any way like how I chose to operate, would save them the downside risk of a defaulting borrower’s collateral continuing to fail.

Was this the only thing happening on Solend yesterday? Oh no, feel free to read their official spaces for details. I’ve successfully mined the anecdote for my teachable moment now.

Am I a DeFi fan as a result of this experience? Not so much, no, but at least in reinventing things from first principles the folks seem to have created a very concrete demonstration of what leverage management (and mismanagement) looks like.

Hopefully that helps demystify leverage and leverage management to you. You will, in both the near future and long after that, read a lot about leverage, and how it blew someone up. And hopefully you will remember that it is not some dark art only accessible to the high priests of finance.

And now you know enough to understand what I'm saying with an otherwise opaque statement like "Tether is 35:1 levered on risky assets during market contagion."

Although this was obliquely phrased, the big two were poor decisionmaking on behalf of Russia (invading Ukraine) and Europe generally (being fundamentally unserious about power policy), leading to ripples that would affect the price of inputs to many participants in the Japanese grid. This would have the knock-on effect of potentially bringing down the grid when salarymen get home from work and turn on their air conditioners.

“Of course, if Russia starts a war during a hot year then there will be a risk of power outages in Tokyo; that’s just science” is perhaps not the most straightforward conclusion to draw, so it’s useful to go into a bit of background into how electricity markets work. This will, I promise, help explain why Softbank wanted me to install a demand response app targeting specifically my 5 to 8 PM air conditioner use.

A few disclaimers to get out of the way: the physics of electricity generation are the same everywhere but the market structures which govern pricing it are not, sometimes due to path-dependent infrastructural reasons and sometimes due to regulatory regimes. The energy markets are a very deep topic; you’ll probably want to get your week-to-week updates from someone like Doomberg (an anonymous group of energy consultants, whose work often feels like smuggling details about hidden power structures into the public view) rather than from me.

I’m going to repeatedly use two common acronyms that it is possible some readers might not understand. For their benefit, “capital expenditures” (CapEx) means long-term investments into things which will generally keep producing over a multi-year expected lifetime, and “operating expenses” (OpEx) means the day-to-day costs of running a business, like payroll or gas for a taxi. Some businesses are CapEx-intensive but OpEx-light, some are vice versa, and (relevantly) the energy business includes a lot which require both huge CapEx and high ongoing OpEx.

Electricity is ubiquitous, and other polite fictions

The first thing to understand about the energy markets is that everything written about them is a lie. Or, more to the point, everything is a simplified model, with some discontinuity between the model and behavior in the real world. This is because energy markets are almost as complex as all commerce combined and, unlike counterparties in commerce, physics tolerates infinite detail and doesn’t care about your feelings.

One useful mental model: there is a Grid™ and it is everywhere, like the Internet is. Suppliers of electricity and consumers of electricity are hooked up to the grid. You push or pull your electrons as you would like to and someone, perhaps the power company, sends you a bill every month.

There is no Grid™, there are multiple grids, there are ways of transmitting power over distance that are not grids, and physics actually doesn't let you ignore distance at will. The topography of your local energy market likely includes some power sources which do not hook up to a grid; they’re frequently built next to industrial users outside cities because transmission lines are costly (in CapEx and constantly lost electricity) and because industrial users have better ability to plan than society does. Power planning turns out to be an extremely deep topic, and if you have unified insight into future needs at a minute-by-minute level and do not have to worry about your neighbors, you can get tremendous savings on power.

This colocation of generation and demand leads to an interesting phenomenon called “stranded” generation, when the industrial user colocated with power generation ceases operation. The power asset is often still functional (the CapEx, heavy in most generation businesses, has been paid and if you put in OpEx you will get electricity out) but useless without a local-ish consumer of a large block of electricity.

This is much remarked upon by, of all people, Bitcoin enthusiasts, because it turns out that Bitcoin miners are almost an ideal consumer for electricity at fire-sale prices, in that you can site them just about anywhere, they have by-design absolutely insatiable demand for electricity at all margins including exactly the amount you have stranded, and capital is available to construct them in hundred million dollar increments.

(Bitcoin miners are in the business of bidding electronics depreciation and electricity consumption against each other to win a tournament in generating random numbers with mystical properties. From this tournament arises a mediocre transaction processing network and a speculative asset. You get more of the speculative asset if you’re the global highest bidder in electricity used in generating random numbers and, hence, the huge overlap in professional Bitcoin miners and people with deep expertise in quirks of the power generation market. For more on this topic from someone who thinks more highly of it, see generally the work of Nic Carter.)

Electricity supplied and demanded are necessarily equal

Electricity is extremely expensive to store relative to its value. This is a limitation of present battery technology but is closely related to the physics of the universe we live in. Most ways we have of generating power in the first place are only economical because they are creative cheats around physical reality, for example by burning millions of years of work by underpaid bacteria to generate a few seconds of heat and light.

As a consequence of this, the supermajority of all electricity is consumed approximately contemporaneously with being generated. Electricity moves through the less complicated bits of power infrastructure at about 0.6c, or about 110,000 miles per second. The physical processes required to generate electricity frequently require hours of time to spin up or spin down. These are fundamentally incompatible timescales and yet we somehow make it work anyway.

Demand for electricity fluctuates over the day and over the year

Speaking generally, most readers live somewhere where there is a grid operator who coordinates energy supply by various producers to meet instantaneous power needs. Those literally change on a millisecond to millisecond basis as light switches get flicked, computers wake from sleep, and aluminum smelters ramp up during work cycles. The grid operators have demand prediction down to a science (and extremely consequential statistics exercise). This is enormously positive for you since tiny mispredictions generate brownouts.

Tepco maintains the following graph for Tokyo, current as of… the rest of today. (I love the Internet.)

On it, you can see extremely common behavior: energy demand peaks in the late afternoon (early evening, for workdays). The marginal loads include everything from restaurant kitchens to washing machines but, on weekdays, are mostly residential air conditioning or heating.

If you draw a line around 20,000 MW, you’ll see that Tokyo has a certain level of work per second required to sustain civilization. That is called “base load.” The 18,000 MW gap between that and peak capacity is dynamic load. (I’ll note that you will generally see power generation figures quoted in MWh or kWh in English sources and, fascinatingly to me, the choice to quote in increments of 10 MW here is because it was easiest for the localization engineering team going from units of 10,000 kW, which is convenient to express in the Japanese language.)

The power markets have substantial complexity and human effort in making sure a diverse selection of sources with different physical and economic characteristics can instantly clear markets at all levels between these.

The power generation mix

You might naturally expect that, if you’ve spent billions of dollars on CapEx for a power plant, you would want it running continuously or as close to it as possible. This is desirable only for certain types of power plants. It turns out that the spin up and spin down times of various plants are different, and the marginal cost of their operation (fuel, mostly) and (increasingly) marginal carbon impact determine who gets lit up when to dynamically manage supply.

Speaking of lighting up: some forms of power generation are less amenable to being dynamically managed by humans. Despite millenia of research, while we have become extremely precise at predicting when the sun comes out, causing it to rise earlier or set later remains elusive. In addition to the daily cycle, which changes with the seasons, day-to-day and minute-to-minute changes in local weather greatly effect the amount of energy thrown off by solar installations.

But returning to base load power: nuclear generators operate virtually continuously (feckless political decisions notwithstanding). This is both due to the extremely low marginal fuel cost of nuclear power and the physical properties of, perhaps surprisingly, water. You can turn off a nuclear reaction in on the order of seconds but the steam produced by your reactor, which turns the turbines to generate power, will continue being hot (due to its own energy and remaining energy in the system) for a few hours.

Other base load sources include coal-fired plants. Here again, inclusion in the base load layer is a result of a combination of physics and economics. Physically, you have a some choice as to whether you burn more coal in the near future but relatively little choice as to whether coal continues to burn from the near past. Coal is generally speaking cheap per kWh and available in sustained quantity in the supply chain; if your local grid counts on it, it very literally counts on it.

Geothermal and hydroelectric are also base load where they are available; while hydroelectric plants have some degree of sensitivity to the weather, both of these sources are broadly predictable and relatively constant over the clock and calendar.

The layer after base load is intermediate sources. These include solar and wind, which are unavailable to cover base load, and also generally some fossil fuels which are amenable to being started and stopped relatively quickly but have somewhat poorer margin characteristics. Natural gas commonly ends up in this layer. (The price of liquid natural gas (LNG) in Japan is sensitive to geopolitics, critical for the Japanese economy, and is currently the subject of much commentary that you can read elsewhere.)

Finally, there are “peaker” plants, which spend most of their time off and economically justify themselves by keeping society running for a few hours every day and, in some cases, only a few weeks out of every year. Peak generation is almost exclusively fossil fuels, generally the ones with the worst margin characteristics but the best minute-to-minute ability to change generation. Gasoline is one example; although gas electricity generation looks a bit different than an internal combustion engine, they roughly share the property of stopping on a proverbial dime when you shut off the flow of liquid into the controlled explosion.

Demand response and virtual power plants

So why am I getting push notifications from Softbank to strongly consider decreasing my energy usage between 5 PM and 8 PM today? It’s part of a strategy called demand response.

Because adding new MW to the grid costs a lot of CapEx, peaks which exceed rated capacity are extremely bad news. Those peaks include many individual decisions which could, in principle, be shifted. Demand response sets up economic, communications, and infrastructure layers to encourage some consumers of electricity to either shift their loads off peak or to curtail (cease using electricity) during peaks.

This has lead to an innovation called “virtual power plants.” Someone, probably in financial engineering, realized that there is an economic equivalency between a peaker plant, with lots of CapEx and emissions associated with it, and a well-balanced combination of software and contracts.

While there are many ways of implementing them, the typical way is to sign up a roster of customers, purchase a commitment to ongoing power from other operators to service those customers, then sell the grid an option to buy back its own power at peak times only.

Get that? Maybe it’s clearer in numbers: “My customers need 40 MW which you can count on, and I will commit to purchasing that level of service, but I can make due with only 30 MW given five minutes of notice. This is economically equivalent to having an invisible 10 MW power plant powered by clean-burning unicorn flatulence. You should include it in your supply management strategy. Now, pay me for my entrepreneurial energies in finding you 10 MW of power available at the worst possible times.”

How does a virtual power plant actually find those 10 MW? By contracting with customers to suggest they move power usage and, if necessary at the margin, curtailing their use (economically, contractually, or in some cases literally by having software installed on-site to remotely shut down their machines).

In Softbank’s case, with regards to this residential user, demand response is a bit more optional; they give about 24-48 hours notice of when they expect peak demand (based on the forecast from the grid operator, based on the forecast of the weather among other factors). Software calculates what typical usage for a family like mine would be over that ~2 hour window and offers me an incentive payment for each kWh that I use less than my predicted use. There is no penalty for going over, and I’m only paid if I (trivially, one-tap) indicate my intention to respond to the call for more power. Softbank pulls the stats off of my smart meter, calculates an incentive payment the following day, and immediately pushes it to me. (Most electricity companies would do a database update and credit my account against future usage but, because Softbank invested in a payments company which offers zero marginal costs sends, they’ll happily instantly transfer me 17 yen.)

A quick aside about meter reading

Smart meters are one of the most economically important hardware technologies that you’ve never thought of. Thirty years ago, your monthly power consumption was likely read by a (human) “meter reader” coming to your building, who would read your meter’s current number, subtract last month’s number, and eventually set the power company up to bill you for the difference. This occasions what industry calls “truck rolls”, which is the activity where an expensive human gets in an expensive piece of movable capital for a few hours. There is a known, high marginal cost for truck rolls and ongoing CapEx for the fixed infrastructure which makes a committed number of them possible.

A smart meter is simply a dumb meter plus a microchip plus a communications channel back to the mothership. They cut down on truck rolls by more than 99%. Upgrading the fleet to smart readers cost an incredible amount of money and paid for itself very, very quickly. (Some readers of this column still have legacy meters, and it would be an interesting dive into switching costs, regulation, and labor economics into who does and why.)

We still had to balance supply and demand prior to smart meters, and indeed prior to computers, and that this worked at all is a triumph of civilization. But smart meters are a much, much better way to do things than the status quo ante.

Why this sudden interest in power?

Power is one of the pervasive bits of infrastructure that civilization sits atop of and largely doesn't think about. This year seems like a particularly poor year to not have good intuitions about how it actually works.

We'll return to this subject later, including a fascinating angle on it that only a payments geek could enjoy: how subscription billing for utilities works and the financial consequences of it.

The humble bank branch. You’ve probably been in one before. You might not realize that that forward deployment of banking capital represented a multi-million dollar commitment to almost a century of service. For an institution which touches most people, particularly in the middle class and above, they’re poorly understood.

As it turns out, my father Jim McKenzie spent most of his career doing real estate acquisitions in the Chicago area. I remembered many of the recurring characters in his dinner table stories to be bank branches, so I had a chat with him about it over my family vacation. He then replied with the sort of fatherly enthusiasm only possible when your son finally takes up the family business of negotiating curb cuts with the Illinois Department of Transportation for suburban Chicago bank branches.

The below is my understanding of his experience, with a bit of color from me on the inside-the-financial-industry point of view. All mistakes are, as always, my own. I work for Stripe, which does not endorse what I write for myself.

Bank branches as privately funded public financial infrastructure

A recurring theme of this column is that banks are privately funded public infrastructure. Bank branches continue this trend.

The typical bank branch, and for here we’ll concern ourselves mostly with full-service bank branches of large institutions since they dominate the amount of new branches in most years, is a complex outsourced real estate transaction followed by bespoke development, funded by both the bank and external investors. Let’s break this transaction into parts.

First, after the scale of community banks, branches are no longer one-off projects. Banks think in terms of branch networks, where the value of the network is greater than the sum of the parts, and then work backwards from that to a spreadsheet with rows for individual branches. The magic jargon word for this in real estate is footprint; this covers both the actual sites, their serviced areas, and the demographics and business opportunities in the relevant commute corridors. We’ll see those concepts return in a moment.

So when a large bank wants to expand its footprint in, without loss of generality, Chicago, they have exactly two options. One is to buy an existing bank with an existing branch footprint and local customer base, and indeed the large national banks grew to their current size by buying many smaller banks over the years. The second one is to kick off multiple parallel real estate acquisition and development projects.

Banks, like most large firms with brick-and-mortar presences, have real estate departments but lack the internal infrastructure to entirely acquire and develop properties. Few bank employees swing hammers or wire structures for a living. Instead, they work with local real estate developers to either identify sites or get sites pitched to the bank.

The ideal branch site

“The ideal branch is in an upper middle-class neighborhood on a corner of Main Street and Main Street”, said my dad. (About which, more later.)

Why on a corner? In commercial real estate (CRE), the world is split into destination locations—places you’d leave your house or office to go to—and Everywhere Else. Think of the last few times you made the decision to leave your house and what caused you to leave. Medical checkup? Grocery shopping? Fancy dinner with the spouse at that restaurant with rave reviews on Yelp? All of those are destinations.

Bank branches are not destinations. Like Starbucks and cell phone shops, they rely on capturing your day-to-day custom when you’re out and about. In the U.S., that mostly means being maximally accessible by cars. (In Japan, and other places with different transit behavior, bank branches are among the most likely user for large parcels directly adjacent to hub train stations, with smaller light branches and ATM-only locations being deployed close to far-from-station workplaces.)

The dominant understanding in the financial industry is that customers choose their main bank due to convenience, with convenience being defined as “close to home, close to work, or (ideally) on the commute corridor between the two.” If you work in commercial real estate, you develop a fascination with the SimCity-like logic of commute corridors and which corners collect what type of worker as the exit which neighborhoods. (A corner has twice as many roads fronting it as a non-corner property and therefore sees many approximately twice as many cars and many, many more slow cars than a non-corner on either of the two streets.)

Banks can substantiate, with voluminous data, that banking relationships are very sticky. I covered this previously and won't repeat it here.

The economics of banking suggest spending generously to capture sticky business, particularly sticky business that will grow. This is one reason they spend tens of billions of dollars a year on advertising. The branch network is an extension of advertising, sometimes extremely literally; there are branches which exist for no purpose other than “had a city-approved large billboard adjacent to a thoroughfare with hundreds of thousands of desirable commuters daily.” The bank built the branch and staffed it with about half a dozen professionals as the cost of being able to put their logo on the billboard for half a century.

Branches are expensive

In general, bank branches are new construction, because banks have functional requirements that few other businesses have.

You might think “Aha, they have security concerns and need a vault. How many commercial spaces have drill-proof rooms in them?”, and if you think that you are creative but not well-calibrated. Physical security drives surprisingly few decisions about bank branches; vaults have gone from the defining feature of a bank to a bit of an anachronism.

The things which a bank needs which are difficult to retrofit into a space that was previously a dentist office are numerous. One example is “a drive-through window with, ideally, a specialized pneumatic system for conveying cash and documents between the customer and the teller”, for many deployments. (You’d think that ATMs would have largely obviated the utility of these, but the businesses which build and service them remain extremely healthy.)

Another example is parking. Banks have extremely weird behaviors by the standards of parking engineers; the typical user behavior is to stop in for only a few minutes but the behavior the bank wants to optimize for, new account opening, can take half an hour to several hours. Through what turns out to be a simple result of queuing theory, bank branches end up with a lot of parking that appears mostly underutilized almost all of the time, and this is close to optimal.

Most retail users don't have this characteristic and so don't calibrate their parking in this fashion. A sitdown restauraunt will want their parking almost saturated on Saturday night (or cars are outcompeting tables). A dry cleaner with hundreds of weekly customers might only need two or three parking spaces.

A physical location that has been a bank branch is poorly calibrated for other users and vice versa.

Bank branches are expensive; the construction of a full-service branch, though it will vary based on size and location, runs roughly $20 million. They also represent an enormous future commitment, because banks largely lease rather than own.

Financing branches

The typical bank branch sits on land the bank has leased from a real estate developer, and the most typical terms are a 20 year lease with a combined 75 year option. This means, decoding from CREease, that the bank is obligated to pay for 20 years of rent and then gets to extend the period it can use a site up to 55 years after that. There is substantial variability in contracts and terms beyond those two.

Why don’t banks own branches? In part, this is part of optimizing the bank’s capital stack. If banks owned branches outright, this would tend to increase the amount of not-directly-productive assets they require to operate a given level of business; this would need to be financed with the bank’s cost of capital. Whether CRE operators or banks have a lower cost of capital is an interesting deep dive which depends heavily on, among other things, marginal regulatory guidance on funding sources.

A larger reason is that, because banks intentionally site branches in neighborhoods that they think are on the up-and-up, and because real estate represents an essentially permanent revenue stream of rents, current owners of up-and-coming land do not generally want to sell it to banks at prices that the banks would accept. A lease represents a trade; the bank believes that this parcel is a great branch on a 20-to-75 year timescale but also that the current owner overvalues it over the 76th through infinite years.

As a result, when you dig into bank filings, you’ll often see the properties they actually own have been in the business for (sometimes) centuries, represent flagship locations like headquarters, or had interesting quirks in the deals which resulted in them coming under bank ownership. The typical branch you’d walk into in Chicago has the land, and more rarely the building, owned by a real estate firm you’ve never heard of.

Prices are prices and are difficult to generalize about, but the right ballpark to think of for the price of leasing a full-service bank branch in Chicago (and many American metropolitan areas) is about $1 million per year. This is on top of $20 million or so in build-out costs.

Value add in commercial real estate development

What value do real estate firms bring to the table? Because bank branches are economic infrastructure and because banks are incentivized to predict (and influence!) the future trajectory of neighborhoods, they are often deployed in advance of a neighborhood or site being “ready” to host a high-traffic commercial location. This implies a diverse range of skill sets.

One of them is micropolitical action. The politics of zoning in America would fill several books, and they are often intensely local. Banks lack the block-by-block context of characters at village meetings and key decisionmakers, particularly for areas which are newly coming into their footprint. They also are extremely sensitive to the dynamics of banks interacting directly with public servants to get their preferred commercial outcomes.

Thus developers end up doing this. Part of the political action is as simple as showing up to the process; real estate developers attend village meetings about zoning decisions more frequently than almost any other profession. It is, after all, literally their job.

Part is about making arguments to stakeholders about land use policy. “If you approved the addition of a bank at this corner, the village would get a steady stream of taxes guaranteed for 20 to 75 years, where many other commercial tenants might or might not survive a shorter lease. Do you have any infrastructure improvements you want guaranteed funding of?”

Part is outsourced project management. The product a bank is buying is a colorful candy shell to conduct the business of banking in; everything outside that shell is someone else’s job. A fascinating detail about this is that your corner location in prime real estate requires curb cuts.

A curb cut is authority granted to you by the owner of the road (often the state government) to make a physical change to your property and the road to allow customers access. Curb cuts alter the properties of traffic management at a block-by-block engineering level. More cuts along a particular road tends to decrease the rate at which traffic flows through it; this ripples through the network and is therefore often very intentionally designed. The (without loss of generality) Illinois Department of Transportation has engineers who are responsible for individually reviewing and potentially approving curb cuts.

These decisions are some mix of the deterministic engineer-applies-written-rules-designed-to-ensure-even-application-of-the-law and politics at its most concrete and raw. The engineer making a decision on a curb cut has a boss, who has a boss, who has a boss, who eventually depends on the good graces of various elected officials. A bank branch is dead on arrival—it cannot be made—without approved high-quality curb cuts. (What makes a high quality versus a low quality curb cut? It comes down to everything from whether the angle a customer would need to turn at is subjectively an easy one to do at speed to factors like “is the physical cut large enough to accommodate the typical vehicle for this branch’s customer base.” I love the fractal nature of detail here; every detail about the built world was a painstakingly made decision at some point.)

Sometimes getting a high quality curb cut means identifying political leaders in, without loss of generality, Springfield and engaging with lobbyists with the intent of having those leaders make a quick call to the Department of Transportation to suggest that a senior official have a quick call with an engineer and suggest they review plans again with a more… persuadable eye.

Phrased like this, the process can sometimes sound corrupt, and a lifetime of hearing about retail politics conducted in Chicago makes it difficult for me to entirely deny that. But for the record, sometimes lobbying means bringing local knowledge to the attention of people who are poised to make decisions based on it. For example, you could tell a political leader that a particular neighborhood with their constituency in it is underserved by the financial industry and that you are one curb cut away from improving that.

Bank siting is not a free choice

McDonalds has almost free reign under capitalism to site new McDonalds franchises at any corner they want to. Banks do not.

Because banks are expected from society to operate as public infrastructure in addition to operating as profit-seeking businesses, in return for privileges society grants to them that are not available to other businesses, regulators and lawmakers care a lot about where bank branches are and aren’t. Branches serve neighborhoods; neighborhoods without branches lack access to credit and other valuable banking services.

One formal lever for this is the Community Reinvestment Act, but less formal pressure from state or federal banking regulators also heavily informs siting decisions. If a large bank consistently applies only for permission to open branches in upper middle-class neighborhoods, which is often optimal for the bank, they will draw negative attention from their regulators.

The culture that is branch banking

Because access to bank branches brings cash management and credit availability into a community, society cares a great deal about whether that access is available. But how does one measure access?

One way, often used regarding so-called food deserts, is to drop pins on the map representing consequential retail locations, draw a circle around them, and count who is and is not within one of the circles using census data. This is fairly naive and therefore is much beloved of social scientists, who can get grad students to do it for them.

Some people in society have to actually be right about what they publish, and they have developed an important novel insight into human behavior: people cannot fly. And so banks long ago supplemented the radius method with painstaking work looking at road networks to see whether someone is within e.g. 1 mile of a bank branch using available and convenient transportation routes, and to calculate the expected catchment areas of candidate sites for new branches.

All models are wrong; some are useful. It turns out that this is a better approximation for user behavior, but people do not strictly optimize for distance in banking, even though industry often believed convenience trumped most other considerations.

A man named John Melaniphy, since deceased, did some groundbreaking work that industry would call user research and that academia might call ethnography. He and his team interviewed thousands of people about how they actually chose which bank branches to drive to.

They found out that there are truths evident on maps which distances do not full capture which influence customer behavior. One, extremely relevant in Chicagoland and having no rational explanation, is that users prefer not to drive through forest preserves on the way to their bank branch; they’ll go substantially out of their way to avoid mixing greenery with their money.

Another is that bank customers feel a strong sense of identification with neighborhoods. Bank branches are, by law and practice, open to the general public. While there are occasionally residency restrictions in whether they’ll open an account for you, typically a bank branch will allow you to open an account if you live anywhere inside the bank’s footprint, at a county-by-county or metro-by-metro level of resolution. I, for example, easily opened a bank account in San Francisco with an address in Chicago just by walking into the local instance of a large national bank; that wasn’t even odd for them.

Bank customers do not feel this is true. They will avoid using a bank branch a block away from their house or workplace if it crosses a neighborhood boundary. People familiar with the microgeography of Chicago can probably guess individual roads where there are somewhat rational and somewhat unfortunate reasons for that reluctance, but this turns out to generalize even between neighborhoods that are very similar in character/demographics/etc.

And of course, the community boundaries that live in people’s heads are not always the ones which are drawn on maps. This complicates Seeing Like A Financial Institution (or regulator), because bureaucratic processes often assume that the map is in fact the territory. One of the reasons large banks with national presences tap local real estate firms is specifically to understand available but opaque local knowledge.

What’s an example of this sort of knowledge? To use one from my current stomping grounds in Tokyo, Aobadai and Nakameguro are adjacent to each other but since Aobadai doesn’t have a train station people living in it frequently identify with Nakameguro (which does) but perhaps counterintuitively not with Daikanyama. Daikanyama station is almost the same distance but the hill up to it is a steep walk or bike ride. You would (mostly correctly) predict that the (generally well-off) people who live in Aobadai bank in Nakameguro and not in Daikanyama, but it would be difficult to guess that from most non-topographic maps of Tokyo.

Now imagine how fun this gets when you mix in factors like “Some customers in Chicago are acutely sensitive to what parish they are in”, “This community has substantial oral lore about a turf war between two gangs which no longer exist and people still avoid crossing that imaginary undeclared illicit boundary if they can avoid it”, and similar.

Does this even matter?

Many technically inclined people would say that this is interesting historical context but that banking happens on phones these days. The financial industry would, as discussed previously, fire back that many account services happen on phones (and they are enthusiastically in favor of transitioning more towards mobile and web channels) but that a large portion of their depositors and an even larger fraction of their deposited dollars are acutely sensitive to the availability of their local branch.

We’ll talk later about some of the fascinating hybridization of service which the industry has experimented with, which received a major accelerant when many branches were forced to temporarily close or limit services during the global pandemic.

This is counterintuitive and sounds like it is trying a bit too hard to be clever. You should believe it.

Crime rates are a policy choice

If you enjoy simulation games, you might be familiar with the mechanic where you click a button and some statistic in your civilization moves radically in response. In real life, cause and effect is more subtle, but this relationship exists, and there are (both historically and at this very moment) legal regimes which are radically different than your status quo, and which achieve(d) very different outcomes as a direct consequence of policy decisions.

A glib way to phrase this is that crime is a policy choice, both definitionally (you could simply agree something was not a crime anymore and bam, crime down) and, more interestingly, because crime responds directly to things which are within your control. Most of the world has taken most of the easy policy choices which have few tradeoffs available! But there are still arbitrarily severe options to control crime from where you are, from “increase the police budget” to “ban alcohol totally” to “implement an Orwellian dystopia.”

Fraud is a unique subset of crime which occurs, to a major degree, subject to the enforcement efforts of non-state actors. A commanding majority of all fraud which is stopped, detected, adjudicated, and even punished (!) gets those done to it by one or more private sector actors. And the private sector has, in this case, policy decisions to make, which, like the public sector’s decisions, balance the undesirability of fraud against the desirability of social goods such as an open society, easy access to services, and (not least!) making money.

Scoping down to payments fraud

To prevent this conversation from being painfully abstract, let’s scope it to one particular type of fraud against one particular type of actor: the bad guy steals a payment credential, like a credit card number, and uses it to extract valuable goods or services from a business. This is an extremely common fraud, costing the world something like $10 to $20 billion a year, and yet it is actually fairly constrained relative to all types of fraud.

This fraud is possible by design. The very best minds in government, the financial industry, the payments industry, and business have gotten together and decided that they want this fraud to be possible. That probably strikes you as an extraordinary claim, and yet it is true.

Before we get into the how, let’s get into the why.

Who pays for payments fraud?

Liability for payments fraud happens in a waterfall, established by a combination of regulation, contracts, and business practice. The specifics get complicated but, for ability to concretely visualize this, consider the case of consumer credit card users in the United States.

You might assume that, if a credit card is stolen/hacked and used by a bad actor to buy something, the cardholder would be liable. They will suffer the first loss, certainly, but society has decided by regulation (specifically, Regulation E) that that loss should flow to their financial institution, less a $50 I-can’t-believe-it’s-not-deductible. As a marketing decision, the U.S. financial industry virtually universally waives that $50.

The card issuer will, following the credit card brand’s rules (which developed in symbiosis with regulation), automatically seek recovery of the loss from the business’s payments processor. It will, similarly, automatically seek recovery of the loss from the business itself.

In the overwhelming majority of cases, that is where the waterfall ends. While insurance is available (both specialized chargeback insurance and general business insurance), overwhelmingly businesses simply absorb fraud costs in the same way that they absorb their office rent, staff salaries, and marketing expenses.

That $10 to $20 billion number we threw around earlier? This is what happens to it, in the ordinary course of business. This allocation of loss is mostly automatic, virtually never involves a court or lawyer, and only sometimes takes human effort at the margin at all.

Fraud as a necessary business expense

Pretend you are the newly hired Director of Fraud for Business, Inc. You know you are ultimately liable for most fraud that happens in this pattern. What target do you take to the CEO for how much fraud you should suffer?

Zero?! Do you think the Director of Marketing desires to spend zero on marketing!? That would be an objectively silly goal. They would clearly be fired and replaced with someone who understands marginal returns.

The marginal return of permitting fraud against you is plausibly greater than zero, and therefore, you should welcome greater than zero fraud. You can think of it as a necessary expense, just like rent or salary or advertising is. You can even write it off on your taxes. (Ask your accountant; businesses frequently misunderstand the rules here.)

The reason for this is that Directors of Fraud are aware that the policy choices available to them impact the user experience of fraudsters and legitimate users alike. They want to choose policies which balance the tradeoff of lowering fraud against the ease for legitimate users to transact.

Costs and benefits of policy choices around trust

Maybe the frame of talking about fraud predisposes people to view the space of choices here negatively. Here’s an equivalent function with different emotional valence: how much do you trust people, and under what circumstances?

All fraud is a) an abuse of trust causing b) monetary losses for the defrauded and c) monetary gain for the fraudster. You could zero fraud by never trusting anyone in any circumstance.

Trust, though, is an immensely socially useful technology. Human civilization has a fundamental limitation in that all humans can be trivially killed while sleeping. Huge portions of society’s efforts go toward establishing conditions where this trivial vulnerability virtually never gets exploited. God has, reportedly, closed all bug reports claiming that it is a feature and won’t be patched any time soon.

Anyhow, trust is also fundamental in commerce, where it’s a layered concept, with different people having different levels of trust in different situations. To increase trust generally tends to frontload the cost to generate that trust, and decrease transactional friction afterwards. You trust your accountant more than most regular employees, you trust your employees more than your customers, you trust your customers more than a person you’ve never met, etc.

This cost falls on both parties in a trust relationship. To employ an accountant, you (the business) need to identify and interview several prospective accountants and employ one winner for years, and you (the accountant) need to have spent years of your life to get a professional credential and then to have worked your entire career to demonstrate yourself worthy of trust. This is one reason why accountants are routinely trusted with the holiest-of-holies secrets of companies and governments.

Clearly, e-commerce would cease if, prior to buying a pair of sneakers online, you required someone to go to that degree of effort. You’d almost never lose a pair of sneakers to a fraudster again, but you’d also sell very few sneakers.

Making a customer of someone you’ve never met

The payments industry has to solve many foundational problems. One of the core ones is quickly bootstrapping a business over the decision to trust someone they’ve never met, enough to allow them to consume valuable goods and services, based on nothing more than a promise of future payment.

A promise! Mere words! Billions upon billions of dollars have been spent on marketing to make you think that a payment is more than a promise. It’s a lie, and it’s a lie we all choose to believe in part because it’s a vastly more effective model to run the world under than the truth is.

Businesses prefer attracting new customers to not attracting new customers, citation hopefully not needed. They have a choice as to how much friction they want that new customer to need to go through prior to being offered goods and services. Many businesses have found that decreasing friction results in getting more new customers, who spend more, and who stick around for more transactions. (These are, incidentally, the “only three goals of marketing.”)

You could subject first-time customers (or even repeat customers), to an elaborate underwriting process, in part to increase your trust in them / decrease your perception of the risk that they would defraud you. You could, for example, ask them to give you a firm handshake as a condition of doing business.

The requirement for a firm handshake is, actually, an effective anti-fraud measure. The requirement that it happen face-to-face decreases the number of international professionalized fraud gangs which can target you, because they’re not physically close enough to shake your hand. Unfortunately, for the same reason, it also decreases how many customers you can sell to; most people don’t live within commuting distance of your retail presence.

Anti-fraud loops used in online commerce

You’ve probably had a shopping experience impacted by an anti-fraud loop, though you might not have recognized it as such. Ever been asked for billing address in addition to shipping address? That’s for AVS verification. There is an obvious user-experience hit there, and it’s quantifiable; removing fields from checkout forms increases conversion rates nearly as a rule. (Conversion rates are an industry term-of-art describing the percentage of prospects who successfully purchase something.)

Wonder why everyone under the sun wants you to have an account on their site? One major reason is that it gives customers a history that allows a business to direct more of its anti-fraud attention to (more risky) first-time users than (less risky) multi-year regular customers. Allowing guest checkouts is a business decision to accept more fraud (and less ability to market to the customer) in return for marginal sales.

Some of the savvier interventions operate in the background or don’t surface for all users. For example, you could imagine asking the purchasers of especially high-risk orders to first confirm possession of a phone number (via typing in a code you text them), or even to talk to a human in your fraud department before completing the transaction. Both of these are aimed at breaking the economics of scaled fraud; phone numbers and voice calls are expensive relative to synthetic identities and tend to leak information about fraud operations, which can further inform defenses.

We’ll talk about this some other time; risk scoring and marginal interventions is a fascinatingly deep topic.

Different businesses have different tolerance for fraud

Margins create margins. A business with high margins will, all else equal, tend to spend more on marketing and sales than a business with low margins; if they don’t, their competitors will “bid up” the cost of attracting customers out of their own fat margins.

Businesses with high margins also tend to be more accepting of payments fraud than businesses with low margins. Consider businesses which sell IP, like video game companies, streaming services, or SaaS. Because their margins are often 90%+, if you were to present them with a menu of strategies which traded off conversion rate and fraud rate, they’d maximize for conversion rates until fraud at the margin reached levels not seen in even the most corrupt places imaginable.

Businesses selling valuable resalable goods with much lower margins, such as Apple hardware or game consoles, have to be much more careful about who they transact with. When they’re offered a conceptual slider for who to do secondary transaction screening on, they screen more marginal orders. They accept painful tradeoffs like, “We’ll have a fraud department review every new order and hold every first-time order for shipping until we can talk to the purchaser.”

Between these two there exists a spectrum of fraud regimes, and this is broadly a good thing. Society gets to make choices, and here it is choosing through the activities of private agents. It is optimizing for how many resources to let leak to bad actors and much societal effort to burn on policing them versus how much low-friction commerce to enable by good actors. This is often missed in discussions of fraud; one reason it has increased over the past few decades is that legitimate commerce has exploded, as the world becomes richer and as barriers to commerce have come down.

This extends beyond payments

So hopefully you buy that Internet merchants can happily accept non-zero levels of fraud. This argument generalizes, and it has some important ethical considerations. We should, as a society, accept non-zero amounts of benefits fraud. We should accept non-zero amounts of cheating on taxes. You personally have benefited from the financial industry’s decision to not expend the maximum possible effort on defending against so-called identity theft.

These tradeoffs are often intensely difficult to pursue openly. Who wants to be known as the politician in favor of benefits fraud or the financial CEO who thinks they are not laundering enough money?

One of the interesting questions here is who gets to resolve tensions like this. Generally speaking, it will be private actors applying their own cost-benefits decisions. There is substantial space for regulations to help with cases, like identity theft, where actors can choose to spend other people’s risk budgets to maximize for their own interests.

If you have other fraud subtopics you’d love to cover, drop me a line.

Bank branches are sales offices

You probably mostly think of bank branches in terms of tellers, ATMs, and routine transactions. This is not why bank branches exist; it is an incidental offering of the bank to secure your business.

Bank branches exist to sell new accounts. They are sited to maximize new accounts and the value of those accounts. They are staffed to maximize new accounts and cross-sells to existing customers (which will often be called “relationships” at a bank). Everything down to the physical layout of branches and sometimes even the relative paucity of non-branch options ("channels" in the lingo, like phone or online banking) is designed to maximize new accounts.

You’d be surprised how few accounts they sell! The typical full-service bank branch costs about $20 million to put into service, with running costs near $1 million a year, and for this… it will generate about 300 new deposit accounts and a roughly similar number of loans. Banks are pretty discreet about these targets, but they appear in both survey data and can occasionally be backed out of public filings.

The lifecycle of consumer(-ish) bank accounts

Why are banks so solicitous about putting sales offices close to you? Because they expect that once they earn your business they will keep it for a very long time.

Churn rates for typical retail checking accounts cluster around 8%. This is in terms of relationships (counting by, typically, the household; banks have views of how income and gender mix which would not endear one to the modal San Franciscan professional). Churn rate is lower when measured in terms of dollar amount of deposits; the more you keep in checking (and, incidentally, the more products you use, like mortgages, credit cards, and the like) the less likely you are to churn in a given year. This means that the bank assumes that, in expectation, well-qualified customers are making a 10+ year commitment.

A commitment to what, exactly? A deposit account will allow a bank to make money via fees, net interest margin, and (if cross-sold with a debit card, which banks train their people to do very aggressively) interchange on the purchases a customer makes. Debit cards are especially important for Durbin-exempt banks; regulated debit card interchange is better than a kick in the teeth but only becomes material for the customers who swipe the most frequently.

Fees tend to be pretty uninteresting intellectually speaking and competition has crammed them down for most customers in the middle class. The “free checking” era largely did away with monthly fees in most markets, largely (and controversially) by replacing them with insufficient funds fees (“NSFs”) levied against people with unstable incomes who overdrew their accounts. This has been an ongoing target of regulatory scrutiny.

The dominant engine for profitability of deposit accounts is net interest margin. A bank’s advantage versus every other business in the economy, and the reason why they are the economy’s major source of credit, is the ability to leverage their capital via the use of borrowed money. Much of those borrowings are cheap or free to the bank, in the form of deposits. Net interest is the difference between the rates banks pay on their deposits and what they take in from their loan book. It fluctuates based on the interest rate environment and other factors, but if you were to estimate it at three plus or minus half a percent, you’d be in the right ballpark.

Note that 3% of a college student’s checking account annually is… not much. Branch banking is about building portfolios of accounts. Some of those accounts will be unprofitable or barely profitable. Some, like many college students, will grow over the years. But most of the retail deposit base is, structurally, older, wealthier depositors who are at or near retirement. In a real sense, the branch exists for them and your ability to use it is a courtesy. (We’ll talk more about branch siting in another issue, as the bank is constrained in how much it can lean into this reality.)

This is also the main reason why branches have not disappeared despite their immense cost and the feasibility of doing almost all bank transactions through other channels. The banks’ favorite customers like them a lot.

The unit economics of bank accounts are fascinating and relatively closely guarded. In broad strokes, the median consumer deposit relationship (across all deposit accounts) generates about $200 a year in revenue. The bank pays almost all of this back in servicing costs; even the stamp for mailed paper statements adds up when statements are monthly and margins are thin.

People are often extremely surprised that banks spend a lot on account servicing. The $20 million real estate project which exists for the pleasure of ~2,500 people has only the cheapest Bic pens and discount coffee! Clearly it's close to free!

An aside: A surprising portion of the intellectual effort of retail banking divisions is on getting more people to opt for paperless delivery. My favorite example of this is the very large Japanese bank which for various reasons thinks it needs to send a certain type of paperwork to customers and record a printed reply. It spent many millions of dollars inventing a system which would allow the bank to print the received reply on behalf of the customer and thereby save one of the two stamps needed for the process. Lest I be accused of making fun of Japan here, I will note that the United States Postal Service has an employed senior official whose only job is making sure the largest banks in the U.S. don’t go paperless-by-default.

But, again, retail banking is a star search model: the median relationship isn’t nearly as important to the bank as their top 10-20% of retail relationships are. Importantly, those top 10-20% are almost exclusively “the mass affluent”; they’re retirees, dual teacher families, software engineers, small business owners, and doctors, not real estate moguls or billionaires. The industry aggressively segments based on wealth and somewhere around $10 million or so a bank doesn’t want you talking to their regular branch employees anymore; you’ll deal with a different breed of commissioned sales professional who carefully presents as a different social class than most branch employees.

Small business banking

Most banks structure small business banking in the same part of the organization as consumer (retail) banking, due to their sales and service models being very similar. Small business branch banking is… largely not a very good experience. It basically can’t be, given the bank’s cost and revenue models.

They can’t afford to deliver most forms of specialized offerings throughout their branch network, and so almost all small businesses receive lowest common denominator service. Small business accounts, even ones which touch subjectively a large amount of money, are not very lucrative for banks to service.

The typical small business is thinly capitalized and pays out almost all of its revenue every month. This means a business with almost $10 million in annual turnover, which is about the cutoff for a bank to consider a business small, probably has an average balance of only $400,000. That works out to in the neighborhood of $1,000 a month of revenue for the bank. This scales down linearly with the business’ turnover; most small businesses have only 5-10% as much turnover.

Phrased like this it might surprise software people that small businesses are bankable at all; the typical SaaS founder would go apoplectic if someone suggested putting a constantly staffed office within about a mile of every customer to charge them $50 a month.

Larger banks treat small businesses largely as a necessary detail to attract wealthy retail depositors, because a large portion of wealthy people own or operate businesses. Two particular forms are extremely interesting to them: real estate, because mortgages remain a major contributor to their businesses, and medical professionals. (Note that the culture which is banking, and the opinions of regulators and other stakeholders, would strongly counsel keeping small business banking open even if there were no direct business case for it. This is not the only line of business for which this is true. Savings accounts for children are more a statement of values than a product.)

What's the commonality between real estate and medicine? Why court medical professionals, and not e.g. lawyers or gardeners? One reason is that medical professionals, as individuals, generally over-accumulate cash relative to other career paths. They are willing to either keep it in checking or invest it based on the recommendation of bank-based financial advisors.

The other reason is that, while the rate of medical professionals working in their own practices is declining, a medical practice with licensed professionals is a proverbial license to print money. That printer has equity value and is durable and resilient in a way that individual professional careers are not. Many professionals will “buy into” their practice or buy out the equity of a retiring partner; many banks make a speciality of writing large, relatively low-risk, collateralized loans to enable this.

All of these factors are superior to those present with more typical small business loans. Banks would originate almost no loans to acquire small businesses but for government programs like e.g. Small Business Administration loans, which effectively use the banks as interfaces to government financing. (The covid-era Paycheck Protection Program economy-wide fiscal airdrop was similarly executed through banks, and one of the reasons it was characterized as loans was to maintain sufficient plausible deniability that bank stakeholders could say “Yes, this is clearly banking business” with a straight face. Banks are, in addition to being private enterprises, a policy arm of the government; this is the thing I think crypto enthusiasts are rightest about.)

Staffing bank branches

If you’ve ever seen Mary Poppins, you might remember a famous scene (and wonderful ditty) set in a turn-of-the-20th-century British bank. It had dozens of employees conducting the quiet symphony of ledgering and deal-making, lead by a prim and proper expert who was a leader in his community.

This is not the reality of bank branches these days. As banks got more sophisticated about using information technology, rather than deploying all bank functions to substantially all branches, they largely centralized them and made them available to branches via telephone or computer applications. Many things you go into a branch to do will actually be done on your own telephone while a banker helpfully points out which buttons to push. This even includes account opening in many places! (There exists a truly excellent writeup of the history of this deskilling of the branch workforce for the U.K.; similar essays could be written about the United States or Japan, though Japan has lagged by a few years.)

While it varies based on market and whether the branch is “full service” or one of the newer concepts, a typical bank branch will have about six employees: one manager, one to two tellers, and the remainder bankers.

Tellers handle routine bank transactions. They’re almost entirely substitutable by ATMs (automated teller machines) but, in a graph which never fails to surprise people, absolute employment has increased since the ATM has rolled out. (This is largely due to an explosion in the number of branches offsetting a reduction in tellers per branch.)

Titles for bankers depend on the bank but could be “community banker”, “personal banker”, “private banker”, or similar. They are basically all commissioned sales representatives who are generally in a high-pressure sales environment, with the branch manager being a quota-driving line sales manager.  Bankers will, for historical reasons, handle higher complexity routine transactions (like e.g. wire transfers) than tellers will, but they are discouraged from spending their time on transactions except to the extent necessary to sell more accounts.

In the past, branch managers were far more akin to CEO of their branch, with substantial authority to influence underwriting decisions on loans or make accommodations for customers; this is largely on the wane. At most banks they are sales player-coaches with some vestigial customer service and regulatory functions. (You can still use a branch manager to achieve an escalation of a routine issue into a bank’s call tree, because they likely have a special number for that, but for anything over low single-digit thousand dollars you should just escalate immediately to paper addressed to someone the bank will trust to make consequential decisions.)

If you’ve ever wondered why you were ambushed upon walking in and directed to a banker for something you’d normally do at the teller window or ATM, sales quotas are why. Bankers have a script to read to see if your needs have changed recently. If they aren’t aggressive enough about sourcing business, from people walking in the door or from calling existing customers or new prospects, they will be replaced with someone who can meet their quotas.

This quick sketch of the realities of incentives at bank branches should suggest a good analogue for you: while a branch might have the professional vibe of a doctor’s office or be opposite the street from a fast food restaurant, the business it is closest to in character is actually a cell phone store. It’s a high pressure sales environment, with most of the sales professionals not exactly being industry experts, where substantially all of the money is made “on the backend” (over time) of service delivery lasting many years.

Will branches change?

Technically-inclined people have been confidently predicting the death of the bank branch since I used Prodigy to access the Internet. While covid likely accelerated the transformation, we’re likely to see not a vanishing but a continued gradual decline of branch networks and more experiments with different branch experiences.

Non-full-service branches are popping up in many places, sometimes as new builds and sometimes as shrinkflation of existing branches. These generally cut staffing down to three to four people, often partially by sharing a branch manager with other locations. It remains to be seen whether this is going to be a permanent part of the mix or whether they are expensive investments in educating marginal users how to sign up for online banking. Somewhat surprisingly to technologists, only about half of the deposit base has figured out online or mobile access.

Apps are, of course, getting better all the time. Covid was a real accelerant in particular to the ability of banks to do account opening online; this was once something of an advanced feature and has now become de rigueur at banks up and down the technical sophistication spectrum. We will inevitably see increasing pressure to move account opening online, even if “online” means “on your phone as you’re standing in a branch”; the marginal improvements in utilization really matter due to how expensive square feet and person-hours are relative to pixels.

Now that you’ve had a quick primer in branch banking, you can come back later for an extended dive into the real estate side of it. See you then.

This issue will be more U.S.-centric than I like, mostly because I understand the Federal Deposit Insurance Corporation, which runs the U.S. deposit insurance scheme, far better than any similar government agency. In broad outlines, I’d expect Japan, the UK, European Union countries, and similar to have similar mechanics, but can’t opine intelligently on them without doing more reading.

I'm going to talk fairly confidently about the fintech industry in a moment, and will repeat my usual disclaimer that I work at Stripe. Opinions in this space are mine alone and were not e.g. run past a lawyer for an accuracy check.

The covered peril

Every insurance policy has a notion of “covered perils”: it will pay out if and only if an event matching a limited description happens. Deposit insurance covers precisely one covered peril: the loss to depositors (1) caused by the failure (2) of an insured financial institution (3).

From this you can deduce a lot of things which are not covered perils:

Did you lose money on a bank-issued bond or equity? Not covered; your higher rate of return was specifically to compensate you for the risk you were taking. As we covered previously, your money protected depositors. Thank you for your service.

Did your bank eff up a given transaction with you, in such a way that you lost money but the bank is still open for business? Not covered; your recourse is with the bank, its regulators, or the legal system.

Did a business which has an account at an FDIC-insured institution, but is not itself an FDIC-insured institution, fail? Not covered. If their bank is still open for business, your princess is in another castle, and you’ll very likely have to follow a bankruptcy proceeding with interest.

This last one is a bit of a sticking point for financial technology firms, which experience a dilemma in building products which mimic some features of deposits. On the one hand, they want to message those products to the market as secure. On the other hand, they typically need to keep customers money in the banking system, generally at an FDIC-insured institution.

But they are not themselves FDIC-insured institutions.

A thing which some app marketing teams will say is that deposits are secure up to applicable limits, which may mollify customers but may not answer the question they truly have. The thing which is often mumbled in these situations is that customers’ money is insured against the failure of the underlying bank, whose risk of failure is known to be very remote, but not insured against the failure of the technology platform, whose risk of failure is almost always orders of magnitude larger than that of the bank.

This is a heavily facts-and-circumstances-dependent detail, and even fintech product teams and lawyers often fail to understand the nuanced difference in mechanics between various implementation options for deposit-adjacent products when exposed to various tail risk events. Regardless of those differences, which could fill volumes and occupy many billable bankruptcy attorney hours, you can round this to “The FDIC does not insure against the primary sources of risk to users of fintech products.”

Anatomy of a bank failure

But let’s talk about what deposit insurance does do. Bank failures are auto-catalyzing processes, similar to meltdowns of nuclear reactors. Deposit insurance is designed to make failures less likely and limit damage caused by them, through a variety of technical means. You could analogize it to some combination of monitoring staff, control rods (and technical measures for deploying them), and cleanup crew.

Most bank failures happen slowly and then quickly. Most banking crises happen sporadically along the periphery of the banking system and then suddenly everywhere all at once. Deposit insurance is designed to limit and contain the damage on the individual bank level to minimize the chance of failures cascading in a systemic fashion. (Making depositors whole is both a necessary prerequisite of this and, to a degree, a happy side effect of this core function.)

The thing which happens slowly to banks is making bad loans. This takes a substantial amount of work, generally spanning many people’s efforts over years. Lots of smart people have to go into work every day, produce a lot of paper, and bend all of their professional efforts to the task for a bank end up with a pile of bad loans.

This sounds like a joke but it isn't one. A good portion of banking regulation is making it hard to do things that blow up banks. That is why banks don't routinely have e.g. 25% of their loan book concentrated in loans to single overleveraged hedge funds, a pattern that the crypto industry has recently discovered the riskiness of.

Recall that deposits are liabilities which allowed a bank to leverage up their capital to purchase or manufacture assets, which will overwhelmingly be loans. Many things can make a loan bad. The most obvious one to non-specialists is abysmally poor credit quality; a loan which was doomed to never be repaid is, obviously, a bad loan. Very few loans are bad in this fashion.

The more likely failures of underwriting loans are pricing credit quality poorly (not earning enough interest to cover the risks associated with a loan) and poor portfolio construction.

The first was a major contributing cause to the global financial crisis; for complicated reasons, the United States via a combination of policy and market forces ended up in a situation where the financial industry greatly mispriced risks in so-called subprime home loans and overproduced them relative to true market demand for those assets and, relatedly, for the homes. (Most discourses about the financial crisis miss that the malinvestment is not just in improper operation of a spreadsheet but in bending the productive resources of the nation to build particular physical instantiations of homes, using bricks and concrete and labor and similar, in particular places, such that no buyer actually existed for the home near the price the home was believed to command.)

Anyhow, back to failing banks. Suppose you’ve made some bad loans, a process which took you years. Suppose you’ve actually lost money on some of these loans, which is not co-extensive with making a bad loan; you can lose money on good loans (and frequently will!) and you can have bad loans which are, for the moment, paying as agreed.

You can, through either ill will or desire to provide your valued customers with the experience that keeps bringing them back, paper over those losses by e.g. lending commercial real estate operators more money for new projects, with them using some of it to make payments on their old loans. This is one of the dangerous auto-catalyzing processes for bank failure: the hole is getting deeper by a mechanism which prevents you from seeing there is any hole.

In principle, bad loans are enough to cause a bank failure. In practice, there is almost always a catalyst and that catalyst is almost always a liquidity crunch.

Keeping your bank hydrated

Banks need to be liquid—to have assets which can be easily converted into money at very close to the value they are marked as having—for day to day operations. Partly this is to e.g. make payroll and pay vendors, but overwhelmingly it is necessary to service depositors.

Predicting depositors’ demands for liquidity is one of the core boring challenges of banking. It isn’t anything close to constant over time; it tends to surge around payday, for example, and holidays, and during periods of broad financial stress. (Somewhat counterintuitively, banks often become more liquid during crises, as e.g. depositors sell financial assets and move cash into the bank, or as e.g. a community bank which recently saw its community hit by wildfires sees insurance payouts move billions of dollars of settlements into the bank one account at a time.)

What happens when a bank is not as liquid as it predicts it needs to be? It sells assets, and generally the best assets go first. A bank with marketable Treasury securities (debts of the U.S. government), for example, can find a willing buyer for them with virtually no slippage at basically any time. A bank might have some bonds of publicly traded companies, and perhaps it would take more of a loss selling those, but in most market conditions those sales can be done quickly.

Run out of things which you can push a button to sell? Well, you still have options. You’ll look at your loan book, and start with e.g. high-quality residential mortgages which are conforming. There is almost always a buyer for that product, and it can be done relatively efficiently, but not as efficiently as Treasuries or marketable securities.

Then you’ll start looking at your non-conforming mortgages and commercial loan book. And around here is where you start to run into trouble.

See, your bank is heavily entwined in the microeconomy it operates in, as are all of your counterparties. (If your bank is under stress, it is overwhelmingly likely to be a community bank. That said, this generalizes to basically all sizes of banks.)

Your loan books are likely to be very correlated. Basically every effective process to grow loans introduces more correlation. You site your branches in places where you think they will get attractive business brought to them, and your loan book starts to concentrate in those neighborhoods. You hire loan officers who are good at getting deals done, and your loan book concentrates in the professional networks of those loan officers, of whom you probably have less than a few dozen. You provide excellent service to your customers and they refer their friends, and their friends tend to be from the same industry with the same rough credit profiles doing business in the same areas.

When you attempt to sell concentrated packages of risk, the buyer, who is likewise a savvy financial institution, will do two things.

One, they will want a discount to what you think the package is worth, both to make it worth their while to absorb the friction of the deal and also to compensate them for correlated risk.

Two, they will blab to everyone they know that you are shopping blocks of your book.

In particular, many of your loans are to commercial real estate developers and operators. Here it is useful to understand that CRE professionals are the most indiscrete industry on God’s green earth. The industry runs on secrets and alcohol, and both are exchanged to lubricate relationships. Some enterprising bartender should mix a cocktail and brand it Non-Disclosure Agreement; it would sell swimming pools.

Commercial real estate players are local rich people and pillar members of the community. Birds of a feather flock together, and CRE players talk to people much like themselves, at the bar, on the golf course, at church, at barbecues, etc. It is literally their business to talk to most of your deposit base.

And the thing they say will be that your institution is undergoing stress, and that the first people to withdrawal their deposits will get 100% of their money, but that later depositors attempting to withdraw might not.

And then you end up with a bank run, the most dangerous auto-catalyzing part of bank failures, where your depositors race to get their money out.

In most cases, if you’re killed by a bank run, the damage was done long before. You earned your fate via years of diligent work making bad loans, and became insolvent. The bank run revealed the insolvency.

Failing bankers often don’t agree with this. They think e.g. the liquidity constraint caused by the bank run made them need to sell off assets at a discount to their true value. If they had realized the true value of the assets, if people had just been patient, they argue, the bank would have survived.

Back to deposit insurance

Deposit insurance schemes include what game theorists would call a commitment strategy. One way to maintain trust is the messy and complicated business of building it over time. In a bank run, that trust, carefully cultivated over years, can evaporate in a matter of hours.

Another way to maintain trust is to say “An algorithm, administered by someone much bigger than any of us, who has much less emotional skin in this game, is going to absolutely steamroll all the facts of this particular situation and do exactly what it is designed to do.”

The FDIC’s algorithm, in simplified form, is “If an insured financial institution fails, we will make absolutely positively sure that each depositor gets their deposits back, up to a limit of $250,000.”

The actual recovery formula is substantially more complicated. That coverage limit is per account type, a nuance that only financial planners could love. The definition of a depositor is exactingly specified down to what happens when people share ownership of accounts.

But what the FDIC tries to do is to make information-sensitive (“This particular bank is failing!”) debt, the deposits, again information-insensitive to most depositors. “Don’t worry, the U.S. federal government is good for more money that you’ve ever had. Don’t feel the need to come to the bank on Monday, unless you otherwise would have, in which case the money will absolutely be there.”

Businesses, which frequently have more than $250,000 to their names, have treasury management practices to limit counterparty exposure, including to banks. We’ll discuss those in depth some other time. This is also available to individuals as a product at e.g. many brokerages, to somewhat artificially boost their FDIC-insured limits while staying within the letter of all regulations. (The FDIC is not thrilled about this, but the products work as advertised for the moment.)

Orderly bank failures

How to ensure that the money is there on Monday? Well, the bank didn’t fail in a day. It has been making bad loans for years. Its supervisors (regulators) have almost certainly noticed its deteriorating health for a while. They told the bank to correct its loan practices and raise more capital. That didn’t happen.

So eventually, on a Friday, the supervisor (which is not the FDIC) tells the bank that it has failed. Concurrently with this, the FDIC swings into action. The micro-mechanics of this are fascinating; they resemble a police raid on the bank headquarters except mostly conducted by people who look like accountants (and in some cases, are).

That action is, in almost all cases, selling the deposits and assets of the bank to another financial institution. Banks benefit from scale. This is a core reason that they open new branches at the margin. The FDIC’s proposal is “Hey, a bunch of perfectly good branches with perfectly good bankers just came on the market. They’ve also got some assets and… well… nobody gets here if the assets are also perfectly good. But almost any pile of assets is good at the right price. Let’s make a deal.”

In cases where the bank is not actually insolvent—where they truly are just having liquidity problems—subsuming them into a larger, healthier bank solves the problem outright. The acquiring bank gets their assets at an attractive price, and the losses (the difference between the value of the assets and that attractive price) are borne by equity holders in the original bank, who will often be zeroed out or close to it. The FDIC prioritizes depositor recovery at lowest cost to the FDIC’s insurance fund, not the interests of bank shareholders. If you have reached this point, you have been called upon to perform the sacred duty of equity in a bank: take the L to preserve the depositors’ interests.

But what about in more advanced cases, where the loan book is so bad or market conditions are so stressed that the bank is insolvent? In these cases, the FDIC attempts to throw in a sweetener to the acquiring bank.

That sweetener often takes the form of a Shared Loss Agreement (SLA). Suppose, for example, that the FDIC models that a failing bank with approximately $100 million in deposits and $100 million in loans will probably take +/- $5 million in loan losses over the next few years. They could write an SLA with the acquiring bank saying “Here’s a $5 million cash payout which we will make to you immediately, covering these doubtful loans. You are contractually obligated to continue servicing them. If you actually get any recovery, wonderful, keep 20% for your efforts and send 80% back to us.”

The SLA is so-named because the acquiring bank and the FDIC’s insurance fund split the cost of loan losses. It could also be called a form of private leverage on the insurance fund. The private-sector equity in the acquiring bank absorbs much of the risk of the loan book, versus e.g. the FDIC having to actually sell the loan book in what is likely a distressed market. This theoretically minimizes the cash outlay of the insurance fund.

You can actually read through the list of bank failures and see how boringly functional this all is. For example, if you were a depositor at a tiny $100 million bank in Florida that you’ve never heard of, when it failed in 2020 you lost zero dollars and zero cents at a cost to the FDIC of about $10 million.

The Deposit Insurance Fund has about $120 billion in it.

The cost of insurance

The cost of deposit insurance is borne by banks, as a mandatory cost-of-doing-business. It generally scales with their total liabilities (prior to the GFC it was deposits only), and is weighted based on the perceived degree of risk, similar to most insurance pricing.

The insurance costs between 3 and 40 basis points per year, which you can compare to the interest rate you earn on your deposits. This magical failure-limiting technology is very not free. The exact determinants are mostly interesting to banking nerds.

One worthy of comment: a bank is charged modestly higher rates if they are heavily dependent on brokered deposits. As we’ve previously discussed in Bits about Money, brokered deposits are critically important to the economic model for brokerages and many fintechs. They’re also ruthlessly price sensitive deposits operated by professional money managers; most deposits are not. If professional money managers think a bank is imperiled theirs will be the first money out the door. Dependence on brokered deposits increases the risk of liquidity flight during times of stress and therefore increases the risk that the insurance fund will have to pay out more to cover a bank’s failure; accordingly, there is a surcharge for them.

This is also designed as a policy mechanism to encourage more banks to build more stable deposit bases. Building deposit bases takes hard work over many years doing the boring business of banking; that is why picking up the phone to a deposit broker is so attractive even where it is more costly to the bank.

The ultimate backstop(s)

The FDIC has at least two advantages backstopping its insurance scheme which more typical insurers do not.

The first is that it enjoys the “full faith and credit” of the United States government. If it depletes the entirety of the insurance reserve in a crisis, a) that is exceptionally bad news which the entire world will read about and b) depositors still get paid because the federal government cannot run out of money. This is, again, to maintain depositors’ trust in the money-ness of all deposits at all institutions, even the bad ones, “come hell or high water.”

The second is that the FDIC exists in a constellation with the Federal Reserve, which is the “lender of last resort” for banks. During crises, when the prices of many pools of assets tend to get even more correlated and organic buyers disappear for them (or charge increasing discounts to their notional values as the premium for liquidity increases), the Federal Reserve has the option of buying financial assets or lending money to financial institutions so that they can do the same.

This was one of the mechanisms for the Troubled Asset Relief Program (TARP). A full accounting of it is outside the scope of this essay, but most non-specialists believe it to have been more of a handout than it was. To an underappreciated degree, it substituted government liquidity for dried up private liquidity and successfully charged a premium for it.

Deposit insurance as ubiquitous infrastructure underlying everything

There are vanishingly few bank failures. This did not use to be the case. Deposit insurance was, along with the larger supervisory apparatus (and substantial ongoing work from the private sector), a major component in breaking the negative feedback loops which previously caused individual failures and repeated, frequent, systemic banking crises.

It is difficult to overstate how important this technology is. You rely on it to the same degree as you do electricity, running water, and stable Internet connections. Like much infrastructure, it is so good you’ll hopefully never even have to realize it is there.

Behold the terrible majesty of the humble bank deposit.

Deposits are money

The model we are taught as children is, like most models, useful but inaccurate. “You take this $20 bill to a bank and deposit it. They will keep it safe for you, and then give you $20 back in the future, plus a little extra for having the use of it in the meanwhile. We call that interest.”

Let’s start dissecting this transaction. You don’t deposit a $20 bill. You purchase a $20 deposit, coincidentally using a piece of paper with the same number on it. The deposit is a liability (a debt) of the bank to you. The bill which you gave the bank in return for the deposit is now theirs, the same as if you had bought a cup of coffee from Starbucks. On their balance sheet, it is now an asset.

The core action which one takes with a deposit is not actually withdrawing it (plus a bit of extra interest). Most deposits will not be withdrawn by the person who originally put them in the bank.

The actual core feature of deposits is that you can transfer them to other people to effect payments. Big deal, you might think. You can also transfer cows, sea shells, Bitcoin, an IOU from a friend, or bonds issued by Google to effect payments.

But deposits are treated as money by just about everyone who matters in the economy, including (pointedly) the state. Economists can wax lyrical about what “treated as money” means, but the non-specialist gloss is probably just as useful: anything is money if substantially everyone looking at the money both agrees that it is money and agrees at the exchange rate for it. This is sometimes referred to as the "no questions asked" property; money is the Schelling point for value transfers that all parties to a transaction are already at.

This is so fundamental a feature of deposits that, in developed nations, we don’t remember that it isn’t automatic. There isn’t an ongoing exchange rate between Chase dollars and Bank of America dollars. Even attempting to imagine that invokes images of chaos. You’d have to specify a bank while doing a salary negotiation. Your rent might swing up or down based on the landlord’s judgment of your bank’s credit position. Newspapers might print daily charts, for the benefit of the business community, of the exchange rates for deposits of every bank in town, so that they knew how much extra to ask for when you wrote a check to settle a $20 purchase.

All of these, and more, used to happen.

They also happen today, in places without well-developed banking systems. Crypto is a good example, to avoid stigmatizing developing nations. There is an exchange rate, constantly changing, between the stablecoins USDC and USDT, between both of those coins (independently) and the dollars they theoretically represent. Different rates prevail in different places and different transaction sizes. This makes stablecoin-settled commerce very rare relative to money-settled commerce.

Heavily engineered structured products pretending to be simple

From the consumer’s perspective, deposits are “my money,” functionally riskless. This rounds to correct.

From the bank’s perspective, deposits are part of the capital stack of the bank, allowing it to engage in a variety of risky businesses. This rounds to correct.

The reconciliation between this polymorphism is a feat of financial and social engineering. A bank packages up its various risky businesses—chiefly making loans, but many banks have other functions in addition to the risks associated with any operating business—puts them in a blender, reduces them to a homogenous mix, and then pours that risk mix over a defined waterfall.

The simplest model for that waterfall is, in order of increasing risk: deposits, bonds, preferred equity, and common equity. Holders of all of these complex financial products have committed capital, under various terms, to the operation of the bank’s business. The more risky the product, the higher in general the return and the more risk of loss.

Careful balancing of the amounts of capital in the various categories, and of the portion of risk absorbed by equity in particular, is supposed to make deposits so safe as to never get any of the risk sludge spattered on them at all.

This happy just-so story has been told by every bank in history, and yet depositors actually periodically lost lots of money, and the banking system had large systemic crises, until it was backstopped by deposit insurance. That’s a deep enough rabbit hole that I’ll cover it in a later essay.

Why fund the risks of a bank with deposits, as opposed to funding them entirely with bonds and equity (and of course, revenue), like almost all businesses do? From the bank’s perspective, this is simple: deposits are very inexpensive funding sources, and the capability to raise them is the one of the main structural advantages banks have vis-a-vis all other firms in the economy. This allows banks to price loans much cheaper than non-bank lenders can afford to, sell vastly more loans (due to more demand for cheap loans), sell vastly more loans for any given level of capital, increase their return to equity holders, and similar.

From society’s perspective, the wide availability of cheap credit is generally considered a good thing, as it allows for productive investment, consumption smoothing over consumers’ lifetimes, and a form of risk-pooling not entire dissimilar to public support or insurance programs. (It is underappreciated that consumer credit is, effectively, one of the largest welfare programs in the United States. Chargeoffs of e.g. credit card debt effectively transfer a private benefit to the defaulting consumer in return for a diffuse cost to the rest of the public, mediated by the financial industry; the net amount of them is almost as much as food stamps.)

Deposits as pink slime

Pink slime gets a bad rap. It’s delicious, economical, and about as healthy as meat generally is. (Cultural note for non-American readers: pink slime is a weighted and generally disparaging name for what the food industry generally calls “processed meat product”; much of American processed protein is made of it. Chicken McNuggets, for example, really are chicken meat but they’ve been blended into a slurry then formed into the classic nugget shape, rather than being contiguous cuts from any individual chicken.)

Deposits, sitting on the liabilities side of a bank’s balance sheet, balance out a bank’s assets, including ones which are risky or illiquid such as the loan book. If your deposit at the bank corresponded 1:1 with an identifiable asset, you’d have to care about that asset, quite a bit. If your deposit was not merely spiritually funding my mortgage but actually backed by my mortgage, you’d watch my financial situation as if it were your own, because it would be. You’d be tempted to e.g. sell out of my mortgage and into, well, anything else if you thought I suddenly developed a taste for mixing expensive alcohol with more expensive poker tables.

And then the people doing business with you would be inconvenienced, as well, because while your payments would notionally be denominated in dollars they’d actually be backed by… something you understand a lot better than the person receiving the payment. And so they’d be less likely to transact without trust in you, demand a risk-based discount, delay transactions to do their own due diligence on me, and similar.

Specialists refer to this property as information sensitivity. (See this interesting read [PDF] on this topic.) The value of my mortgage is very tied to facts about my situation that an interested party could, through expenditure of effort, have an advantageous viewpoint on. It is information sensitive. The value of a deposit for 100 yen at the bank issuing my mortgage is precisely 100 yen, every child knows it, and no amount of malfeasance will allow you to harm someone by engaging in an otherwise fair transaction for 100 yen and satisfying it with that deposit. Deposits are information-insensitive debt.

Another benefit of deposits, quite controversial, is that they allow money creation via a public/private partnership between the government and commercial banks. We'll have to address that another day.

Many things are quasi-deposits

Banking is, when executed well, a very lucrative business to be in, and many firms and financial products replicate parts of it. These are sometimes described as “shadow banking”, which is a word which might conceal as much as it illuminates.

Be that as it may, unsophisticated customers perceive many things where a database shows an entry in dollars as deposits, and (sometimes worrisomely) those are actually sold as being equivalent to deposits. If they’re not money, they are not deposits.

Starbucks balances are a useful thing to have if you drink coffee, and are denominated in dollars (or here in Japan, in yen, etc etc). Starbucks balances are not deposits because nobody takes them except for Starbucks. (If someone asks for money, maybe you should give them money. If someone other than Starbucks asks for Starbucks balance, hang up because there is a 100% likelihood you are being defrauded.)

Many products are not as clear-cut as Starbucks balances. The fintech industry has not covered itself in glory here. Sometimes firms misclaim a product to be a deposit where it is not. Sometimes they actually institutionally misunderstand the nature of the product they have created.

One would hope that that never happens, but just like the most expensive mistake in the history of programming being the assumption that programmers can count, smart people are doomed to continue discovering that just because a deposit is a complex structured product involving a bank which has a stable dollar value, not every complex structured product involving a bank which appears to have a stable dollar value is actually a deposit.

Voyager, a publicly traded company, marketed a deposit-adjacent product to users, paying a generous interest rate. Then a cascading series of events in crypto, outside the scope of this essay, blew up a series of firms, including one which had taken out a loan of hundreds of millions of dollars from Voyager.

Suddenly, the information-insensitivity of Voyagers not-deposits was pierced, the pink slime appears both undermixed and undercooked, and customers now need to follow a bankruptcy proceeding closely.

(An interesting tangent is that, since all deposits are interchangeable by design—they are money, after all—that invariant is relied upon for the orderly operation of financial infrastructure. When something which was believed to be a deposit is discovered to not actually be a deposit, infrastructure around it breaks catastrophically. Matt Levine has an excellent extended discussion about how Voyager discovered that attaching the ACH payment rail to their deposit-adjacent product became a huge risk once they went under.)

Making the magic happen

Capital adequacy ratios, reserves, and the other not-so-secret sauce in banking are worth many books. Deposit insurance, a key ingredient in making deposits as reliable as they are perceived to be, is actually explainable in a reasonably compact form. Tune in for that next time.

As always, Bits about Money is my own opinion. My employer Stripe, which frequently has differences of opinion with me regarding cryptocurrency, has recently announced a product which uses a particular stablecoin (USDC). I have made de minimis usage of USDC personally partially out of technical interest and partially to use a prediction market. I find prediction markets intellectually interesting and have been an occasional user of them for almost 20 years. They have poker's thrill of intellectual ritualized combat and take advantage of my absolute incapacity to resist any textarea that could carry the message Someone Is Wrong On The Internet.

Stablecoins in a nutshell

A stablecoin is designed to be a deposit (i.e. money) recorded in a slow database which is negotiable among other actors who use the same slow database. This compares to deposits at e.g. banks, which are typically recorded in a faster database and are negotiable either at the bank or, through the banking system, at other users of money.

Stablecoins are typically contrasted with other cryptocurrencies, such as Bitcoin, because their unit of account is linked to a government-issued currency (overwhelmingly, the U.S. dollar) rather than more speculative assets one could record in a ledger maintained in a slow database.

Stablecoins are big business relative to most startups and miniscule relative to the money supply. There are currently a bit more than $150 billion issued, which (if they were consolidated) would be in the same weight class as a large regional bank like e.g. Fifth and Third.

Uses of stablecoins

In principle, you could use stablecoins as money, like how you use deposits as money. Stablecoins are not used like money; rather than facilitating almost the entire diversity of transactions in the economy, they are overwhelmingly used for a few niche use cases.

The cryptocurrency community often explains that the core use for stablecoins is for moving money between cryptocurrency exchanges to assist with arbitrage. This is not the dominant use of stablecoins. The dominant use is actually collateralizing investments in popular products with embedded leverage, such as Binance’s USDT/BTC perpetual futures contract. Perpetual futures are themselves a fun rabbit hole, which might have to wait for an essay of their own. Exchanges like them because they allow fat largely sub-rosa fees; institutions like them because they’re extremely capital-efficient; retail likes them because they allow high amounts of margin (which gamblers perceive as amping up their fun).

An emerging use case for stablecoins, which is not yet dominant, is that they’re programmable money that can be easily operated on by smart contracts in “decentralized finance” (DeFi). DeFi is something of a term of art; we’ll come back to how decentralized some popular offerings actually are in a minute. DeFi’s current raison d'etre is borrowing/lending cryptocurrency to allow increased leverage by traders, if one is charitable, or creating financial games which are Ponzi-adjacent (memorably described once as “boxes”) if one is less charitable.

In principle, stablecoins could be used to settle transactions. In principle, two crypto users could pull out their phones, share a QR code (to show the receiving wallet address), and send stablecoins over without their wallet providers needing to further coordinate. In practice, this is uncommon, because it is a higher-friction more-expensive slower Cash App which does not yet have the society-wide network effects of competing ways to settle transactions.

But the principle is interesting! It’s virtually impossible to find a tech investment fund which thinks that bank wires, for example, have the appropriate amount of friction and ceremony associated with them, and some tech investment funds use stablecoins to settle investments. Depending on the slow database used, this costs about the same as a bank wire (negligible relative to the investment) or less, is much faster, requires no coordination with a banker during bank hours, and may be substantially less likely to be disallowed if e.g. one of the counterparties is in another nation.

And slow expensive Cash App is a Cash App one can download without e.g. having sufficient legibility to the banking system to use fast cheap Cash App. Plausibly that is at least an intellectually interesting point in the multidimensional vector space of all possible Cash Apps.

You’re probably not here for a deep dive on slow database technology or fast database technology, and even if you were I'm not here to write it. Fast databases are complicated, impressive technical artifacts.

A more interesting question is how privately issued money gets and maintains parity with publicly issued money. There are a few different mechanisms for this.

Money market fund style stablecoins

A money market fund is a specialized form of investment vehicle designed to have the desirable characteristics of a deposit (liquidity on demand and virtually riskless) while having more yield than deposits do. This is typically achieved by having the money market fund invest in short-term high-quality commercial paper or government-backed securities. Money market funds had a rough go of it during the seize up of the treasury repo market during the global financial crisis, a story underappreciated but told in many other places, but be that as it may: fix the image of a money market fund in your mind.

Got it? OK. Now change the money market fund’s fast database to a slow database, make individual units of it movable without cashing out, and set the management fee to equal 100% of interest income. The tickets are now diamonds. The money market fund is now a stablecoin.

USDC, issued by Circle, is the largest money market fund style stablecoin (and 2nd largest stablecoin overall). Their backing is held (currently) in cash and low-duration U.S. government issued securities.

The money market model is, relative to other ways to construct stablecoins, boring. Boring is a feature! Boring means that the stablecoin operator can’t get seigniorage income through digital alchemy. Boring also means that the stablecoin is unlikely to see its value vaporized under conditions of market stress.

Stablecoins are “pegged” to something, typically the USD. A peg is a story about why two things which are not the same are, in fact, similar enough to be treated interchangeably.

The story for money market fund style stablecoins is that, while in normal circumstances you would just hold the stablecoin or move it around, you could at any time return it to the operator and receive actual money at par. Like money market funds, these stablecoin operators have high confidence that their net asset value (NAV) is always almost exactly $1 per unit outstanding. Even under conditions of substantial market stress, one does not expect e.g. short-term Treasury bills or high-quality commercial paper to become illiquid or trade at a discount, even at large sizes.

That’s not an ironclad assumption! Again, it was false in 2008, when money market funds collateralized by Lehman Brothers commercial paper or Treasury repurchase agreements (repos) suddenly found their collateral impaired or illiquid. But it is what these boring stablecoins go with.

Another similar stablecoin is Paxos’ USDP, which is much smaller than USDC. Partisans between the two would probably say the main thing that differentiates them is the regulatory regime each operates under. As someone who is Switzerland here, I’d say “Meh, pretty much equivalent, and pretty low risk by the standards of cryptocurrency things.”

Part of that judgment for low risk is that these coins have enthusiastically courted engagement with U.S. regulators. Unwillingness or inability to bow to the demands of regulators, some of which make the coins worse qua products, is one reason why other stablecoin entrepreneurs went with the models discussed below.

One demand, for example, is that the stablecoin sponsors comply with Know Your Customer (KYC) and anti-moneylaundering (AML) regulations similar to the way other money services businesses have to. Enthusiasm for KYC and AML regulations is, to put it mildly, not universal in the cryptocurrency community. It will inevitably result in having to tell the user that the user can’t do the thing they want their money to do, at least some of the time, but current practice suggests that even enthusiastic compliance with these regulations results in an equilibrium far less frictionful than prevails for e.g. wire transfers.

Another demand was for more conservative collateralization than USDC previously used. That was unfortunate for USDC, since it costs them interest income, but regulators remember 2008.

Equity-backed stablecoins

Equity-backed stablecoins are sometimes called “algorithmic” stablecoins, to suggest they have the predictability of a well-operating computer program (on the way up) and blame the loss of billions of dollars on software rather than identifiable people (on the way down). I don’t think this is a particularly helpful frame. The interesting engineering is financial, not software.

Say you have a business. That business has equity value. If that business also plugs into many counterparties, keeps a ledger of money it owes to counterparties, and allows those debts to be transferred via any mechanism, that business could theoretically function as a payments rail. The business could choose to do this via letters carried by courier, via a slow database, or via many other methods.

How does the business maintain confidence among its counterparties that its debts are always worth face value, even if it doesn’t actually pay those debts back in any given period? By reference to its equity value.

If you continue to believe that e.g. Netflix has a large equity value, and equity takes impairments before debt does, then a Netflix bond (or Netflix Dollar) should maintain its value, all else being equal. Matt Levine has a great explanation of this.

Netflix is in the business of overpaying for mediocre content and distributing it really well, not in the business of facilitating payments, and so there is no convenient way to swap Netflix Dollars. People holding dollar-denominated liabilities from Netflix mostly just ask Netflix for their money, and charge Netflix an interest rate when Netflix desires not to repay those liabilities for a while.

But in principle, if Netflix invested engineering and partnership effort in making Netflix Dollars transferable on demand, Netflix could easily issue a stablecoin valuable only by remote reference to Netflix equity. Sophisticated marketplace participants would continue to believe that Netflix was good for its debts because we live in a society and because other sophisticated marketplace participants seem willing to believe that Netflix will eventually produce positive future cash flows meriting a generous current equity valuation. And, this is really really important, the size of the equity totally dwarfs the anticipated number of Netflix Dollars in existence.

What if you wanted to make Netflix Dollars but didn’t want to spend decades on a DVD business, streaming infrastructure, and content deals? Or if you wanted to issue a lot of Netflix Dollars, like billions of them, so many that short-term swings in the value of Netflix’s totally-real-all-must-acknowledge-some-people-do-pay-money-for-their-thing business might imperil the statement “There’s a whole lot more Netflix than there is Netflix Dollars”?

Well, you can conceivably make a Netflix Dollar out of any business. Even a fake or fraudulent one.

Let’s talk Terra USD, which was vaporized earlier this month.

Terra USD, which I’ll call Terra for convenience, was an equity-backed stablecoin. The equity was in the form of a sister token called Luna. (Cryptocurrency enthusiasts sometimes like to feign ignorance about tokens being equity claims, principally as a form of regulatory arbitrage. Luna is worse-is-better equity; worse in that it has far less protections than equity, better in that it could be sold to retail without getting one put in jail (yet).)

Why is Luna equity valuable? The argument was that Luna was an equity interest (“Utility token!” Quiet, you.) in the business that was the operation of the Terra slow database. Terra Labs and other parties would make the slow database available to software developers in return for ongoing fees, which would make Luna valuable for the same reason “sell an underwhelming database subscription with a complicated pricing model over time to developers in exchange for money” makes Oracle equity valuable.

Terra Labs added a feature to the slow database which would allow one to use the slow database to exchange Luna (equity) for Terra (the stablecoin) at par. If Terra traded below the $1 peg, arbitrageurs could buy it, redeem for Luna, sell the Luna (somewhere), and profit from the difference. If it traded above the peg, you could run that trade in reverse: buy cheap Luna, redeem for dear Terra, sell the Terra, now you’ve got more money than you started with.

All pegs are stories. This story didn’t sound like a very good one, unless the Luna equity was valuable. Software company equity is more valuable when lots of people are doing interesting valuable things with the software.

So Terra Labs concocted the existence of an interesting valuable thing. They wrote a program using their slow database called Anchor. Anchor was an automated program to allow moneylending. There are many of these in DeFi land.

Many of them use an aggressive growth hacking strategy, which is saying “If you use my program, I will give you equity in my company.” This strategy is commonly called “yield farming.” Anchor was very aggressive about this, promising 19.5% APY yields on their stablecoin. To use their program to get 19.5% yields you needed to buy into their database software, which made their database software seem to be more valuable (“Look at all the users!”), which caused the value of their equity to go up, which allowed them to redeem their own equity for more money to throw at user acquisition, which…

… created a Ponzi scheme. With extra steps.

Was this a complicated bit of financial engineering? Was it conducted in secret by an elite priesthood who needed years of education to even understand the acronyms at play? Nope. It wore its I’m Going To Blow Up And Take You Down With Me heart on its sleeve. I looked at it about two minutes and then confidently tweeted about the mechanism and inevitable fate.

The cryptocurrency community did not cover itself in glory regarding Luna, because Luna made the right people an awful lot of money in 2021. As cryptocurrency venture capitalist Nic Carter has explained at length, it was apparently “the trade” of 2021. (His On The Brink is the best podcast in crypto, by miles.)

In early May, Terra Labs announced that it was going to stop subsidizing users of its computer program. Users stopped using it, never having had any reason to use it other than collecting Terra Lab’s subsidy. This caused the perceived value of Luna equity to decline, which put pressure on the peg, which caused people to exit the pegged stablecoin, which decreased the use of the slow database and further hurt the implicit equity value of owning the slow database, which…

The technical term for this is a “death spiral.”

On May 8th, Terra was the 3rd largest stablecoin in the world, with $18 billion in assets. Luna had recently been worth more than $30 billion.

It is, as of this writing, less than two weeks later. The shenanigans aren’t over, because Terra Labs thinks that it can trick people again, but many tens of billions of dollars were lost and will not be recovered.

All pegs are a story. This story will probably fill a book. The part of the book before the depegging is fiction; the rest is history.

I consider this effectively inevitable for the seigniorage model, because no business, not the most valuable business in the world, can continue having a high equity value relative to “all the money anywhere”, and to the extent a business incubates a non-neglible seniorage-backed stablecoin in it, increased adoption of that product will forever represent a (growing!) threat to the business. The better that product gets and the faster it grows, the worse the threat.

Choosing to subsidize the use of the product was an accelerant but probably wasn’t even necessary to cause the collapse, and indeed we have seen several similar schemes (Iron Finance, Basis, etc). This is a product unsafe at any size. (I really enjoyed Preston Byrne’s writeup of Basis Coin back in 2017 and, if you read it, Terra was visible a mile away.)

Fraud-backed stablecoins

Tether. Enough said.

Don’t want to comply with pesky government regulations? Do want to award yourself a license to print money? Then pretend to be a money market style stablecoin, but just lie about it. Cover redemptions with other people’s money that you misappropriate. Spread the wealth around to co-conspirators both directly and by driving up the price of assets you mutually depend on.

You are ride-or-die on the lie.

As long as everyone is making money, no one will look at you too closely. Your detractors will sound like crazy people; your co-conspirators will hopefully be extremely skilled at regulatory capture. Buy a president or prime minister. Buy several. Buy an entire sovereign nation and use it like Saul Goodman would use a nail salon.

You can afford it.

Any reason for optimism?

Successful proofs-of-concept are one way that the world gets better. It isn’t a law of nature that the right amount of ceremony and cost for international money transfers is the amount it would take to do things between banks. Wise (nee TransferWise) experimentally disproved that, and brought an excellent product to market.

Money market style stablecoins don’t look to me like the obvious future of money movement. But they’re an argument with executable computer code attached. There is at least some possibility that that argument produces a user experience compelling enough, at risks society can stomach, that something which looks somewhat like them might end up a large, enduring part of the landscape.

I don’t buy it, but some smart people do.

I look at arguments like Wise or Cash App and think “Hmm, there is an excellent argument that these should interoperate and slow databases don’t necessarily need to be part of that interoperation. Plausibly it might even be so obviously true to society that they get mandated to do it.”

What follows is the long-form written explanation, in case you don’t want to viscerally experience excruciatingly work-like pseudo-productivity for an hour.

(My usual disclaimer applies here: this is directly relevant to my professional interests, obviously, but both Bits about Money and Keshikomi Simulator are personal projects rather than endorsed by my employer. I’d like to acknowledge JT and Kay for helping me build Keshikomi Simulator; it would be a lot worse without their contributions or those of our illustrious designer.)

Reconciliation: your money has insufficient bits in it

How much human attention should be paid to money? Quite a bit, say most societies. How much human attention should be paid to individual money movements? Asymptotically approaching zero. Money is, after all, just a message between parties about their desires for the future, and while the content of that message might have involved the work of ages, technology has driven down the marginal human effort require to move a message from point A to point B to a number so low it is undetectable compared to historical scales.

An interesting habit I have when reading sweeping sketches like the above is thinking of counterexamples. Which sorts of messages still have material delivery costs? Why do they? Will those continue into the future?

Japan manages to spend about $500 million a year on telegrams (telegrams!), which is a surprising fact about the world until you realize that functionally all of them are part of wedding rituals. All cultures agree that setting money on fire is a major function of weddings; specifics differ. You might reasonably think that, far into the future, aunts of the bride will still pay NTT about $20 to have an RSVP thanking them for the invitation but begging off due to unavoidable necessity delivered to the ceremony hall; $40 or so extra if you want the telegram accompanied by decorative Mickey and Minnie statuettes.

One place human effort is currently required and glaringly should not be: B2B payments. For the last several hundred years, these have followed a very well-understood dance. A deal is struck. The exact amount of compensation is decided upon and memorialized by the sender in an invoice. The purchaser receives the invoice, reads it, then wraps money around a metaphorical brick and throws it through a metaphorical window. Someone on the other side of the window then applies forensic science to the question of what caused this particular brick to arrive. This process is called reconciliation; "keshikomi" in Japanese.

Reconciliation is a business process which arises almost entirely because of a lack of structured data in the pipelines that convey money between businesses. Because of the lack of structured data, to a greater or lesser degree worldwide, many people must spend a great deal of time every month reconciling incoming payments to invoices. Because of technological and organizational barriers, the people who do this work (in Accounts Payable or Operations or office management or any other seat at the company) often do not have direct visibility into the customer relationships which gave rise to the invoice or the instrumentalities of money movement. They’re only allowed to see the epiphenomena of the work.

How reconciliation works in Japan

As we covered previously, Japanese bank transfers hold effectively four pieces of metadata: receiving account number (bank code, branch code, 7 digit number), receiving account name (effectively guaranteed to be correct by the banking system given that the account number is correct), amount, and sender’s name.

The task of reconciliation is taking these four pieces of information, of which two are generally useless because they’re about you rather than the transaction, and associating the transfer with the previously agreed-upon debt between businesses it satisfies.

Why do this? The biggest reason is accounting; you can’t know how much money entering your bank account represents actually earned revenue and how much represents Something Else unless you’re able to trace payments back to someone agreeing to buy goods or services for money. Tying payments to an invoice is an important piece of that puzzle. Reconciliation also helps answer the questions “Which of my customers haven’t paid yet?” and “How much earned-but-uncollected money do I have on my balance sheet?”

(An important question! Uncollected revenue is not quite cash, but it is an asset, and you’re likely to collect the supermajority of it in the very near future! Your company is in the constant state of being owed money and that money being the difference between life and death, just like you yourself are in the constant state of being the air in your lungs and about three minutes away from asphyxiation. It’s really important to know your next three minutes of air are in the bag.)

Any unstructured data field will be abused by users

96% or so of Japanese business transactions, many trillions of dollars worth, go over bank transfers and are only reconcilable because of a hack of those two remaining data fields. The dominant form of the hack is “Please override your name as the sender and replace or supplement it with the following alphanumeric invoice ID.” The recipient can then read that ID and quickly look up the corresponding invoice, rather than looking through all of their outstanding invoices.

This feature is built into most payment systems that are used at scale. Credit card payments, for example, are broadly designed so that a point-of-sales system or online portal knows specifically which transaction they’re charging you for when you charge them. Operations done on that transaction in the future trivially flow over to the associated movement of money. If you go to McDonalds and your hamburger is not to your liking, when the clerk voids the sale, you get your money back so transparently that nobody needed to consider that voiding the sale and giving you your money back are in fact two different operations on two different parallel events.

This is possible with cards, and most payment systems, because of a carefully coordinated dance between the parties at the time the transaction is made. Mostly for historical reasons, bank transfers assume that this dance will be coordinated not by software but rather by people.

There is no intrinsic reason this is so! Humans don’t actually add value via e.g. a review process to most transactions! There is no one sitting at the Tokyo Stock Exchange refereeing in real time between a foreign hedge fund and domestic marketmaker saying “Yep, I’ll certainly allow this multi-million dollar transaction to go through.” Everyone has precommitted to a system which, through a great deal of ongoing effort but with virtually no marginal human attention, balances books, tidies up entries, and catches problems.

But for business payments, the lifeblood of the economy, we still rely on many millions of hours of human labor every month.

The trouble with people and unstructured data

By far the most common problem with giving people the instruction “Put the following reference number in your bank transfer” is that they simply don’t follow that instruction. Much of that behavior isn’t stupid or malicious; there just physically isn’t a pathway between the person in receipt of the invoice, the accounts payable department, and the bank which would allow overriding the company’s name on the outgoing transfers. (Surprising, but accurate.)

The company’s perspective on this problem is that if you want to keep getting their money then you’ll deal with it the same way they deal with it. You will employ a floor of people who do basically nothing other than this a week every month. You will tolerate 15 business day average confirmation times on a ~500 ms latency payment method.

Once the invoice ID is omitted from the transfer, all bets are off on attempting to reconcile it, and the sleuthing must begin. People do not have unique names. Companies don’t necessarily pay from the entity that the receiver was expecting; this could be a subsidiary or parent company (in the more sophisticated case) or from the owner’s personal account to save a few hundred yen in fees (in the less sophisticated case).

Your bank does not have any knowledge of your suppliers’ invoice IDs or expected formats, but will expect you to enter them all by hand, leading to a truly disheartening number of typoes. Customers will never cease to amaze in finding other numbers on their invoice or previous correspondence to use in place of the invoice ID. Dates, phone numbers, the Giants’ current box score, etc are all useful information to someone but probably less than optimal for your Accounts Payable or operations team.

How reconciliation could work

Recall that, of the four pieces of metadata that bank transfers carry, two are broadly useless. How about the third one, the amount?

Many Japanese businesses do factually override the amount of the invoice to help their reconciliation process. This is a thing that really happens. The mechanism is sneaking a line-item onto all of your invoices for an unspecified arbitrary discount for the customer, such that the final two digits of your invoice are unique among all invoices you expect to send that month. Then, to find the correct invoice to attach NoUsefulNameHere’s payment of 234,235 yen to, you just scan down your list to find the only one which ended in 35 yen.

This, of course, breaks down at scale, but it is a common trick in the arsenal of small business professionals throughout Japan.

The thing which actually works at scale, but is barely adopted yet, is treating the receiving bank account as being useful information rather than being the same for literally every payment into your company.

This is generally done via a mechanism called Virtual Bank Accounts (VBAs), which are a product available from the banking industry in Japan, the U.S., Mexico, Brazil, and many other countries. You contract with your financial institution of choice to reserve a block of bank account numbers corresponding to a far smaller number of actual bank accounts. You give out those numbers to your customers rather than giving out your “real” bank account number. You then take action based on which account number your customers use.

Due to technical and social issues within the financial industry, the banks offering VBAs generally expect you to bring your own implementation work at this point. Should you e.g. re-use VBAs within your block? They probably don’t have a straightforward answer to that question; up to you. Should you treat them as secrets? Up to you. Should you share them between customers? Up to you. What should you do once you know one of your VBAs has received a transfer? The bank will give you your money, what you do with the data is up to you.

Stripe does basically the simplest thing that works: give each customer/business pairing a unique VBA, shared across all invoices for that pairing (to avoid e.g. a customer not updating their supplier management system with the new bank account number on the second month’s invoice). Use ability to introspect invoices (and their open/closed/etc state) and inferences to tie incoming payments to the invoice they’re most likely associated with. Kick all the exceptions to a human or computer system, whichever the user specifies.

This works really well. Which shouldn’t be surprising: virtualized addressing is essentially to how your email, phone calls, and web visits get delivered. Billions served an hour, error rates far below any human-mediated process, cost almost too cheap to meter. The only thing preventing its use in payments is not deciding to do it. (One could similarly use virtual addressing for postal mail, by the way, and Japan has started doing so.)

The implications are straightforward: this lowers the total cost of payments by eliminating human effort, which is expensive, and human errors, which are expensive and numerous. An underappreciated consequence is that freeing people from drudgery gives them the ability to do more important, meaningful work.

Story time

I was once a peon in the bit mines of a systems integrator, and had to implement a system quite like Keshikomi Simulator, except much more serious. It was designed so that operators at a particular Japanese university could reconcile tuition payments against tuition invoices. It did not automate their paper-based process; it moved their paper-based process onto a web interface, for (real, but modest) gains in observability and recordkeeping efficiency.

For two weeks of every year, from the founding of the university through the present day, they had endured an absolutely hellish crunch quickly reconciling 80-90% of incoming payments and then dealing with edge cases.

Don’t dun Hanako; her tuition was paid by her dear uncle, who doesn’t share her last name, but did send in tuition in a timely fashion. Don’t dun Yusuke; his father paid through the account of his care dealership rather than from his personal account. Don’t dun Esmerelda; her parish took up a collection to pay her tuition and you can find it under the name of the parish treasurer d/b/a the small town church.

These facts, and hundreds more like them, were discovered every year by painstaking work on phones, faxes, and gossip trees. Any uncaught mistake in the process would result in a student being unable to register for classes, often thrown into substantial doubt as to whether they had a future in college or whether a substantial portion of their family’s life savings had been irretrievably lost. The people who performed that work (and who were in close proximity to mistakes that were made) were highly educated professionals who the university and society would have preferred do absolutely anything else. They were reconciling invoices because there was simply no other alternative; the invoices had to be reconciled or the university would cease to function.

That is a silly constraint on the world, like phone calls needing to be manually switched between circuits was a silly constraint on the world. It should be done away with. The optimal number of staff hours to spend on that process was zero. Zero dinners with family missed during the crunch weeks. Zero students dunned unnecessarily. Zero registration holds.

Zero is a bold number, but it is an achievable one. We can do it; we have the technology.

Beyond reconciliation

As payments start to carry more useful data and be more natively operated on by computers, they get far more powerful. The blockchain enthusiasts are pretty convinced that programmable money is going to win the future. I entirely agree with them on this score; I just think that the most programmed money is going to be the dollar, yen, or similar.

A big question in network economics is where in the network the value accrues. The Internet stitches together a lot of disparate data siloes but relatively little value accrues to the network itself and almost zero to the protocols. Banks and companies are siloes with some (fairly primitive, by modern standards) wires connecting them; it remains to be seen to what degree and in what fields intermediaries between them succeed. The market opportunity, though, certainly isn’t a small one.

Recently, I got back into fantasy miniature painting, after a 17 year hiatus.  (Quick plug: YouTube is the best thing ever for people getting into a hobby or profession which involves tacit knowledge. Lyla Mev did more in 7 minutes for my painting skill than all of my previous practice. You can lose many, many hours watching extremely talented people break down exactly what they're doing, and watch high-def video translate sorcery into replayable, replicable hand motions.)

What was once a traditional IP-rich industry (a great writeup by Byrne Hobart on Games Workshop can help non-geek readers get up to speed on it) has had disruptive innovation happen through a coalition of Shenzhen manufacturers, bicoastal U.S. tech companies, mom-and-pop factories throughout the Western world, and a global network of small firms producing mass-customizable sculptures.

Let’s start with this from what the end user sees and then trace the supply chain backwards.

(Disclaimers as always: while I’m not directly conflicted with any company named below, some of them are likely clients of my employer, Stripe, and I may have purchased their wares recently at standard retail prices. As always, all views here are my own, and nothing is non-public knowledge.)

“Your high elf wizard’s cheekbones are not quite what I imagined for mine”

Some people paint fantastic miniatures purely for the aesthetic value, but the dominant use of them is to play games. Depending on the game, there may be an in-game benefit due to painting, but I’ll spare you that rabbit hole; suffice it to say that some hobbyists who spend a lot of time in a fantasy world get attached to their avatar(s) and want them to look both good and the way they think they should look.

This provides a challenge for traditional manufacturing industries, because the number of SKUs a game retailer can stock is finite and the number of D&D characters is not. Even for just minimal correctness on “vaguely appropriate for a gender/race/character class” there are over a thousand combinations possible, which already is more SKUs than a hobby store can afford to stock for this offering, and players sensibly demand that their character not look generically like every other high elf wizard in existence. Some want the character a bit plumper, some want them in a wheelchair, some want them holding the magic item they strived for in the campaign, etc.

Historically, the license to manufacture D&D miniatures was held by Wizkids, which produces a few hundred different SKUs at any time. Wizkids has a model not too dissimilar to Games Workshop and has a similar production function. They hire artists directly or on a contract basis to sculpt elven wizards. They arrange for a factory in, without loss of generality, China to produce a large number of copies of that single wizard. They send those copies to distributors. Retail stores order from distributors, mark up their wholesale price of of the wizard by about 100%, and sell to the public.

Then the Internet happened, and then some improvements in plastic manufacturing happened, and then the Internet happened again.

Today, if you go to e.g. Etsy, you will find more plausible high elf miniatures available, for about $5-10 each plus shipping (about the retail price point for single miniatures), than Wizkids could create in a corporate lifetime, with hundreds more being released every month.

Consider, for example, ScatterMaster’s rendering of Amlund Mageon, a wizard who bears a striking resemblance to many depictions of Gandalf, in a bit of remixing extremely well-known in the hobby and by Tolkien as well. A hobbyist unfamiliar with this might think that ScatterMaster is a truly talented sculptor or has hired one.

ScatterMaster may be an artist in their own right, but relevantly to a transaction on Etsy, they are a small U.S. based manufacturing firm which has licensed the rights to print Amlud Mageon for profit from a French sculptor who does business as Galaad Miniatures. The terms of the license is simple: ScatterMaster pays Galaad ~$30 a month for a license to reproduce any of their sculptures in any quantity as long as it is strictly 3D printed only.

It is a non-exclusive license (and that's why the artist enjoins the manufacturer from producing it with a technique amenable to industrial scale; to maintain the value for many licensees rather than reducing Mageon to his weight in plastic). Dozens of firms are paying Galaad the same $30 a month, along with a hundred-odd individual hobbyists.

Contrary the starving artist cliche, Galaad receives approximately the median French salary from Patreon (virtually) guaranteed at the start of every month, and to earn it the next month they just have to crank out another 8 sculptures to add to their library.

They can (and factually do) earn more with taking commissions, running Kickstarters or similar where they provide (or arrange for) their own fulfillment, or license STL files a la carte. If you really like Amlud Mageon, own a 3D printer, and wanted to print him yourself, the 3D “source code” is available by Galaad on MyMiniFactory for $5, of which Galaad gets to keep $4.50.

The average hobbyist does not own a 3D printer, and specialization in the hobby (and hobby-adjacent artisanship) is what enables ScatterMaster and Galaad to both have businesses. The small-scale manufacturer fronts the capital for maintaining (probably) a printing farm producing miniatures as fast as orders can come in, and does both the labor required to turn semi-toxic resin into pretty figures and also the entrepreneurial application to (basically) make sure they show up on top of Etsy and Google for searches for “wizard miniature for D&D game maybe a bit old and wizened?” The artist gets to largely outsource per-order customer service questions, fulfillment, owning and operating a printing farm, the complexities of international trade, etc. They can focus on marketing to end-users (and perhaps more importantly to potential manufacturers) and sculpting miniatures which fill the holes in their users' product lineups or tabletops.

There are many other firms one could name here, on both sides of this market. Some of them are strictly mom-and-pop manufacturers or individual designers. Some are small but thoroughly professionalized firms, like Loot Studios in Spain, which has an in-house painter and video crew to produce marketing collateral showcasing the work of their team of sculptors to potentially licensees.

And interestingly, in sharp contrast to most global plastics manufacturing, it is extremely useful to this supply chain that both the final producer and final consumer are in the United States, because it enables low-latency low-cost shipping. Most plastic you consume originated in China and got to you via, at one point, a rather slow ship, and the latency between the factory and your door was on the order of several months not a few days. Most plastic doesn't particularly care about the delta of a few months or staying in inventory, but a malevolent lich has a tight schedule to keep or he might be dead (permanently, this time) before his physical instantiation arrives on the tabletop.

But there is still a supply chain implicated here in China.

Shenzhen is eating the world

There are broadly two types of 3D printers in common use. One uses thermoplastic filament sourced from a spool and extruded through a heated nozzle attached to a gantry with three axes of motion to build a printed object from the build plate on upwards. This was the first widely commercial available 3D printing technology for home or small business use, and while it has a lot to recommend it for many applications, it did not take off for the miniature use case.

The other type of 3D printer is a resin printer, which are a technological marvel of chemistry and hardware design. If you want a full technical explanation I recommend the Youtube-delivered seminar Ph.D Chemist Explains 3D Printer Resin by a fellow painter who also found his professional life colliding with his hobby unexpectedly.

3D printer resins are liquid photoplastics; they cure (harden) in the presence of UV light. An LCD screen beneath the transparent bottom of a vat of ooze exposes a layer 30-50 microns thick to harden it; a single-dimensional screw then rotates to pull the build plate upwards (to remove it from the film) then downwards (to get more liquid resin stuck to the newly-solid layer then re-adhere to the film). The process then repeats until the print is done, in something a calculus teacher might describe as integration by parts and that a timelapse videographer might describe as absolutely mesmerizing.

And here physics and calculus made an unplanned splash into the economy for printing 3D fantasy miniatures. The thermoplastic option has build time increase linearly with the weight/volume of models printed. The photoplastic option does not; build time increase linearly only with their height, but the other two dimensions are free up to the limit of your printer’s build plate.

Recall that printed miniatures are numerous, extremely valuable relative to their weight, and…  very, very short. So you could, with a prosumer-grade printer costing only a few hundred dollars, print miniatures with a retail price equal to half the cost of the printer in a single print run lasting approximately two hours and requiring on the order of $3 of expendables (resin, cleaning materials, paper towels, and gloves) and less than an hour of skilled labor.

Intel would kill for the ROIC here, is what I am saying.

So where does Shenzhen come in?

The many faces of ChiTu

There is a thriving market in 3D printers both for industrial applications and for pro-sumers, with price points for the later offering clustering in the $300 to $1,500 region. They are largely made by a small cluster of companies which, unlike most manufactured Chinese products by tonnage, are consciously branded, with global marketing campaigns and customer service. Famous names in this sector include Elegoo (based in Shenzhen, short for “electronic googol”, a reference to a bit of math geekery which has proven auspicious in the tech sector before), Phrozen (based out of Taiwan), Anycubic (back to Shenzhen), and similar.

Much like paper printers, it turns out there is a flowing river of money in the resin printing business in rough proportion to the economic utility of all printed things. Most of that money is literally liquid. The hardware business is an extremely rough one to be in; my quick estimate of the BOM (cost of included materials) for a low-end pro-sumer printer is about half of the purchase price. But the printer literally exists to drink resin, and resin can be optimized and branded. Paper printers have fought an unending war against third-party ink toner, which needs to solve a difficult but tractable problem of repeatedly transferring pigment to paper. Resin, on the other hand, exists in an 8-dimension product space before you even start seriously thinking about it; if (for example) it doesn’t cure effectively on the wavelength of light your machine outputs then it is virtually useless to you.

This results in the printer manufacturers getting to price their resin at a premium and keep it, while maintaining a commanding share of kilograms consumed by their installed base, rather than being predated by third-party resin. Resin is sold in 500g to 1kg bottles at price points in the mid tens of dollars; it is produced in multi-tonne increments and likely at 98%+ margins. The chemical and economic properties of it make it basically ideal for putting in a shipping container even when the things it will be used to make would find that container a terrible economic fit.

And so the printer manufacturers likely make most of their money from selling consumables, primarily resin but also replacement parts for the machines (the physics of them currently invariably cause disabling damage to key parts over time, the LCD screen will eventually burn out, etc).

The printer companies are extremely sophisticated at marketing, to a degree which would be notable in almost any industry. They all operate first-party online storefronts and e.g. Amazon stores in their key national markets, and they aggressively promote themselves via e.g. ensuring that widely-watched hobbyists like Uncle Jesse receive review copies of their printers and all the resin that they can disclose. (One could write another essay just about the information economy for printer/resin/etc buying decisions; it’s a lot of the affiliate links, starred reviews, influencer marketing, online advertising, Amazon gaming, and similar you would expect.)

And inside substantially all the resin printers… is one board to rule them all, brought to you by ChiTu. ChiTu has a mortal stranglehold on the software and hardware that turns exported 3D designs into a series of tightly coordinated instructions for the motor and LCD screen. The printer manufacturers can all source final assembly and commodity electronic and mechanical components, but (in this segment) are not able to build the board yet.

ChiTu has vertically integrated itself from the embedded software segment up the stack to the slicing software as well. Slicing software is a requirement from 3D resin printing; it helps the operator turn an artist’s sculpture into something that can be physically realized on the hardware they operate. The most basic function is chopping the sculpture up into a scripted series of 30-50 micron slices (hence the name). It is also used to e.g. add additional planned plastic to support the structural integrity of the print. A fairy needs to hang upside down from a build plate fighting real-world gravity while perhaps being supported by a gossamer dress and the dreams of an imaginative child. Tolkien didn’t have to worry about that; operators do, and so you both have to put on (in software) and take off (with your own carefully gloved hands) plastic supports.

This is another reason why the mom-and-pop factories pumping out prints have an economic reason to exist: there is substantial skill involved in correctly operating one’s slicing software, one’s printer, one's post-production process to produce a model which is realizable in the physical world and survives its rigors in an aesthetically pleasing fashion.

(I have three scaly 30mm lizard rumps attesting to the fact that dragons might be vulnerable to arrows of dragonslaying but are absolutely destroyed by face-first contact into curing resin if their vicious claws can't stay attached to a milled aluminum build plate. Each of them cost me hours and a thorough cleaning of the printer to prevent the congealed puddle of dragon from penetrating the tank’s film and drenching my WFH desk in mildly toxic resin. Clearly the economically rational thing for me would be to procure my dragons from Etsy, but this is more fun than I’ve had in ages.)

Anyhow, ChiTu has started forcibly integrating ChituBox, their slicing software, with the leading printer manufacturers, and they have them over a barrel (of resin). The manufacturers hate going to their customers and saying “The hardware you buy from us now comes with a we-hope-this-won’t-be-mandatory $169 a year subscription”, particularly as they get to keep 0 cents of that subscription. (There exists competing slicing software, most interestingly the Belgian/French Lychee, but they don’t manufacture the baseboards).

So if you are following the money:

An end-user buys something from a small manufacturer on Etsy or a similar marketplace.

That manufacturer made a capital investment and OpEx purchases from a printer manufacturer. They have ongoing non-exclusive IP licensing agreements with multiple artists, brokered by software/financial services like Patreon and MyMiniFactory.

The printer company buys their boards from ChiTu, which (separately) may charge the plastic manufacturer an annual software subscription.

This is an impressively multilayered, global supply chain. Prior to buying my own printer, a single miniature routinely involved a manufacturer in Utah, a printer (and resin) from Taiwan, a board from China, a sculpt from Spain, and logistical and financial knitting by Etsy (East Coast, USA) and (multiple uses of) software-heavy financial platforms (West Coast, USA).

It's useful to ponder why the money movement here is more complicated than simply "banking" or "payments." The models themselves look like a classic e-commerce transaction, but their supply chain implicates a high-volume aggregated-then-distributed subscription commerce service needing to make multinational pay-ins and pay-outs. That is something that you can't conveniently get from either a bank or from the credit card networks. And the capability to offer this allowed entrepreneurs to create a business model which looks very little like existing firms; it is neither vertically integrated nor does it recapitulate the supply chain for most physical items with embedded IP.

Kickstarter: Marketing and capital stack for B2B art producers

Another wrinkle here: we’ve discussed the operation in steady state of an artist, plastic manufacturer, and similar, but cold starting an artist on the subscription model is almost as difficult as cold starting a SaaS company. As you might expect, VCs are not exactly wowed by the pitch “I am going to sculpt a lot of ogres; the IP will eventually be worth tens of thousands a month” and many artists cannot afford to bootstrap themselves into having a library worth licensing.

Enter Kickstarter, which both has an impressive amount of reach in hobby-adjacent spaces (people who enjoy board games have written more about the mechanics of board game Kickstarters than you could read in a week) and critically can raise I-can’t-believe-its-not-capital in advance of the IP being fully printer-ready.

And this itself has created a thriving little ecosystem of software firms! And enables advancements into more capital-intensive manufacturing!

A worked example: Dragon’s Hoard (the editor in me might suggest "Horde" instead since if a gamer wanted the pile of gold and gems a dragon slept on they would find it from a separate sculpting firm) is a set of miniatures which had a successful Kickstarter raising $250k. The campaign's proof-of-competence was enough beautifully sculpted goblins to demonstrate likelihood of success, but probably not enough to sustain a firm via the above-described ecosystem. This both paid for years of artist time and also a substantial amount of fulfillment work, because they are selling miniatures and not computerized descriptions of miniatures.

The physical properties of the models they describe are not easily achievable by the production processes used by traditional miniature manufacturers or the resin 3D printing ecosystem. They suggest to me that they might be adopting SioCast, which has the wonderfully evocative (to industrial engineer)s tagline “the link between 3D printing and injection molding” and which has to be seen to be believed. SioCast, if it is widely adopted, would be another quantum revolution in miniatures manufacturing, because it combines the (very high and low-marginal-labor-cost) scalability of traditional injection molding with the (very low) setup costs and lead time of 3D printing.

And while that might not make all that much difference for the long tail millions-of-SKUs that the hobbyist market wants, it makes a huge difference for e.g. wargamers, who need hundreds of models and can tolerate many of them looking substantially identical. You could imagine the Games Workshop of the future shipping new boxes of models directly to subscribers (and perhaps or perhaps not to hobby stores) on a monthly cadence, not on an annual-or-longer refresh cycle, basically as soon as their artists could sculpt them.

Anyhow, that is my guess at why it costs about $5-10 for a single artisanally produced fantasy figure via 3D printing but Dragon’s Horde is pre-committing to more than 100 for less than $100. They have successfully re-achieved economies of scale in plastic production, but at a scale more easily obtainable than in traditional plastic production, with a product which is substantially better for purpose than the one produced by traditional techniques.

Kickstarter apparently leans into their marketing and I-can’t-believe-its-not-capital-raising side but largely punts on fulfillment, which has caused a flourishing of boutique software-and-financial-services providers called “pledge managers” to step into the gap. A pledge manager is something akin to a CRM crossed with Shopify crossed with a physical fulfillment workflow. It lets backers pay for their (country-specific) shipping costs once the product is actually ready for delivery, lets the campaign not drown in Excel while managing thousands of orders, and it sell “late pledges”, which are post-Kickstarter-campaign purchases of the product. You could see Dragon Hoard’s here on GameFound.

And, playing forward a little bit, these Kickstarter campaigns buy artists/designers/entrepreneurs enough runway to produce assets which have enduring value, both the IP created and the customer relationships and reputation within the community. Reputation both leads directly to more customers and indirectly to more demand for prints of things in your style/range on e.g. Etsy, which causes manufacturers to license your IP. This causes a virtuous cycle not entirely dissimilar to the economic engine behind Disney or Marvel, but heretofore unknown in artisanal scale fantasy sculpting.

HeroForge: AI eats enables amateur sculpting

If you’re paying attention to the intersection of artistry and AI you might have seen it roiled by Dall-E, a language model which spits out plausible-if-vaguely-hallucinatory visual artworks in a variety of styles. Many commentators have opined on whether or not this will endanger the jobs of visual artists.

I tend to think that artists, like writers and programmers, generally succeed via achieving symbiosis with emerging technologies rather than being supplanted by them. Every programmer you’ve ever met is utterly dependent on an AI to do their work; we just call it a compiler or interpreter. The principal reason that the programmer/interpreter symbiote is not classified as an AI is that a book describing their operation isn't filed under science fiction but rather history or current affairs.

So imagine you’re someone with an idea for a fantasy character but not the time or sculpting skills to chisel them out of the ether. You might use HeroForge, a truly remarkable technical achievement which is essentially an in-browser CAD software which specializes in fantastic characters. Their software is free to use. The business model is that you can export the character you design to either a 3D printing business located in China (for about $20-$45 plus shipping, depending on whether you want it colorized) or to an STL file for printing on your 3D printer (for $8 a la carte or ~$3 if you want to commit to a monthly subscription).

The results are…  impressive. (See below.)

And despite the fact that they offer almost uncountably many models which would be at home in a D&D game, they do not in fact result in technological unemployment of fantasy sculptors, who are enjoying the best market for their skillset in history. A moment's thought will answer why: HeroForge is extremely good at producing a beautiful model which will look to all cognoscenti like it is a HeroForge model. Something of the job-to-be-done of a fantasy miniature is looking like you belong in the world with your friends but are nonetheless very different from them. In this, they are not dissimilar to fashion (or art generally).

Bringing it back to the real world

And there you have it: that’s the geekiest combination of chemistry, hardware/software, and international finance to have recently colonized my desk. I find these sort of deep dives into tiny parts of the real world are good in providing context to the broader forces which are reshaping society, such as globalization and increasingly software-mediated economies. If you’d like to see more (or less) of them, let me know. I promise that the next one won’t be quite this geeky.

To echo a line from Stripe's annual letter, we're still in the earliest stages of the broadly participatory cultural (and economic) dynamism unleashed by the Internet.

This issue will unfortunately be a little U.S.-centric both because Americans have a distinct charitable tradition (leading the world in donations by every metric you can imagine) and because a lot of the infrastructural complexity is downstream of a societal decision to subsidize most public charitable works via tax law. Tax treatment of charitable giving varies wildly between countries. In Japan, for example, you can only deduct charitable donations given to the Japanese government itself, which creates its own truly fascinating ecosystem that I’ve written about elsewhere.

(My usual disclaimer here: I’m currently employed at Stripe, was previously the CEO of Call the Shots, Inc. from its formation through dissolution, and am speaking solely for myself.)

How charities are organized in the U.S.

Most organizations with a formally recognized legal existence in the United States are created under the laws of a state. This sometimes surprises non-American observers; companies in Delaware and companies in Illinois very literally aren’t the same thing (except when companies in Illinois were incorporated in Delaware because… that’s another column).

Charities are generally very similar. Some states have particular registration processes for them which are distinct from for-profit companies (or other forms of organization); in other states they’re organized as regular companies which happen to have some differences in their mission, founding documents, and similar. And indeed the rest of the world mostly interfaces with charities as if they were businesses. Mostly. A charity can rent an office, hire a person, buy a carton of printer paper, open a bank account, invent a work of art, or be sued, just like any company.

Most differences in how charities operate compared to for-profit companies are a matter of choice. Much like companies, they are generally given extensive latitude in how they conduct their own affairs, and the rest of the world has extensive latitude in how they choose to interface with them. There is one very conspicuous exception: tax-exempt status.

Lay people often say the word IRS (Internal Revenue Service, the U.S.’s federal tax agency) at this point. Individuals I know who worked at the IRS are extremely, extremely exasperated about this and will expound upon given any provocation or none at all: the IRS does not make tax policy. The American people, through their elected representatives, make the tax policy that they wish to be governed under; the IRS is then in charge of converting those wishes into forms, ingesting (and distributing) a great amount of data and money every year, and assisting taxpayers through the process.

One thing that the American people are overwhelmingly in favor of: charitable works, broadly characterized as “Come on, everyone knows what charity looks like”, are things we want more of in the world at almost all margins, and the government should therefore encourage them via tax policy.

This results in charities having two linked but crucially distinct tax benefits. One is that donors can deduct charitable donations from their income prior to paying income tax. This has the effect of subsidizing donations; the out-of-pocket cost to convey $10,000 to a charity is, for most donors, a few thousand dollars less expensive than spending the same $10,000 on almost anything else.

The second is that charities themselves are exempted from a range of taxes, most prominently (at the federal level) income tax, but again the American public is extremely, extremely in favor of charities and so many state- and locally-assessed taxes also have a carveout for charities. Sales tax, often surprisingly to people not involved, frequently exempts any purchases made by charities, which can very literally result in charity employees flashing paperwork to confused clerks or businessmen at checkout and trying to (sometimes) patiently explain that the total printed on their receipt is illegal and please try again.

This is an enormously valuable privilege and one which is extremely vulnerable to misuse, and so the American people have instructed their elected representatives to square a circle: they don’t want the government making value judgments about charities, because government shouldn’t be in the religion business or pry into the affairs of civil society’s diverse panoply of do-gooders, but they do want the government to make sure not to subsidize putatively charitable organizations which are clearly not actually charities.

And so our benighted friends at the IRS find themselves caught, not for the first time, between multiple priorities the public holds in equally high regard which conflict with each other.

The way the regime (which, again, a creature that the American public caused to be written into law, not a thing the IRS cooked up) works is that the IRS cares not at all about charities per se but exercises pre-authorization and ongoing monitoring of charities which seek to avail themselves or their donors of tax-advantaged status. They explain it in patient and exacting detail on their website.

Recordkeeping for charitable donations

Many (but probably not most, interestingly) transactions made in the economy need to have records kept of them for tax purposes. The IRS generally awards taxpayers an extreme degree of latitude in choosing the form of their records. So much so that, as a one-time first-time small businessman, I spent hours looking for a sentence which would clarify whether I could keep my records in a database or not—the IRS is so neutral about format that they used to not mention software at all. The exceptions to this format neutrality are records which would be needed to substantiate tax positions that have historically been extensively abused.

Charitable donations fall into that category, and so (over a low limit, currently $250) donors seeking to deduct their donations need to get a written acknowledgement from the charity substantiating the donation.

The requirements for that acknowledgement are simple enough that a seven year old child could produce a conforming one with two minutes of instruction. So of course there are billions of dollars of professional labor and software spent on charitable recordkeeping every year.

My opinions on this are, candidly, complicated.

One is that infrastructure always sounds easier to build at societal scales than it actually is, and given the American people’s decision that they would build and subsidize a national fundraising mechanism which is materially implicated in literally every community and literally every field in life, the American people were of course deciding to invest in some expensive infrastructure. That infrastructure is extremely operationally complex, costs what it costs, and the public benefits enormously from it.

One of the numerous non-obvious requirements is that the charity needs to be able to produce a report at the end of the year (Form 990 Schedule B) listing any donors who made more than $5,000 in aggregate tax-advantaged donations, which causes data processing and retention issues that regular businesses don’t have. (Those probably seem like pretty minor issues, but I would say with a high degree of confidence that none of AppAmaGooBookSoft could today answer the question “How much did I spend with you last year across all transactions?” to a degree of rigor their accountants would be comfortable with signing off on to the IRS.)

This requirement serves dual purposes: it allows the IRS to cross-check claimed deductions with the charity’s own return. This helps foil tax cheats getting a convenient seven year old to forge charitable receipts. It also allows the IRS, and by extension the public, to verify that the charity isn’t primarily rich people self-dealing. The test used effectively (partially) outsources the judgement to the public. You are a charity if you do good deeds, and you are presumptively doing good deeds if you can convince widely disparate Americans that you are so effective at doing good deeds that they should part with their own money to have you do more of them, without getting any direct benefit in return.

A fun two-opposing-ideas-synthesized part of charitable tax policy: while almost all charitable tax returns are public records and have to be both published by the charity and released by the IRS on request, the identities of donors are kept private. The American public both doesn’t want people to cheat the system but also expects privacy in their dealings with e.g. their churches, schools, medical providers, etc. If you find a charity’s Schedule B on their website, they should have all large donations listed with one line per donor but the name and address of the donor redacted.

My third thoughts, though, whisper that the charitable-industrial complex partially arises because charities are extremely price insensitive when it comes to taking donations, because donations are (of course) free to them at the margin, and so they tolerate leaking very large amounts of money during the fundraising process. The tolerable degree of fundraising (and administrative) expenses in the charitable-industrial complex has caused substantial debate over the years, both inside and outside of it. We won’t resolve that debate today, but just know that every e.g. portal you’ve ever used for online giving has as one of its primary selling points to charities “We replace part of your massive recordkeeping operation with software for less money than it would cost you to hire people to do it, but that software won’t be exactly cheap.”

There is a surprising bit of nuance in the real world, one reason it is so infinitely interesting.

Charitable donations and time travel

My elevator pitch for why anyone should care about finance: it is in the business of teleporting value through space and time, no more and no less. This is a stupendously useful capability for society to have, which is why substantially all societies place an immense deal of importance on it.

The subsidy for charitable donations causes some donors to have a problem which needs a time machine: the times at which they gain access to money, at which they earn that money for tax purposes, and at which they want to donate that money to charity may be several years apart. This complicates receiving the subsidy, because income taxes are assessed on an annual basis.

Enter an infrastructural innovation which relatively few people have heard about: donor advised funds. Despite having something of a hobby in Weird Types of Accounts You Could Find At Financial Institutions, I had never heard about them until a few of our donors told me that they would, of course, be donating through their DAF.

A DAF is a chimera which holds itself out to the world as a freestanding charity whose primary charitable purpose is funding many other charities, and holds itself out to donors as being essentially a financial institution that can time travel in return for allowing you to direct but not withdraw your deposits.

Here’s why they matter: for people who have lumpy income, like business owners or employees of Silicon Valley startups, putting money into a DAF is a charitable donation effective immediately. If, for example, you’ve just sold a company (and made more money on that transaction than you expect to over the vast majority of years of your working career), you can sweep your expected charitable contributions for many years into the DAF now, take the tax break in the year which it is most advantageous (right now, when your income is higher than in other years), and then get to actually distribute the money from the DAF to charities at any time of your choosing in the future.

DAFs aren’t a secret, per se, but they are extremely well-known in a relatively small community of practice and not broadly understood outside of it.DAFs are some of the largest charities in the world. Fidelity Charitable, for example, has 250,000 donors and has distributed more than $60 billion to other charities. (Including, disclosure necessitated not by law but just by my personal sense of propriety, the one I ran, at the recommendation of two donors.)

The “recommendation” thing is… interesting. After you give money to a DAF, it is the DAF’s money, to fulfill its charitable mission however it sees fit. Do donors really believe that? Well, they mostly believe that they have a phone number, website, or mobile app they can operate where they can make a polite recommendation to the universe that a particular charity quickly receive a particular amount of money. The DAF might, in a purely optional way you understand, use that recommendation as part of its charitable decisionmaking process. Perhaps that process might include having a member of an operational team confirm that the suggested charity is indeed tax-exempt. Perhaps they might choose to note that fact in a log. Perhaps they could reach out to that charity to briefly discuss the best way to get them money. Perhaps they could send them an amount of money exactly corresponding to the amount that their donor recommended.

Or they could do anything else at all, as long as it was charitable.

Just like the little old lady next to the pachinko parlor who has an odd hobby of buying exactly the brand of pencil the pachinko parlor awards as a no-cash-value gift prize for an oddly high amount of money, DAFs have quixotic preferences and are very effective at convincing sophisticated donors that they will maintain those preferences in the future.

Described like this it probably sounds like a loophole, which, the IRS might explain in an exasperated tone of voice, is a synonym for a very thick book of regulations exhaustively written to be compliant with the desires of the American people as expressed through their elected representatives.

How the money actually physically arrives at the charity is interesting. DAFs are, like all organized charities, corporate entities which have full access to the banking system. They’re capable of making electronic funds transfers like e.g. ACH payments and wires.

But a surprising portion of donations are sent via paper checks. Why, when paper checks are the worst way imaginable to move money? Because paying someone with a check requires only three things: you need to know their official name, you need to be able to put the check physically in their hands, and you need to know that they will quickly deposit the check after they receive it.

Paying someone with a paper check does not require telling them that you’re going to pay them with a paper check, getting their approval prior to sending the check, or coordinating with them on check receipt. This is unlike e.g. credit card payments; you, a church school in Nebraska, need to have taken some action to enable credit card payments, but you don’t need to have taken any specific action to be able to cash a check (beyond having a formal existence and a bank account).

And the DAF can ride on the coattails of tax administration to determine your official name, tax-exempt status, and address you can receive mail at. Those are all public records. Almost all charities in the country have their name and address in a database that you can get from the IRS or a variety of commercial providers.

So if a DAF reaches out and contacts a charity to give it money, it doesn’t strictly speaking need the charity to respond to their phone call or email. It might choose, in its charitable wisdom, to just send the charity a check at the address the charity has promised the government has a responsible officer who quickly opens any mail addressed to it. (Why would a charity not respond to that unsolicited phone call or email? Many charities are quite busy, and particularly for smaller charities, the pitch "I represent a multi-billion dollar pot of money. We've decided we like what you're doing and might want to give you some money. Tell me a bit about yourself and we will probably decide to do that. Can we begin with the number on the top of your tax forms and your banking information, please." sounds a bit suspicious.)

DAFs and their donors rely on the (reasonably good) assumption that a charity which finds an unsolicited check in the mail will quickly deposit it. If they don’t, eventually the DAF will void the check and have to decide to do something else with its money, perhaps in reference to recommendations made by people who have previously donated to the DAF.

If a charity does answer the phone/email the flow is good and pretty straightforward, I say from experience. I was asked to briefly explain our charitable aims, produce a copy of the IRS determination letter, and (optionally) confirm banking information in case we wanted to receive electronic payments in preferences to checks if the DAF decided, in its charitable wisdom, to make donations in the future. A check arrived in the mail within about a week. It was more straightforward than almost any other channel money came into our organization by, seen from our perspective. (The joys of charitable fundraising and e.g. grant approval processes deserve a book, probably titled something like Painful Processes One Hopes Exist in the World Without Ever Encountering Personally.)

An additional feature of DAFs is that, between the donation of the money and the disbursement for the charity, the donation can be invested, in the same way that all charities can invest money that they do not need for current-year operations against their future needs. The investment returns are tax-free from the perspective of the original donor; after all, they never receive the investment returns, they merely gain moderately more karma with the DAF for making future recommendations, which karma the DAF chooses to denominate in dollars and keep exactingly precise records of.

This feature forms most of the business model of DAFs: Fidelity Charitable is an offshoot of Fidelity Investments, and invests most of its money in vehicles that Fidelity has put together. Fidelity charges Fidelity Charitable to manage those investments, between 0.015% and 0.99% per year. This model replicates well enough that many large financial institutions targeting the mass-affluent have an associated DAF; Schwab, for example, does.

As a taxpayer with obligations to both Japan and the U.S. there is literally no entity on earth that all laws I’m subject to consider worthy of tax subsidy, so I do not have the typical use case for DAFs. But since I'm fascinated by infrastructure, I started using Daffy, a startup by some Wealthfront-associated folks who apparently don’t care about unbeatable competition in their brand search results from cartoon characters. I find it useful just from a management perspective; DAFs (and Daffy) allow you to donate anonymously, and the app functions as a single “Figure out how to give these do-gooders money for me” portal, enough to justify the $3 a month that they charge me. I’d recommend it if you want to play with one casually.

In a strictly personal capacity, I will mention that Fidelity Charitable basically created the category and is the commanding favorite among people I know who have an opinion on this, largely due to being the commanding favorite of the professionals they ask for advice. Strongly consider asking your competent professional advisor for recommendations if e.g. you work in the tech industry and will, over the course of your career, want to donate appreciated stock.

(Your advisors can tell you about the tax consequences of that, which are a compelling bonus given that you are charitably minded. An important operational consideration is that large DAFs know how to use the ACATS system to move shares of stock around and your local public basketball program probably does not. You can use the DAF as an intermediary to transform the stock into a check to the basketball program; they absorb the complexity there on behalf of many thousands of charities in parallel.)

This is a pattern seen elsewhere

So charities have at least two purpose-built financial products that the rest of the world doesn’t need, DAFs on the donor side and specialized recordkeeping systems on the charity side. Where else do we see this phenomenon?

Charitable bank accounts basically look like all bank accounts, at least in the status quo, but that will likely change over time. The advantages to having software which is aware that you’re a charity and aware that you accordingly have charity-specific needs are overwhelming, and the people who specialize in the needs of charities will likely partner with the banking sector to bring better charity-specific banking products to market.

Charities are not the only sector with a relatively small number of extremely high-salience requirements, often caused by regulation. Increasingly, these requirements are going to be administered not just by professionals who have specialized in those requirements for most of their career, but by software systems catering to those professionals. You’ll see a lot of the dollars follow the software, rather than the software merely keeping track of the dollars.

Intellectual property lawyers, as one example, frequently concern themselves with discerning the colour of bits and then pronouncing whether a program operating on those bits is behaving legally. (Computer scientists, on the other hand, believe that answering whether a program will behave legally is formally undecidable.)

Money is also widely believed to be fungible. A dollar is a dollar and a yen is a yen. On the one hand, this is a foundation on which our civilization now rests. On the other hand, human society doesn’t evaluate money like math does. Money also has colour.

A salary payment is coloured differently than a gift, even if they’re the same amount paid by the same person the same way into the same bank account. Some dollars were re-coloured by implication recently, without those dollars ever being touched or named or even intended to be acted upon.

Accountants frequently concern themselves with tracking the colour of money, because business managers, owners, and tax agencies care a great deal about this question.

Which brings us to a question which sounds deceptively easy: if a customer pays a business money, does that money have the colour of “revenue?”

Why revenue recognition matters

If you were paying attention, you’ve probably guessed that the answer is “Money paid from a customer to a business is sometimes revenue.” In the general case it is undecidable, but it is nonetheless extremely important that businesses quickly come to a defensible conclusion about how much revenue they earned in e.g. a month.

The process by which a business discerns that some money has the colour of revenue is called revenue recognition. The exact details depend on your jurisdiction, accounting standards, and a myriad of factors, but broadly the accounting profession has three tests.

Money is revenue when:

1. The amount to be collected is known exactly.
2. The thing the business is being paid for has been performed; the customer has either taken delivery of the product or received the service completely.
3. The business has reasonable confidence that the money will actually be collected.
   

Why do we do this to ourselves, when we could simply count money as it is received? Because accounting is designed to perceive the world as it actually exists.

A business which has performed all their obligations to a creditworthy customer should certainly expect to receive money; they are in a better position than before they performed their obligations. If society did not agree they had received revenue, they would both be allowed to do some incredible tax gamesmanship by delaying receipt of payments until convenient for them, and also they would experience artificial distortions in their apparent health caused by e.g. the banking calendar or implementation quirks of their payment processors.

Revenue recognition in SaaS

In a past life, I founded several software-as-a-service (SaaS) companies. They were much smaller than the publicly traded giants who were thrown into a tizzy by IFRS 15, an accounting standards update which tried to better align how we performed math about contracts with their economic impact in the actual world. But even though the businesses were small, revenue recognition still came up.

A fun example: many SaaS companies operate against a non-revenue metric called, to the great disgust of accountants, monthly recurring revenue (MRR). MRR is the number SaaS entrepreneurs selling on the low-touch model live and die by.

SaaS companies that sell month-to-month services also frequently offer annual plans, generally trading the knowledge that the customer won’t churn for a year and the upfront use of cash for a discount. And then they tie themselves into knots trying to build dashboards which track MRR but adjust for the impact of a 12 month plan sold for the price of 10 months. Almost nobody gets this right.

Frequently seen bugs include adding all 10 months of the pricing into MRR (will overstate it by a lot since the user will, by definition, not pay you next month), adding a full-priced subscription worth of MRR despite the 2 free months, or having MRR dip during the last two months of the year despite there being no change in the customer relationship.

Accounting says “We have a simple solution for this. You should recognize revenue for all your plans, monthly or annual, on a daily basis over the lifetime of the period the customer is committed to using you. If you want to predict how much money you’ll make next month, you’re almost past what accounting can do for you, but to a first approximation try multiplying your last day’s revenue by the number of days in the next month.”

So what happens to the money which is received but not recognized as revenue? It gets a different colour: deferred revenue. Accounting says deferred revenue is a liability; it is something you owe society, which you will repay by either servicing your customers as agreed or returning their money.

Who cares about what the balance sheet says? Acquirers, for one. SaaS companies are largely valued based on revenue multiples. (At the smaller end of the scale, technically seller discretionary earnings multiples.)

When I sold my SaaS business which had a mix of monthly and annual contracts, the buyer sensibly didn’t want to pay similar prices for differently coloured dollars. The MRR would almost certainly (net of churn) recur next month, but the annual contracts had (on average) about six months left before the buyer would get paid, and they’d be responsible for performance of those contracts immediately. They’d still have to keep the servers running, pay the CS staff, etc. And so that revenue was valued lower. (This surprises public markets SaaS investors and frequently VCs, because after a certain scale annual contracts are more valuable than monthly contracts all else equal.)

I work for Stripe, which has a Revenue Recognition product that helps do the heavy lifting of prorating subscriptions over their lifetimes. (As always, this is my own space and these are my own opinions.) This has exposed me to an even geekier accounting question: how do you recognize revenue for sale of imaginary swords?

Revenue recognition in virtual goods

Computer games are one of my hobbies. Mobile games are one of my vices. In particular, recently I have been playing a “merge” game. Merge games take resource management games, a genre which can be wonderfully intellectually satisfying explorations of complex systems with an undercurrent of Chinese techno-optimism, take out all the intellectual content, and replace it with a Skinner box to drive monetization.

Which is easy to write when I’m sitting at my computer, just like I can easily write that Coca Cola has no nutritional value, harms health by default, and is terrible for teeth. But sometimes I find myself hankering for that sweet effervescent siren. And sometimes a game like Merge Adventure Puzzle Master hooks me on allowing me to brainlessly click my way through appealing fantasy art.

Which is why I have this screenshot of an in-game offer handy:

(For the benefit of readers who might not see the above image, it is a goblin offering to build a bank for 1,000 gems. The bank will produce 10 gems a day forever. Gems are the game's premium currency and available for a variety of price points, including 1,500 gems for ~$10.)

Now here is a fascinating question: how do you recognize revenue for selling a perpetual bond with a 1% daily coupon denominated in a virtual currency?

To answer it, let’s take a brief detour through the fascinating world of virtual goods revenue recognition [PDF by PwC]. It’s gloriously geeky.

Merge Master, like many mobile games, sells a premium currency (gem) for real money. You might naturally think “OK, so when you give them money and they give you gems, you’re even. That payment is revenue within a second after they credit you the gems.”

That’s a reasonable first guess. It is the first guess of every engineer who has ever considered this question, including myself. So if you had it, you were in excellent company in being totally wrong.

It is wrong because accounting tracks the world as it actually exists, and in that world, I’m not paying Merge Master for gems. I don’t want gems. I want the things gems let me get access to in the game I trust Merge Master will continue making available to me. A necessary implicit part of our bargain is “You’re going to be around in a minute so that I can trade in my imaginary gems for imaginary swords to kill the imaginary dragon, right?”

And so virtual goods accounting for goods bifurcates around the question of whether the goods have a necessary services component in them. If you sell someone e.g. a downloadable e-book, you can recognize the revenue about as quickly as you can if you sell them a real book at a bookstore. (Immediately, modulo their right to return it, which you’ll book a reserve against based on your prior data about… sorry, I love this stuff.)

If you sell someone a virtual good with an embedded service element, there are broadly two treatments based on whether the virtual good is consumable or durable. I like to think of this as potions and swords.

Accounting for potions: This is fairly easy. If you sell someone a potion, revenue is recognized when they drink the potion. Or use their speed boost. Or skip the progress gate to get to the next dungeon. The fiction doesn’t matter. Reality matters, and the economic reality of the situation is that performance has happened after the temporary thing you’ve promised is delivered. (Technically speaking you do have to recognize the revenue over time if your potions last long enough to be close to monthly SaaS contracts, but practically speaking most are over with in a day and most accounting systems lose precision below that.)

Accounting for swords: If you trade gems for swords then you can ratably recognize the purchase price of the gems when you satisfy your obligation by giving the player a new sword.

Hah, just kidding. That would be way too easy.

You actually need to recognize the prorated cost of the gems exchanged for the sword over the economically useful life of the imaginary sword. “Economically useful life” is a concept with a lot of prior art on it in accounting. You are obligated to, and accountants can point to substantial work on, estimating the economically useful life of factories, cruise ships, CNC machines, bunk beds, Bitcoin miners, dairy cattle, and almost everything else that depreciates.

But there is not a huge amount of prior art on imaginary swords. So you get to pick one of two methods..

The first, by far less commonly used, is to put your head together with very expensive accounting professionals and rigorously answer the question “What events in reality, in the universe we actually live in, would cause the owner of this imaginary sword to believe it had no or de minimis future value to them?” And perhaps that conversation would involve questions like power creep, game balance changes, declining player preference for swords now that you’re offering a sale on imaginary nuclear weapons, etc.

Nobody has time for that conversation or the truly gargantuan amount of implementation engineering required to enforce the decision it comes up with, so they largely use door #2: the economically useful life of a virtual good is by nature upper bounded by the economically useful life of a player’s relationship with our game, so use that instead.

You are required to use your existing data to make a reasonable estimate of how long either the particular player or, failing that, the hypothetical spherical frictionless average player will continue to play the game after the purchase. Then you recognize the price of the sword ratably over that time period.

This causes many virtual goods companies to have bookings (player purchases) diverge sharply from revenue. That depresses the value of their companies, in the real world, and those companies then spend substantial amounts of professional labor taking their frustration out on imaginary swords.

Game mechanics as accounting optimizations

If you’re familiar with gatcha games like Genshin Impact, you have probably seen game systems which require sacrificing multiple copies of a character or item to get a new, slightly better version of the same character or item.

Some people cynically believe this exploits the player’s inability to understand how exponential math works. “Eight levels of upgrades, each burning two imaginary swords, requires buying 256 swords, not 16 swords. If Sword of Awesomeness was honest about its price, no one would actually choose to buy it!” There is some amount of truth to this.

But again, if you sell someone a Sword of Awesomeness for value, you have to recognize the revenue ratably over the future economic life of the Sword of Awesomeness. If on the other hand you tell someone to burn two Swords of Slightly Less Awesome to get their Sword of Awesomeness, at least  one (and maybe two, depending on your accountant’s opinion) of the Swords of Slightly Less Awesome has no future useful economic life.

Which means you can recognize the revenue immediately. Which means that your large, perhaps publicly traded video game company is worth more, no longer weighed down by the unearned revenue liability that was the existence of the now-burnt Sword of Slightly Less Awesome.

And that is why accounting standards have destroyed more imaginary swords than all the rust monsters in all the prime material planes combined.

So about that goblin

Alright, let’s return to the goblin banker with the improbably underpriced 1% daily yield non-resellable perpetuity bonds.

Is the “bank” a sword or potion? Sword, clearly.

Do you have to somehow account for the liability implied by the bond? No, you do not, because the bond is a fiction and accounting is concerned with reality. Accountants working on a production of the Merchant of Venice don’t recognize revenue during Shylock’s soliloquies; their gaze is on the audience, not the stage, and the audience has been satisfied once the play is over. (I will note that, unfortunately, many depictions of Shylock and many depictions of goblins share some baggage, in a fashion which many of you already noticed.)

The economic substance of the goblin transaction is not “I will give you an infinite amount of future value in return for a current payment.” The economic substance is “If you pre-pay us now, we will give you a variable discount on your future purchases of in-game items for the duration you choose to play this game.”

And thus the answer is, somewhat boringly, that you recognize revenue for the goblin bank ratably over the course of the player’s predicted future lifetime (again, unless you do substantial professional services work to establish that the 10 gems per day is no longer valuable to the player).

Back to the real world

Why does any of this matter? One reason is that systems which operate with money often grow in scope over time, as they have to become increasingly aware of the colour of money. Revenue recognition is just a single example. Some money is KYC-rich and some money is KYC-blind. The risk ratings of money would fill an entire spectrum of colour.

This is important both for consumers of systems which operate on money (i.e. everyone) and for people who build and adopt those systems. It is crucially important, on a local level, that you pick systems which can actually perceive the colours which you have reason to care about.

On a societal level, the more colour there is in the world and the more people have to care about it, the greater incentive there is to buy systems which manage money rather than attempting to build them. If your model of money is “It’s just numbers, math operates trivially on numbers, let’s just build a spreadsheet or otherwise simple system and call it a day”, then money problems often look easy. This is extremely deceptive, and as you grow to perceive more colours around you, you will often come to regret that the spreadsheet has colour vision deficiency. This frequently results in you having to update it, sometimes for very large piles of greenbacks.

This is an interesting thing for policymakers, who can invent new colours and frequently do, to keep in mind. Every additional colour introduced to the world is a vote that there be less systems better architected to perceive it. That sometimes exists in direct tension with other policy goals, such as e.g. promoting competition, discouraging centralization or correlated failures, and similar.

And yet, this column is called Bits about Money rather than Atoms about Money both because it is a better pun and because software has historically had much more difficulty interfacing with atoms than it does interfacing with bits. This is changing in ways which are underappreciated, including in payments.

Voucher payments

A very frequent concern, as more economic activity migrates online, is whether we have a good access story for people who don’t have networked payment methods (like e.g. credit cards). In emerging markets, and (perhaps surprisingly) here in Japan, a common way to make online payments with physical cash is to use a voucher payment system.

Here in Japan they are mostly conducted in convenience stores (known universally as konbini), but they are also huge in Mexico (at the Oxxo convenience store chain and branded as such), in Brazil (through the boleto system), and elsewhere.

(This might be a good time to mention that this is obviously relevant to my day job at Stripe, which supports Oxxo, boleto, and as of this week konbini payments. As always, these are my own opinions.)

The general user experience is quite similar in every nation: you, the customer, agree on a transaction with some business or government entity. Here in Japan this can be almost literally anything up to about $3,000, from buying movie tickets to paying for an Amazon order to submitting one’s residence tax payment (if you are e.g. self-employed and so don’t have it deducted from your salary).

The business gives you a transaction reference number to bring to the konbini, generally advising you to print it out with barcode and instructions or, in some cases, physically mailing the same to you. You go to the konbini and follow the instructions, which depending on the chain are generally either “Show this to the clerk” or “Go to the automated kiosk in the konbini, select Make a Payment, key in this reference number, and take the piece of paper the kiosk prints out to the clerk.”

You then pay the clerk the exact pre-agreed upon amount of cash, and receive a receipt for it. From the user perspective, you’re done.

Why is this so special?

Konbini are deeply woven into the fabric of Japanese life, and I like to think of them as privately owned public infrastructure. You can rely on there being a konbini close to you, virtually everywhere in Japan, and since this is true for you, it is true for everyone you might do business with in Japan. There are seven within a seven minute walk of my apartment here in Tokyo, but there’s also one in all but the most isolated communities. They’re open 24/7, which is an annoying expectation for some rural franchisees but considered fairly core to the brand promise by the chains.

Konbini welcome all comers. People in our social class often do not appreciate that this is not true of all businesses in society, including banks. You don’t need a brand-name employer, good credit, stable housing, government ID, or even literacy to use most konbini services. If you’ve got cash, they’ll take it.

This is extremely unlike credit card issuers. And it is the dominant reason why people, once they hear about voucher payments, viscerally understand why they are so popular in emerging markets.

So what explains their popularity in Japan, which is a highly developed nation? You can probably immediately intuit why e.g. immigrants and tourists have YouTube explainers in every language imaginable for konbini services. Those reasons extend to members of Japanese society who exist at socioeconomic margins.

But konbini are used, extremely frequently, by all socioeconomic strata. Prior to the pandemic, they had about 20% share of B2C online payments.

In 2005, you could have guessed “This is basically downstream of low credit card penetration” and that would have been very plausible, but the financial industry has done a great job of pushing credit and debit cards since then, and almost all banked consumers have access to one. (The most effective single salesman for the credit card in Japan? Steve Jobs. Recurring bank debits are a pain here and receiving a paper bill in the mail monthly is annoying, so cell phone bills specifically were the transaction that motivated many card applications.)

Konbini continue to thrive even with high card penetration. One reason is lingering fears of abuse of cards if used online. Japan is a high trust society, but remote commerce has burned many Japanese people. One particularly common version was the behavior that is described in the seedier sides of the affiliate marketplace as “rebilling”, where a consumer intended to buy a teaser e.g. beauty supplement and unwittingly ended up with a commitment to purchase them monthly with no clear way to cancel. This was a common enough form of abuse that Japan passed a toothy law [PDF] against it in 1976, when the abuse largely happened via traveling salesmen or the telephone, and it hasn't gone away in the Internet age.

Accordingly, many consumers, particularly older consumers, prefer “push” payment methods rather than “pull” payment methods. You will never pay a yen more than what you expect for a convenience store payment, and you’ll never get charged again unless you go back to the store and pass more cash to the clerk.

Another reason is user privacy. We often think of privacy as a right against uninvolved third parties, intermediaries, or perhaps the business one is purchasing something from, but an underappreciated aspect of privacy is a right against members of one’s own household. This is true worldwide, but is particularly salient because of some social features of Japan, such as multi-generational households being more common, many adults living with their parents prior to marriage, and sometimes contentious norms about division of financial responsibility within marriages.

(The traditional normative expectation is that a salaryman earns his salary and immediately gives all of it to his wife for management, and she gives him a small allowance, to prevent spendthrift behavior. This often surprises foreigners, who would not predict this based on their assumptions about gender relations. Cultures are complex; film at 11. As with many traditional norms, this is in flux, far less hegemonic in 2022 than it may have been previously, and was sometimes observed in the breach even at the height of its practice.)

So you can go to a konbini and purchase anything you want without having to justify that transaction to anyone who reads your mail. You can also receive packages at a convenience store, which is a service they’ll happily extend for free in most cases. This is a win for privacy in some cases but is much more commonly used because the konbini is, again, reliably open 24/7. If you work irregular schedules and might not have someone home to receive a package when your friendly local logistics firm comes calling, send it to your local konbini instead and you can pick it up at your konbiniensu.

This creates a compelling loop of user behavior for all parties. Purchase something online, pay for it at the konbini, wait for it to ship, pick it up at the konbini. The konbini wins because you come in twice, and will likely pick up a drink or snack or something to make the most of your trips. Plus, like all of their offerings, increasing your affiliation with their store specifically helps against the brutal competition for user habit. There are more than ~50,000 konbini in Japan and anything they can do to convince you to come back versus using a competitor is worth doing.

The fascinating underbelly of cash management

After a konbini takes cash, what actually happens?

From a settlement perspective it is fairly straightforward. Konbini payments are operated by networks, of which there are a handful. The networks are overlays on top of individual konbini chains, and share financial infrastructure which the smaller chains couldn’t justify building out. At the point where the user makes the payment, the receiving store owes money to the network. These debts are quickly netted out using the standard banking system. The business receives a notification that the payment was processed “relatively soon” after it happens, and their payment is delivered via bank transfer, generally a few days later.

So what happens to the cash? Konbini historically have had huge cash management challenges, and the solution is genius: buy a bank. Or, if you’re a smaller chain, partner with one.

In both cases the objective is to put an ATM in each of your stores. Your franchisee can then recycle the cash they receive by depositing it into their bank account from the ATM in their own store. Instead of paying an armored car company to come drain the tills periodically, the franchisee and chain earn ATM revenue when payday rolls around. (Payday in Japan is the 25th monthly, and much of the country turns their salary transfer into cash on that day. This causes such a wild swing in cash inventory levels that banks arrange for deliveries of hundreds of thousands of dollars to each of their ATMs in advance of it. It’s a fun logistical challenge of the sort that Japan generally eats for breakfast.)

Being able to mix collected cash into the take of the store and make profitable operational use of it is one reason why konbini payments are so efficient. They’re offered free to consumers in Japan; the cost to businesses is broadly in line with cards.

A less happy detail of konbini payments historically is that you had to be a large business to gain access to them, because you need to do a business negotiation with each konbini network, and their sales processes assume that you’re only worth talking to you if you’re bringing in several hundred million yen worth of payments annually. One nice thing about aggregators like Stripe is that we can amortize negotiation, financial plumbing, and custom engineering work over many, many customers in parallel, and so offer them to any business capable of filling out a brief form and meeting the chains’ requirements. These are within the reach of e.g. Japanese sole proprietors within days of starting their business. Good news for the konbini, the businesses that can use them for the first time, the users who can now pay their preferred way, and of course us.

Making good experiences better

The user experience of konbini themselves is fantastic, so much so that there is a thriving genre on Youtube of tourists gushing about them. The user experience of konbini payments has historically had two really brittle moments in it: moving the transaction from the screen to the store, and incorporating the transaction into the business’ operations.

Most large businesses in Japan implemented konbini in the late 90s and have not substantially updated their konbini flows since then. (Much like in the U.S., checkout flows are designed by people who have idiosyncratic beliefs about checkout flows. For example, if you work in payments, you are almost certainly a credit card geek and can gush about the relative benefits of your favorite three reward programs. As as a result, you’ve probably not used cash-on-delivery or konbini payment or similar in years, and you haven’t even checked your own site for how these work because why would you.)

Back in the late 90s, the transaction was almost certainly happening on a desktop, and the dominant convention was telling the user to print out a particular page to take to the konbini. Web development in Japan didn’t stop for 25 years, and as companies updated their checkout flows the konbini confirmation page became technical debt, and so often e.g. moved to a separate, out of the way location where it still looks like it did in 1996. This is true of firms which you would assume it could not possibly be true about.

Anyhow, in the last few years Japan has introduced this magical technology called a cell phone. (Perhaps you’ve heard of it.) One thing you can do with a phone is carry a web page on it while you walk around town. Stripe’s konbini implementation is particularly slick in that we host that page for users (they can, naturally, build their own if they want, but again the largest and most sophisticated companies in Japan can’t convince their teams to do this work), and gave it features like being able to dynamically swap convenience store chains on the fly. Thus, if you thought you were going to go to the Family Mart next to your office but instead found yourself walking into a Lawson for a quick coffee, one tap and you’re done. (Incredibly, this previously required canceling your order and redoing it in many implementations.)

The other backend improvement supports operations. Many e-commerce merchants structure themselves, technically and operationally, around the assumption that payment is captured at the time an order is made. Konbini payments are one type of asynchronous payment; they’ll be captured hours or days after the order. This introduces two different speed issues at different scales.

The fast one: Users expect confirmation that the thing they just did will have the effect they want very quickly. Historically, konbini networks passed payment confirmations to merchants over a download-the-CSV-file-periodically-via-SFTP methods which added a few hours of latency between the user handing over their cash and the merchant sending them an email saying they would ship the order. That’s a terrible experience; we replace it with webhooks that can move at the speed of typical Internet transactions.

The slow one: You have to make changes to your operations to support asynchronous payment, like having some abstraction for a good order which isn’t paid for yet. That might require e.g. reserving the stock for the order. There are also some subtle downstream operational consequences, like issuing refunds.

Card networks have refunds built in. Konbini payments do not; you can’t tell the user “Can you go to the store an ask for your money back?” But, of course, the payment isn’t final because no payment is final. If the customer returns the goods to you, you will want to (and in Japan, due to the aforementioned law, possibly have a duty to) refund them.

We built a fully automated service for this into our implementation: you can click to refund a purchase from the Dashboard (or trigger it via the API), and we’ll email the user to collect bank account information to send them the refund. This saves you from having to have your operations team send several calls or emails to get the information necessary to return what is, probably, a relatively small sum, in the majority case where the user has access to a bank account or postal bank account. (You can have your operations team send e.g. a postal money order to them if this doesn’t work for any reason, or otherwise solve this yourself.)

Distributed identity verification

Another example of an offline/online convergence is distributed identity verification. One way this is happening is through automated ingestion of IDs and selfies via smartphones, and that is such a fascinatingly deep rabbit hole I think I’ll do an essay on it alone someday.

But smartphones aren’t the only physical technology that is improving rapidly. Shenzhen is transforming our physical environment worldwide, including here in Japan, by making bespoke hardware development orders of magnitude quicker and cheaper than it has been historically. And some of that hardware is designed to bridge the gap between governments and businesses consuming their services.

Governments got into the identity game to raise armies and taxes to pay for them, but in the millennia since then, they offer identity as a service to the wider community. This is widely underappreciated; one reason the DMV in the U.S. is as annoying as it is is because it is the unsung lynchpin in this service, which almost every business consumes in some fashion.

Japan has drivers licenses but they’re less hegemonic as a proof of identity here than they are in the U.S. You can easily find middle class Japanese people without them, for example, where that would be uncommon in the U.S.

Japan also has some higher ceremony identity / authorization proofs, which are necessary to explain before I tell you about the really cool machine at the local konbini.

When I bought my wife a car in Japan, the dealership understandably needed to confirm that I was actually me prior to binding me to the purchase and a substantial loan. I don’t have a driver’s license, and while I do have government issued identification, for high-value transactions, employment, or similar Japanese businesses will often ask you to get either a certificate of residence or a certificate proving registration of your personal seal.

These are issued by your town hall for a small fee. They provide a set of proofs. One is your claimed name and address match what is on file at town hall. One is that you went to town hall and convinced a competent Japanese bureaucrat of your identity. One is that if you are fibbing about any of this you are now defrauding the Japanese state, not merely your counterparty. And one is that you have very high intent to engage in a non-routine transaction that required you to go to town hall.

Why is that last one important? Because licenses and other forms of ID in your wallet are used promiscuously in society and so they leak. Thousands of businesses have seen my ID before. Some of them doubtless have a copy on file. Many of them will lose that copy to a hacker.

So it is not necessarily a given that anyone presenting a copy of something that looks plausibly like my ID is necessarily me, or that a woman saying “This is my husband, he’s totally in agreement about purchasing this car, and here is his stamp and ID” is necessarily my wife and actually authorized to commit me to purchasing a car.

Thus requiring a certificate from town hall. They’re single use by design, unlike IDs, and your counterparty will invariably require a recently issued one. And they prove that I personally went to town hall (or sent a legal designate with some appropriate level of ceremony) with intent to engage in a non-routine transaction. (Town hall generally doesn’t inquire into specifics, unless e.g. you’re a senior citizen taking instructions from a cell phone and they’re worried you might be being swindled.)

That’s the backstory why this exists. Notice that it introduces pain for myself, town hall (which has to staff a line continuously to deal with these requests), and the dealership (which will lose the transaction if I can’t organize myself to do this, which hurts but hurts less in expectation than fraud).

Now here’s the cool machine: My local konbini has a multifunction printer/scanner kiosk that can do about a hundred things. One of those things is that it can read a chip in my My Number card (Japan's I-suppose-I-can't-call-it-new-anymore-even-though-I-got-here-first national ID), ask me for my password, authenticate it with the chip, connect to town hall and present them a certificate request signed by the chip, and print out the certificate for me. This is more convenient than trudging to town hall, has no office hours, required no highly trained public servant, and is approximately as secure.

Moreover, while it is difficult to get that security out of a My Number card remotely, one can request someone submit a certificate remotely. (And, if one cared to, one could call up town hall, read them the certificate number, and ask them to confirm that their copy reads the same as the copy received from a customer. In practice, most Japanese businesses don’t bother calling, because the certificate is such good evidence of benign intent.)

This thus turns the konbini into a sort of offline/online hub connecting and solemnizing remote transactions with a built-in guarantee that anyone committing fraud is doing so against the government, on camera, and (probably) close to where the police can make their acquaintance.

That, in turn, makes certain transactions viable remotely, or partially remotely, which would have required multiple in-person visits before. Certificates move through the mail easily and can also be trivially captured via the camera on one’s phone, allowing one to do higher-ceremony transactions without needing to physically meet anyone to verify ID.

There’s always more to discover

A very smart person once heard me gushing about these systems and said “... They should all be web apps.”

One intuition there, that the world should become more convenient and more orchestrated by software, is obviously correct.

However, I’d expect vastly more deployment of online capabilities in offline lives as time goes on. The most important platform for it will continue to be the smartphone, and increasingly “web apps” will be able to treat the camera and other “embedded peripherals” as a first-class I/O device, but (regardless of where you live in the world) you’re going to start seeing some innovations here.

An obvious one: kiosks at banks which will connect to an officer in a call center rather than an in-branch employee. This will eat the class of transactions which are more complicated than a teller can handle but can be done by a branch banker. The economics supporting that use case are absolutely overwhelming; you can stop needing to staff (and massively underutilize) relatively expensive sales professionals in high-cost metros and replace that fraction of their work with ~100% utilized specialized call center employees. It is already in trials in Japan and, mark me on this, it is coming to your neighborhood before 2030.

Software is famously eating the world, but Shenzhen is, too. The faster hardware cycle times get, and the more amenable they become to orchestration at the speed of software cycle times, the more entirely offline processes will migrate to this sort of hybrid existence.

I wish I had enough space here to cover those kiosks, tax payment machines, and other cases in more detail, but that will have to be another issue. Drop me a line if you enjoy geeking out about this stuff.

I will repeat my usual disclaimer here: I work for the Internet at Stripe. These are my own opinions.

We begin with SWIFT begin with a brief diversion into how money moves. Or, more to the point, how it doesn’t.

Correspondent banking

As we’ve covered previously about bank transfers, “moving money” is a misnomer, a simplification which covers a complex coordinated series of offsetting agreements about debts. When you move money domestically, your bank and the recipient’s bank use some intermediary system to coordinate a series of agreements which result in your bank agreeing it owes you less than it did prior and the recipient’s bank agreeing that it owes the recipient more than it did previously.

This same principle is at play in moving money internationally, with one interesting difference: banks largely cannot hold money extraterritorially directly, for most useful values of “directly.” Instead, they rely on a correspondent banking relationship.

Banks can have accounts at other banks, and extremely frequently do. A major reason to do this internationally is to facilitate payments in other currencies and other jurisdictions.

An example which shows the general pattern (with one tiny fib to save a few paragraphs of irrelevant detail): once upon a time, shortly before the global financial crisis, a young American banking at a small institution in Gifu Prefecture, Japan needed to send in his student loan payment to the servicer working for the U.S. government. The U.S. government, somewhat predictably, strongly prefers dollars over yen, and (perhaps less predictably) has incredible difficulty taking payments internationally.

That small institution, which will remain nameless since I still bank with them, holds some dollars on its books (a few hundred million dollars worth) but does not “physically” control more than the tiniest fraction of them. (That tiny fraction is paper dollars which, if you are a Gifuite anticipating a vacation to e.g. Hawaii, you can purchase at your local branch office in small quantities for a fairly hefty spread.) The vast majority of its dollars are owed to it by Mitsubishi UFJ Bank, the largest bank in Japan.

MUFJ is the largest supplier of yen/dollar liquidity in Japan, but it does not have direct access to the U.S. banking system. (In something of an oddity, it does today control a U.S. subsidiary which has full access, but that was not available back in the day.) Instead, it holds accounts at a variety of U.S. banks.

The one which acted as the intermediary bank on the wire (Wachovia) is no longer with us. MUFJ had an account with Wachovia, which is to say that the dollars MUFJ owned were owed to it by that bank. Neither MUFJ nor my own bank had custody of the dollars they were going to move on my behalf.

MUFJ’s intermediary had full access to the U.S. financial system, including to FedWire, which does domestic wire transfers.

When my local bank executed the wire, it passed an instruction to MUFJ, which passed an instruction to Wachovia, which effected a funds transfer through FedWire, which goes through the Federal Reserve, causing Bank of America to be owed slightly more money by the Fed, which it swiftly agreed that it owed me most of (after deducting a fee). And thus an offsetting series of rapid agreements about changes in amounts owed between bilateral counterparties results in me having less yen and the U.S. federal government having more dollars, plus each at least five entities earning a fee.

In broad strokes, this is how correspondent banking has always worked. Note the absence of an explicit technological substrate here: it could be conducted over TCP/IP, by a telegraph, or with a letter carried between countries on horse. And, indeed, all of those have been extensively used in correspondent banking over the centuries.

Now we’ll return to the present, and to SWIFT.

SWIFT: The world’s most expensive encrypted messaging service

SWIFT is a company and often, like Kleenex, used to refer metonymically to their best-known product, which is an extremely specialized low volume encrypted messaging platform properly called FIN.

(There is a bit of geek humor in there, because “5 billion messages a year” sounds like a lot but is inconsequential scale in modern computing. I’ll note that is 160 transactions per second in case anyone wants to compare that to alternative ways to send very low volumes of messages about money movement.)

SWIFT is almost synonymous with international wires because it is the primary way that banks and their correspondents choose to interoperate with respect to wires. Specifically, they send a MT 103 message, which is a bit longer than a tweet, and then each bank operates their internal books and other banking systems to make the request encoded in the message a reality (or fail gracefully).

Very little about this choice is about the technology, per se. The exact specifics of what fields a MT 103 allows you to send are interesting to professionals but are not why SWIFT is important. SWIFT is a multi-layered network effects business.

Being on the platform means you have many more counterparties you can easily reach. Your regulator and auditors are likely far more comfortable with it than they would be with your second most plausible solution. The new head of AML Compliance you’re thinking of hiring certainly has a lot of experience with the operational oddities of SWIFT messages. The advantages continue and compound upon each other, making each marginal financial institution more likely to join SWIFT, and SWIFT more valuable for each institution on it.

Importantly: SWIFT is not actually co-extensive with international money movement. In the first place, it doesn't actually do anything directly to money. Money doesn't travel over SWIFT any more than it travels over napkins, though each could potentially contain an instruction about money that a bank might choose to implement.

SWIFT does not even monopolize the thing that everyone assumes it monopolizes. There is a document in a binder in Tokyo which describes “How to move money to the U.S. if SWIFT is down” and it does not say “Pause the Japanese economy until SWIFT gets their #%()#% together.” MUFJ knows the phone numbers for the U.S. banks holding billions of dollars of their money and can transact with them in any of the ways that you’d expect a bank to make available to a customer with billions of dollars deposited.

SWIFT is in the news in early March 2022 principally because it has been deputized to fulfill some policy aims of the European Union. As of this writing, they’ve cut a subset of Russian banks from the network.

The intent of this policy has been described variously in various places. In my personal opinion, I think the best articulation of the strategy is “We are attempting to convey enormous displeasure while sanctioning some banks which are believed to be close to politically exposed Russians, while not making it impossible for Russian firms generally to transact internationally nor sparking a humanitarian crisis either inside or outside of Russia.”

A subtle but important note: SWIFT does not substantially control who is allowed to send wires. In the above example, each of the three banks involved has independent authority to accept or reject my business, and each of them has independent responsibility from their regulators to enforce the laws of their jurisdiction. (SWIFT has some products which assist with complying with those laws, but doesn’t independently make rules or develop lists of people to exclude from the system.)

Banks and other financial entities are policy arms

A recurring theme for this column is that banks are expected, in return for guaranteed monopolization of some lucrative franchises, to act as policy arms for the governments they are subject to.

SWIFT is, theoretically, a Belgian cooperative. The Federal Reserve is also, theoretically, a joint-stock company owned by member banks and not at all part of the United States government. These are… consensual fictions.

SWIFT also publishes the following consensual fiction, of which the last four words are of particular note:

SWIFT's Board composition is designed to reflect usage of SWIFT messaging services, ensure SWIFT’s global relevance, support its international reach and uphold its strict neutrality.

I have no great expertise in power politics. Reading much commentary on international financial sanctions causes me to believe that some putatively serious people chose to take the consensual fictions literally despite them being shelved next to Harry Potter (and being approximately as accurate regarding how banking is conducted).

Of course SWIFT is a policy arm. Of course it answers to the EU. Of course it can be directed against disfavored individuals and organizations, including governments, including in an indiscriminate manner. Of course this has happened before. Of course all parties knew this could happen again. Of course this is only part of the response.

Conveying “Commander’s Intent”

There is an interesting intellectual framework in the U.S. military called “commander’s intent”, which balances the world’s paradigmatic example of an institution which follows orders from above with the complexity of ground state that is not possibly knowable to the person writing the orders. Commander’s intent is designed to inform subordinates of what you actually desire, so that they can intelligently embellish upon what you order them to do.

Many commentators confuse the actual effects of severing particular banks from SWIFT with what they perceive as the policy goal motivating it. More important than either is, in my opinion, what it communicates about commander’s intent to the policy arms who are responsible for enforcing it.

Specifically, it communicates that Something Has Changed and that Russian institutional money, specifically “oligarch” money, is now tainted, and not in the benignly ignored fashion it has been for most of the last few decades.

Read: “We will with absolute certainty hand out billions of dollars of fines stochastically over the next ten years. You can minimize how many hit your institution by successfully intuiting who is on the Bad Risks list. We will be sharply less tolerant of ‘Cyprus is an EU country and so banked customers in it are per se low risk’, ‘lots of people buy real estate in London and we couldn’t possibly inquire about all of them’, and things which we have previously turned a blind eye to, and we will probably lie about having turned a blind eye to that, and you will, too, if you know what is good for you.”

I have some deep misgivings about policymaking through “try things and see what we fine”, but be that as it may, it is absolutely pervasive in regulation of the financial industry, particularly with regards to anti-moneylaundering (AML). Much of the activity which will be sanctioned will be sanctioned under AML regimes, because they’re on the books, have existing and relatively capable agents charged with enforcing them daily, are enforced against all human activity which touches money, and can create a “process crime” out of almost any underlying predicate (including ones which might be difficult to criminalize directly).

One reason why SWIFT is currently in the news, and routine updates to your friendly local bank’s AML compliance division from their regulator are not, is simply that SWIFT is, as a nearly pervasive infrastructure layer, one-stop-shopping from the perspective of political actors. But your friendly neighborhood compliance officer is assumed to be reading this news carefully.

Tiniest bit of personal opinion

I am in favor of discouraging offensive wars, but know the financial system to be a very blunt instrument for expressing society’s preferences. In particular, while the global financial system and the governments giving them instructions have not actually banned the usage of oligarchs’ wealth prior to now, they have already given many ordinary Russians (and ordinary Ukranians!) severe difficulties doing things we expect to be utterly routine for law-abiding members of society.

Without divulging any professional confidences, in the wake of the 2014 invasion of the Crimean Peninsula many U.S. banks decided to stop serving customers with Ukrainian passports. No one explicitly made the decision “Your nation got invaded, so you should have less access to financial systems half a world away. This is a natural and just outcome in a democratic society.”

It flowed indirectly through “The Crimea now poses a heightened risk of money laundering”, “We lack the ability to discriminate between the Crimea and the rest of Ukraine”, “We care a lot more about not facilitating money laundering than we do about our infinitesimal Ukraine business so Ukraine is going on the High Risk Country list”, “Sorry, you have citizenship from the High Risk Country list, accordingly I’m not allowed to open this account for you. This is a commercial decision of the bank and will not be reversed.”

Maddeningly, no one—not the regulators, not Compliance, not the front-line employee delivering the decision—believes they are accountable for this result! Which happened! Tens! Of! Thousands! Of! Times!

Adverse actions against innocent individuals will certainly increase as a result of the current conflict. Indeed, some intemperate commentators are calling for harming individual Russians as a policy goal in and of itself.

The logic of collective responsibility for the actions of a despot is justly disfavored. I urgently hope that civil society chooses narrowly drawn regimes to express its preferences, and that it prefers legible and democratically accountable lawmaking over kicking the decisions to Compliance and taking no responsibility for what Compliance comes back with.

Almost no SaaS company I used had these problems.

There are various pitches for embedded finance for consumers (you can sort of cast BNPLs in that light if you squint a bit), but the big pitch for embedded finance is: the SaaS platform your business runs on can arrange financial services for you, which will feel bank-like but better.

It is now time for my usual disclaimer: this is quite relevant to my professional interests. I work at Stripe, which has products for platforms to offer financial services and cards to their users. That said, my opinions are my own opinions, as always.

Why is SMB banking underwhelming in general?

So a good first-pass model for the behavior of banks is that there are some things they do which are lucrative and some things they do because they are the traditional business of banking. Experiences for the first category of things are often really, really good.

Issuing credit cards to retail users, for example, is an extremely lucrative business. The bank will often pay you to use the product. You can sign up for it at 2 AM in the morning. You can be approved in less than a minute without talking to anyone. The mobile app is (if not a heartbreaking work of staggering genius) reasonably functional. If you have any questions you can call the number on the back of the card and get a responsive answer very, very quickly from a human if necessary.

Banks also bank small businesses, but many do this out of a sense of societal obligation. This is sometimes organic from the culture that is banking. This is sometimes out of a more explicit quid pro quo. Society grants banks an exclusive license to engage in a number of extremely valuable activities, such as making consumer loans and funding loans with deposits. In return for this, society (and, specifically, banking regulators) demands a number of things.

One price of admission: you must have some offering for a customer who comes in and says “I just opened a laundromat… what happens next?”

Every entrepreneur who opens a laundromat is providing a valuable service to their community and should be celebrated. However, from a bank’s perspective… this is not the most exciting client in the world to have. A full accounting of why is outside the scope of this post, but some of the issues:

KYC (Know Your Customer): The bank will need to spend expensive staff time getting intimately acquainted with a business which is likely somewhat informal and less-legible than regulators expect it to be.

High servicing cost: The laundromat will have frequent dealing with coins, which are annoying and costly for banks, and which are likely under-monetized for a variety of reasons (this is often something society expects banks to just take care of because, after all, we gave them stewardship over money and coins are money). The laundromat will, likely, consume quite a bit of staff time on an ongoing basis.

Low revenue potential: The laundromat’s primary value to the bank is as a source of cheap deposits, but it will never deposit a lot of money; most of what it earns every month will go out the door.  The laundromat does have credit needs but the bank mostly cannot serve them profitably, because the nature of the business is risky, the business is informal and illegible, the entrepreneur likely does not have legible credit or assets or earning power outside of the business, and the bank does not have any computer program that can spit out a decision at 2 AM for seven cents but rather needs to have expensive human underwriters spend expensive professional time on a business which will probably not consume enough services to justify that time.

This is not a unique failing of banks! If laundromats had to hire top-flight engineering consultants to build internal software to operate their businesses, those consultants would have very similar issues servicing them. Which is why laundromats mostly don’t do bespoke software development.

They instead snap into SaaS, which love their business and can service it profitably.

SaaS platforms are rewriting the operations of small businesses worldwide. This is a very the-future-is-here-but-not-evenly-distributed trend; businesses which many technically inclined people would describe as “legacy” and “not tech-forward” will almost all run on software. It is going to happen much faster than consensus estimates.

Some SaaS companies will also become major financial services providers.

SMB-focused platforms are moving into financial services

Take Shopify, for example, which supports more than two million merchants, the vast majority of which are Main Street businesses which happen to exist at least partially on the Internet. (Shopify is a customer of my employer and is also a very large publicly traded company; all of the facts about their business can be sourced from their public reports.)

Their core business, historically, has been providing SaaS to let businesses that could never hire an engineer still have a competent web front-end. The vast majority of their customers pay less than $50 a month for this.

Shopify’s brief is to make their merchants more successful at selling more things to more customers. Payments is a large and growing portion of their business now, but at the core they’re a software company, and programmatic money movement is a very, very interesting capability for a software company to have.

The nature of payments has historically been that money moves from a customer through some pipes coordinated by a competently-executed software platform into a bank account before the business can actually use it. That last hop introduces delays, costs, and a lot of friction.

Many businesses which use Shopify eventually treat it as the nerve center of their business. It’s a special case of a pattern we see often in more vertical SaaS: “morally an ERP, in addition to the thing more commonly associated with it.” The view into the business exposed by the dashboard and things linked on it is the best legible reality of the business, as far as the owner is concerned.

The hop into the banking system necessarily causes that view of the business and the actual reality of the business to diverge. Reconciling them causes a lot of pain for the managers, for bookkeepers / accountants, and for anyone (banks, tax agencies, potential acquirers, etc) who need to understand the reality of the business.

If Shopify arranged for the provision of financial services, this final context-shedding hop would not be necessary. And so now they do: the offering is called Shopify Balance (which is powered in part by Stripe Treasury). It does the things a business wants to do with money, like spend it, hold it, and send it, but does them more natively in the Shopify context than if the business had to build an ad hoc underspecified version of half of it to interface with the bank on the corner.

Why is this structurally innovative?

It's probably fairly intuitive that most SaaS platforms will ship better software than most banks, but banks can (and do) publish some pretty amazing software. It is just far more incentive compatible for a platform to do it than for a bank to.

One reason is concentration of their customer base. Small business banking is, by necessity, a very, very broad offering. I previously ran several small businesses which banked (in part) with a large U.S. bank, chosen by the very sophisticated SMB owner process of “They had the branch in the food court where I went to college, many years ago, and so of course I walked into one of their branches and asked ‘Uh I think I am actually in business now, what do you have for me.’”

The bank helpfully offered my small SaaS company exactly what they would have offered a landlord, or a dentist, or a freelance copywriter, or a towing firm, or a budding e-commerce company. My business has very different needs from all of those. The bank knows this, because they are extremely intelligent professionals, and cannot act upon it. Society expects them to have a responsive offering for any licit business which walks in the door, and they have an offering, and it is what it is. They cannot customize that offering to exactly match the diversity of needs of any business in the economy. A hundred thousand engineers could not do that.

Their one-size-fits-none offering was relatively expensive. I paid $15 a month for basic business banking. The terms on the business’ credit cards were uniformly worse than the ones the same bank offered to me on my personal cards, despite the personal guarantee, and despite those cards being more lucrative for the bank structurally (due to spend levels and interchange rates). I’m a businessman and I don’t begrudge them their prices, but… I’d happily have moved to a better, cheaper offering.

SaaS platforms don’t have to serve everyone in the economy. They can make very large businesses off of a very small subset, and make their products sing for that subset. Shopify (presumably?) doesn’t have to worry too much about dentists or landlords.

Banks are able or willing to make some services free because they support the rest of the business (for example, the aforementioned coin management). SaaS platforms have the same capability, and because financial services are so sticky, platforms are often willing to “give them away” along with their base offering. BigBank had a five-prong test to waive the $15 a month fee for basic banking; Shopify will happily staple a free card to every new account.

Many small businesses are extremely sensitive to their cash cycle. The few days it takes payouts to land and become spendable are a frequent source of managerial headaches. With tighter integration between the platform handling money collection and the first place the money is spendable, that delay can be almost arbitrarily short.

BigBank has no system in place to see “OK if you just had a purchase on your website then probabilistically you should receive $48 in four days so, hmm, modulo credit risk and cost of funding, I should just advance you $48 to keep you happy”; Shopify and friends can build that. (Shopify offers next-business-day payouts to Shopify Balance.)

(There’s a fun sidenote about creating overlay networks on top of the existing financial system which I will have to cover some other time, but the backend of how this works is fractally interesting.)

How does this make business sense for platforms?

Platforms are rapidly becoming the OSes of businesses running on them. In some ways, I think of them as akin to franchising systems. A typical franchisee is simultaneously a provider of capital, labor, and entrepreneurial skill. The franchise’s operating document and systems attempts to increase their entrepreneurial capabilities by walking them through the huge list of things that they’re suddenly responsible for, from hiring to taxes.

SaaS platforms are evolving in this direction, where their software, services, educational offerings, and similar provide much of the infrastructure that a franchise does, but at wildly lower total take rates and for far less capital investment.

Adding financial services makes all of their offerings better. It makes them stickier. It can even make businesses more effective in a scalable fashion, and given that the platform indexes on the aggregate success of their users, figuring out e.g. “X% of our users and a disproportionate number of underrepresented founders report access to capital is holding them back; what can we do to solve that?” shows up on their dashboards very quickly.

There is also a direct revenue model for the platforms. One major component is interchange on the businesses’ card spend. For reasons which will be obvious to you if you’ve read previous installments of Bits about Money, the cards will be invariably issued by a Durbin exempt institution.

Just like interchange allows one’s suppliers to bid down one’s cost of capital, interchange also allows one’s suppliers to bid down one’s cost of compelling SaaS services.

This dynamic is extremely underappreciated in most popular discussions of interchange, by the way. Interchange historically moves the cost of credit from the purchaser to the business servicing them, by compensating the card issuer such that they can afford to offer credit at lower cost. This is the core engine of BNPLs; in the limit case extending credit can be free to the user of it. Businesses partially pay interchange specifically because they want their customers to come back more often and buy more and offering cheap credit does that.

An SMB’s non-SaaS vendors should in expectation be thrilled that their SMB customer is using a SaaS platform. It makes that SMB more likely to survive. If that SMB thrives it will buy more of their stuff. The SaaS platform and embedded finance team up to essentially make this subsidization happen automatically, without needing to walk anyone through the internal logic.

So does this cut out “the banks”?

No, mostly it takes advantage of specialization of labor in banking. Most embedded finance products heavily involve regulated financial firms, often including banks. Some are specialty players who have a core competence in this sort of thing, versus banks which have SMB banking in the same division as retail banking but also have e.g. ten thousand people working on capital markets, a wealth management division, etc.

(A plug: the annual report of any bank you use is fascinating reading. If you’re not familiar with how to read them, start with First Republic Bank, which is about as straightforward as a bank can be, and then after you’ve figured out what “simple but large” looks like compare that to any of the large money center banks.)

Banking services can be very bespoke on the high end, which probably seems unbelievable to a laundromat. Wall Street won’t return your calls, let alone negotiate terms with you. But a large enough aggregation of small customers without requiring building out a nationwide branch network is very interesting to even the largest banks in the world.

This can be a have-our-cake-but-eat-it-too for stakeholders in society who care intensely about exactly how money moves around. Large investment banks and money center banks are maximally legible to banking regulators. Smaller speciality banks are still, at the end of the day, banks. The diverse panoply of regulated financial entities stays regulated.

There are some people who would be very worried if “tech companies” ended up holding the money, for a variety of reasons that I mostly disagree with but can at least intellectually understand the underpinnings of. Embedded finance offers an interesting compromise solution: “tech” gets a detailed understanding of customer needs and paints the pixels; regulated entities sign up for the usual list of responsibilities and will respond with alacrity to any inquiries.

There is a potential that this is the best of both worlds. And even better, it’s a choice: the traditional small business bank account is still abundantly available. Walk into almost any branch in the country; they probably have a small business offering. But now they have to earn your business, which is better for everyone.

This trend is very, very early

Although fintech geeks have been quite enthusiastic about embedded finance for a while now, this trend is still very early. One reason why the above writeup is a bit North America centric is that offerings here in Japan, as well as much of the world, are still rather anemic. That is changing rapidly, though.

Core infrastructure takes time to build. Consumers of core infrastructure, though, can expand at the speed of software. Many people think SMB owners are set in their ways. You are welcome to share this observation with your SMB owner of choice over the messaging app on their smartphone and see whether they agree with it.

I think people underrate how fast product cycles and adoption curves will move in this segment, and I think that is also true of most people who are professionally involved in making it happen. The product possibilities are much better than existing alternatives and the channels for selling them already exist. This avoids the frequent bugbear in consumer fintech, which is that it becomes a cost of customer acquisition game against some of the best funded and most savvy marketing teams on the planet.