The Diff - Klarna and the Two Pivots of 2022

In this issue:

• Klarna and the Two Pivots of 2022—Klarna's a good reason to revisit the Buy-Now-Pay-Later model. The company's results are impressive, and, while they are benefiting from AI, there was another factor at around the same time that also accelerated their revenue
• Crypto Adoption—You can probably guess who's most excited about transactions that don't require trusted third parties.
• Art Margin Calls—The real collateral isn't on a canvas.
• Liquidations—Scale wins in commodity businesses.
• Corporate Espionage—A sting catches a corporate spy and eliminates a popular excuse.
• DeepSeek—Consider the scale.

The Diff March 17th 2025 16:15 • Click to play audio

Klarna and the Two Pivots of 2022

Buy-Now-Pay-Later companies serve a very helpful didactic purpose, in that they illustrate that "payments" and "credit" are really the same business. At some level of precision, money changes hands at a different time than goods do, and any intermediary who affects the perceived timing of this is taking some kind of credit risk. When credit is offered in bulk—in a bond underwriting, for example, or a loan—the borrowing is often separate from any one purchasing decision. At a consumer level, though, credit is often directly connected to a purchase, and the overall availability of credit, coupled with the exact structure of the credit (and even the number of clicks, taps, and characters typed in order to access it) all affect whether or not a transaction happens. Credit has been a part of retail for a long time, and has always been a lever to get a little more spending and perhaps some loyalty, too. BNPL is a comparatively recent entrant into this business, very much a creature of the post-crisis period of constrained consumer spending and caution about debt. Over time, it's converged a bit with its competitors, but a new organization built around a new-ish product will tend to come out of such a convergence process with a faster operating cadence and some unique advantages. Now that Klarna has filed to go public, we can take a look at what those advantages are.

Klarna's prospectus opens with the now-traditional letter to investors. The usual formula for this one is to throw in some modesty upfront and then get more bombastic over time.^[1] In Klarna's case, they also throw in some complaints about post-financial crisis banking regulation:

So, wasn't competition supposed to fix [late fees, overdraft penalties, revolving debt, etc.]? Sure, but it was sacrificed at the altar of "financial stability" and "privacy" through misguided regulation that raised barriers to entry.

And then they go on to say that Klarna is equally guilty, that they were born as, basically, a traditional bank that marketed itself online rather than through a branch network, but with no fundamental change to the business model.^[2] Klarna got religion and decided to build a different model, which can be described in admirable terms by highlighting what they try to avoid—late fees and revolving credit^[3]—and can also be described in more businesslike terms by what fundamentally drives their company: transaction-level underwriting, variable pricing based on merchants' willingness to pay for incremental revenue, and a compounding data advantage. One result of this legacy is that Klarna is still a bank, with $9.5bn in deposits. They're slightly better-capitalized than some of their larger peers: European banks had a common equity tier 1 ratio of 15.7% as of Q3 of last year, the big US banks are in the low- to mid-teens, and Klarna's at 16.8%.

The fundamental BNPL model is probably well-understood at this point (see The Diff's writeup of Affirm at the time of that company's IPO for much more). The basic idea is that paying in installments is appealing to some consumers, and leads them to complete transactions they'd otherwise give up on, or to buy a bit more than they otherwise would. Merchants pay variable fees, depending on both the economics of dealing with their customers (transaction size, default rate, perhaps likelihood of future spending through the same BNPL provider) and their own negotiating position. In two-sided networks, you can't really talk about one side having the best value proposition, since the network can only grow sustainably if both sides are roughly balanced. But you can talk about the most intuitive take on it: for credit cards, it's easiest to think of the value proposition for borrowers; for BNPL, the natural focus is on revenue lift for merchants. One way to look at it is that Klarna is still a consumer lender, not necessarily targeting the most creditworthy customers, but it's identified a slice of the market where the interest payment that makes the loan worth it is at least equal to the fee that a merchant would pay to get the incremental business that loan brings in.^[4]

Klarna talks a bit in the prospectus about network effects, and these are always relevant for a payments company. More users onboarded means a higher likelihood of accurate underwriting, which lowers the fee they need to charge merchants in order to turn a profit. But there's a bit more to it than that:

We open our network to a broad consumer and merchant ecosystem, similar to Visa, MasterCard and Amex, but also benefit from our proprietary closed-loop network where we issue, fund, process and settle the entire payment, while retaining a direct relationship with our consumers.

Controlling more of the transaction is a capital- and operationally-intensive proposition, but it's also a way to be a bit more than just a payment service. There's a common pattern where companies get big enough to become banks, but the natural counterpoint is that bank-like businesses can amass enough first-party purchase data to be very good ad platforms, especially for long-tail merchants.

Of course, all of this is pointless if they don't make good loans. There's a lot of fun data science involved in underwriting, and for the right personality type it's a rush to find just the right signal to cut defaults by a basis point, or to figure out the perfect subset of hard-to-lend-to customers who are worth doing business with. But to have that kind of fun, you need to figure out the relatively less stimulating business of getting a big enough sample size. But there are different kinds of sample sizes: swipe a credit card at a physical store, and Chase is just guessing how you got there, what items you considered but didn't buy, etc. Klarna knows, so their useful-bits-per-dollar-of-spend ratio is a bit higher. They actually give some detail on this, noting that their Gini score (a measure of how closely their credit model's ranking of creditworthiness corresponds to realized default rates) rose in the US from 0.36 to 0.72. And, to their credit, they don't point out that this happened after they tightened lending standards in mid-2022, a move that would restrict the range of potential borrowers and thus lower the expected Gini score.

2022 was actually a very important year for Klarna, in two ways. One of these, they're happy to talk about at length. They gave less attention to the other, even though it also had a big impact. The two big macro stories of 2022 were, in retrospect:

1. The inflationary wave that had picked up earlier reached unexpected heights, leading to central banks raising rates from a near-universal near-zero, crushing the value of both long-duration fixed-income products and implicitly long-duration growth stocks, and
2. OpenAI slapped a nice frontend on a model they'd released seven months earlier, turning it into the fastest-growing consumer product in history.

Klarna has a great AI story: they are, of course, a business with a vast amount of data of varying degrees of structured-ness. They have a large base of customers who will periodically misunderstand something about their bill, and it's easy for a single customer support call to be the difference between profit and loss when the loan in question is for $80 or so. They estimate that they saved $39m/year by moving more of their customer service over to an AI model (they note that the bots are, depending on the metric, equal to or better than their human customer service employees). They've been able to cut marketing spend almost 40% since 2022, while still growing revenue, and they attribute some of that to AI as well. And the quantitative kicker for all of this is that their revenue per employee rose from $344k in 2022 to $821k in 2024.

Which is incredible! Both in the sense that it's an objectively impressive result and in the sense that it is just not credible that AI did all of that. Klarna's prospectus is going to be trotted out a lot in the coming months as an example of how quickly AI can drive business efficiencies, if you just know how to use it right, and it's true that they've done that.

But it's also true that in 2022, both Klarna and the broader economy changed in ways that would ramp up revenue per employee even if GPT-2 were state-of-the-art. Here's what they say about the adjustments they made once the equity financing available to high-growth companies was suddenly a finite sum:

In the second half of 2022, we implemented a strategic initiative to adjust our underwriting standards in an effort to improve the overall credit quality of our portfolio. The initiative was driven by our strategic recalibration to a more balanced growth and shift towards profitability. These changes included updates to our credit underwriting decision framework, such as launching new risk models to manage risk return trade-off in line with our profitability targets for 2023, including first-generation new-consumer-level risk models, targeted risk-based down-payment policies, updating decline thresholds following the new model implementation and adjusting our risk-based pricing policies for our consumer loans to drive a higher yield on the portfolio. In particular, we increased the number of consumers that were required to make a down payment in order to take advantage of our financing products.

One reason for the shift in sentiment that forced them to get religion on underwriting standards, and to dial down their growth: interest rates went up in 2022, by a lot! Klarna is a bank, and digital banks tend to have a higher deposit beta—their customers can easily move, and have a very easy time comparison-shopping, so Klarna's cost of capital rises quite quickly when rates go up. (To their credit, they actually run a reverse duration gap: they lock in funds for an average of 280 days, and make loans that mature in an average of 40 days. But that also means that, over the course of a year, their cheap funding mostly rolls off and their loans need to be renewed. And even though these loans are mostly zero-interest in an accounting sense, they still charge economic interest paid for by the merchant. So, of course Klarna's revenue per employee went up when rates went up: the thing they're selling is money, and the price of money spiked starting in 2022!

That said, they've managed the transition gracefully, and they are in fact reporting GAAP profits now, with improving unit economics everywhere you'd care to look.^[5] There's a genuine AI story there, even if it happens to coincide with an interest rate story that would drive similar results for a bank. And there's another story, too: a story of a bank that rebelled against the norm in banking and, as a consequence, grew to the point that their economics looked more bank-like: Klarna's shareholder letter talks about how they didn't want their economic model to be based on providing good service, not hitting customers with fees. Scroll down the Management's Discussion & Analysis section, though, and you'll see that of all their discrete revenue sources, one grew the fastest, at 39% annualized from 2022 through 2024. It's "Consumer Service Revenue," and the single biggest component of that is gently euphemized as "reminder fees." It's possible to build a better bank, and Klarna's done that. But the bigger it gets, the more it's just a bank.

Diff Jobs

Companies in the Diff network are actively looking for talent. See a sampling of current open roles below:

• An OpenAI backed startup that’s applying advanced reasoning techniques to reinvent investment analysis from first principles and build the IDE for financial research is looking for a product engineer with experience working on intuitive data-intensive interfaces at a product focused company like Notion, Figma, Stripe, etc. (NYC)
• Well-funded, fast-moving team is looking for a full-stack engineer to help build the best AI-powered video editor for marketers. Tackle advanced media pipelines, LLM applications, and more. TypeScript/React expertise required. (Austin, Remote)
• A well-funded startup ($16M raised) that’s building the universal electronic cash system by taking stablecoin adoption from edge cases to the mainstream is looking for a senior full-stack engineer. (NYC, Singapore)
• A Google Ventures-backed startup founded by SpaceX engineers that’s building data infrastructure and tooling for hardware companies is looking for full-stack and front-end focused software engineers with 3+ years experience, ideally with data intensive products. (LA, Hybrid)
• A company building the new pension of the 21st century and enabling universal basic capital is looking for a mobile-focused engineer who has experience building wonderful iOS experiences. (NYC)

Even if you don't see an exact match for your skills and interests right now, we're happy to talk early so we can let you know if a good opportunity comes up.

If you’re at a company that's looking for talent, we should talk! Diff Jobs works with companies across fintech, hard tech, consumer software, enterprise software, and other areas—any company where finding unusually effective people is a top priority.

Elsewhere

Crypto Adoption

Russian oil companies are apparently settling trades with China in crypto, in order to avoid sanctions. This is crypto perfectly performing its desired function: peer-to-peer exchange of value without the need for a trusted third party, and with a public ledger that means that people can retroactively deanonymize various parties. It is, like other aspects of crypto, a technologically sophisticated implementation of a very old concept. Historically, it was quite hard for Country A to impede some kind of financial transaction in Country B, if these transactions were ultimately settled in gold or equivalent bearer instruments. It just wasn't feasible to track and block the flow of money. Once currency transactions are mostly digital, it's comparably trivial for legal authorities to track them, and to cut off undesirable counterparties from the financial system. But the power to do that is always finite: it slightly weakens the dollar's network effect every time a transaction is blocked, and slightly strengthens the network effect of whatever gets around the blocking. That alone is not enough to really shift the balance between currencies—USD is far ahead of any alternative, and because of the volume of USD-denominated debt on both sides of balance sheets worldwide, it will remain so even if the US economy becomes a smaller share of global output. But if there's enough global demand for trade between countries that can't use dollars, there will be global demand for alternative currency systems that won't, or can't, block such transactions.

Art Margin Calls

Any time there's a business with exclusive access to big-ticket consumer purchases, there's an opportunity to turn it into a more lucrative business by financing that transaction, or selling the leads to someone who will. This extends to the market in fine art, where lenders have been issuing margin calls, demanding either more cash or to swap collateral for more valuable artwork ($, FT). It feels like a financially dubious decision to lend against an asset that has a high beta with respect to equities but produces no income and carries punitive transaction costs. But the actual collateral for these loans is basically a function of borrower ego multiplied by borrower net worth. The well-off have many costs they can temporarily cut, but one thing they can't bounce back from is selling an N-of-1 object.

Liquidations

Two crypto ETFs are liquidating, both of which were getting exposure to cryptocurrencies through futures rather than spot purchases. (They're both described as "active ETFs," but rolling futures contracts appears to be the full scope of their activity.) Both funds outperformed the spot funds for their first few months, and then started underperforming, drastically in the case of Ethereum. So it was a crypto ETF, with a systematic bet-structure overlay that briefly worked and then failed catastrophically. Crypto ETFs were a natural land rush, because the idea was obvious for a long time and was only held back by regulation. But ETFs in general are a scale business, and one ETF, BlackRocks's IBIT, has just under two thirds of the assets and 86% of the liquidity, so, over time, it'll charge the lowest fee for the identical service these funds all provide.

Corporate Espionage

HR software company Rippling has accused a competitor, Deel, of corporate espionage. They even set up a honeypot:

Earlier this month, [Rippling general counsel Venassa] Wu sent a letter to three people, including Philippe Bouaziz, Deel’s chairman and C.F.O. (and father of [Deel CEO and cofounder] Alex Bouaziz). The letter referenced a Slack channel that Wu implied had embarrassing information about Deel—but was really set up as part of the trap... Within hours, [the accused mole] started searching the channel, the company asserts.

There's a default form of internal transparency where young companies don't set up elaborate access rules and logging until they're either dealing with customers and investors who expect it or have some kind of incident. All of those rules create annoying friction and slow the pace of shipping, and the younger a company is the less likely it is to have any information worth stealing. But if they don't tighten up security until it's too late, the next best thing is to execute corporate counterespionage with a bit of panache: if the complaint is accurate, Rippling wasn't just able to identify the mole, but also to narrow down the list of who knew about them. A classic way for companies to deflect bad behavior is to say that it was done by junior employees without C-level knowledge. Rippling made sure that would be a difficult claim to make.

DeepSeek

DeepSeek's models are being deployed widely in China, and have sparked a broader tech bull market there ($, Economist). But it's hard to fight scaling laws. Most of the world's population and GDP are outside of China, and it's that segment of the global economy that still has access to the best chips and chipmaking equipment. In general, technologies with a high upfront and lower marginal cost do best when they can scale over as many users as possible, and even when a lower-scale participant makes some gains, the best bet is that they won't persist.