In the spring of 2020, something odd happened to Americans’ credit scores.
They went up.
During the early stages of the most severe economic shock to hit the country since the subprime mortgage crisis, credit scores went up. According to FICO’s own data, between April 2020 and April 2021, the average FICO score for a U.S. consumer jumped from 708 to 716. And it didn’t stop there! By April of 2023, the average U.S. FICO score had reached an all-time high of 718.
Why did this happen?
Was there some sudden, large-scale shift in American consumers’ willingness to repay their loans?
No. It wasn’t consumers’ debt repayment habits that changed during the COVID-19 pandemic. It was their circumstances. Stimulus checks, plus reduced spending and forbearance programs from lenders and (in the case of student loans) the U.S. government created a significant amount of room for consumers to build up their savings and pay down their existing debts.
In other words, it was their ability to repay that changed.
We know this is the case for two reasons.
First, these pandemic-era score increases were concentrated in the deep subprime bands. According to FICO, consumers scoring between 550 and 599 saw average gains of around 20 points in the first pandemic year. In other words, the score moved most for the people whose improvement was most likely to be temporary scaffolding — stimulus checks and paused payments — rather than a durable change in how they were doing.
Second, once that temporary scaffolding was removed, the performance of lenders’ portfolios began to degrade, in parallel with consumers’ ability to repay. Even as that record 718 average score was being set in 2023, missed payments were already climbing; by FICO’s later accounting, bankcard delinquencies rose 48% and mortgage delinquencies 58% off their 2021 lows.
Most troublingly, that degradation didn’t show up in the national average FICO score until the end of 2023, when it dropped by one point to 717 (its first decline in more than a decade!) It held at that level through 2024, then slipped to 715 in 2025 and 714 in early 2026. At 714, it’s still six points higher than it was in April of 2020, despite Americans’ total debt levels being significantly higher today than they were back then.
So, to sum up, the national average U.S. FICO score jumped precipitously, in the midst of an unprecedented economic crisis, and has drifted down only grudgingly since — four points off its peak in three years — even as Americans’ debt loads climbed and their ability to repay eroded.
That doesn’t seem great, but I want to be clear about why this happened.
It’s not because the FICO score is badly designed. To the contrary, you won’t find a more well-tested and analytically-sophisticated general purpose credit score anywhere in the world.
But the efficacy of that score is constrained by the data it is built on and the competitive ecosystem in which it is deployed. And those structural constraints have, over time, become more problematic.
Structural Constraints
I’ll give you three examples.
#1: Proxies, not fundamentals.
The assessment of creditworthiness rests on two pillars: willingness to repay and ability to repay. You have to be able to assess both in order to make good credit risk decisions.
However, the problem — which was thrown into stark relief during the pandemic — is that the analytical value of traditional credit bureau data is weighted heavily towards willingness to repay.
The reason for this is easy to understand. A credit file is a historical record of a consumer’s debt repayments. That history, whether positive or negative, is a reflection of both their behavior (Were they careful about how much debt they take on? Were they diligent in making payments? Etc.) and their capacity (Did they have enough money to pay back all of their debts?) Behavior tends to be a durable signal. You’re generally either a responsible person or you’re not. Capacity, as we saw during the pandemic, is circumstantial. It’s a function of where you are, at a certain point in your life. So, a historical record of debt repayment may tell you something meaningful about a consumer’s behavior (their willingness to repay), but it likely won’t tell you much about their current capacity (their ability to repay).
So, if traditional credit files and credit scores don’t know, for sure, what a consumer’s current ability to repay is, they need to guess.
My go-to example is inquiries; the record of all the times that a consumer has applied for credit within the last two years. You likely know that having an excessive amount of inquiries on your credit report is considered a negative signal. But you may not know why. The logic is revealing. It goes something like this: Maybe you lost your job, and you know that pretty soon you’re not going to have enough money to pay all of your bills and so, before that happens and your credit score takes a hit, you want to apply for as much credit as you possibly can, even though you know you likely won’t be able to repay all of it.
That is a rather wild chain of interlinked assumptions, which has become formalized as a best practice in consumer credit (Don’t apply for too much credit or you’ll hurt your score) even though it has no fundamental relationship to the actual concern that the lender is trying to validate (Did you lose your job?) The traditional credit file doesn’t include income or employment data (i.e., ability to repay data) so the bureaus and FICO have had to construct a proxy — inquiries — in order to approximate it.
The trouble with proxies — apart from them often being very far off the fundamental reality you are trying to measure — is that they can be satisfied without the underlying reality changing at all. I’ve ranted about fintech credit builder products in this newsletter so frequently, that my doctor and my wife have both advised me to avoid the topic on high blood pressure days. So, I won’t go into a whole thing here. Instead, I’ll just point out that when a score is built on proxies, gaming that score (and the data it’s built on) isn’t a bug that better enforcement can stamp out; it’s the logical response to being measured by guesses rather than facts. Instead of describing the world, the scoring model becomes the rulebook everyone follows. And once that happens, it detaches a little more from the reality it was supposed to represent.
#2: Data is now an asset.
The credit bureau system runs on furnishing. Lenders voluntarily and freely report loan performance so that everyone can draw on the shared pool of results data. That model wasn’t obviously objectionable when the big three credit bureaus were consolidating and digitizing in the 1970s and 1980s. It makes far less sense now, in 2026, when data is rightly understood as one of the most valuable assets that a company owns.
The clearest evidence of this shift is pay-in-4 buy now pay later (BNPL); a huge, fast-growing category of consumer credit that millions of people reach for precisely because they don’t want credit cards, and that is still, years in, missing from traditional credit files. Part of the reason for that is technical (furnishing pay-in-4 repayment data and incorporating it fairly into scoring models is tricky), but part of it is simple economics. The firms that hold that data have little to gain and a lot to lose by pouring it into a shared utility that may then penalize their own customers for choosing what was, for them, the more responsible product.
The traditional credit bureau furnishing model assumes universal participation. That assumption is out of date.
#3: A loss of accountability.
We tend to imagine the traditional credit bureaus as having always been big, centralized, and faceless.
They weren’t.
They started as local cooperatives. In the 1890s, a grocer named Cator Woolford compiled a list of customers and their creditworthiness for his retail grocers’ association and sold copies to fellow merchants to cover the cost. That list became the Retail Credit Company, founded in Atlanta in 1899 — the firm you now know as Equifax.
In the cooperative era, accountability was structural, not sentimental. The merchants who contributed the data were the same merchants who relied on it, which gave them an incentive to focus on only reporting reliable, high-quality data. More importantly, the consumers who were the subject of those reports were also their customers. Sloppy furnishing or unsatisfying dispute resolution could create real business risk for the merchants.
That structure broke on a specific date: 1901, when Retail Credit Company began selling “moral hazard” reports on prospective policyholders to the life insurance industry. For the first time, the buyer of the data wasn’t also a contributor to it. What followed was both predictable and unsettling: Investigators interviewing your neighbors and coworkers, logging your drinking habits, your relationships, your “mode of living,” sometimes inventing the derogatory details outright. The consumer, who in the cooperative model had been both customer and product, became only the product.
It was, in fact, so unsettling that Congress eventually noticed. Committee hearings uncovered systemic errors and abuses and helped drive the creation and passage of the Fair Credit Reporting Act in 1970, one of the first data privacy laws in the world. The FCRA essentially bolted an artificial layer of accountability onto the credit bureaus — accuracy obligations, dispute rights, limits on who could pull a file and why. It did not change the underlying incentive, because it couldn’t. The bureaus still made money selling files to third parties, and consumers still were the product, not the customer.
The FCRA imposed accountability from the outside precisely because the market structure no longer produced it from the inside.
Cash Flow Underwriting
If you’ve worked in fintech for a while and/or are a frequent reader of this newsletter, you’ve probably already landed on a good fix for the problems enumerated above: Cash flow underwriting.
Instead of inferring a consumer’s ability to repay from proxies, measure it directly. With the consumer’s permission, a lender can reach into their bank account — the paychecks landing on the 1st and the 15th, the balance that holds a cushion (or the one that scrapes zero every month), the rent clearing on time — and underwrite against the real thing.
Cash flow underwriting effectively addresses our first two structural constraints.
It goes straight at the proxy problem (#1). You’re no longer counting inquiries and trying to reverse-engineer a story about job loss; you’re looking at whether the income actually shows up and whether there’s money in the account. That isn’t a proxy for ability to repay. It is ability to repay. And because you’re watching real behavior instead of a manipulable summary of it, there’s far less opportunity to game the system. You can keep a dormant credit card open to age your file or sign-up for a credit builder loan that doesn’t actually disburse the proceeds upfront. You can’t fake two years of a paycheck hitting your account.
And cash flow underwriting sidesteps the furnishing problem (#2) entirely. It’s not dependent on lenders volunteering to furnish data for free. The data already exists, sitting in the consumer’s own bank account, and the consumer permissions access to it directly.
The third problem — accountability — is the tough one.
Think about what actually happens when a consumer permissions their bank account data to a lender through an open banking connection. The lender gets the data. The consumer gets, hopefully, a better decision. And the bank — the institution that holds the account, that has the actual relationship with the consumer, that generated the data in the first place — has no idea it happened and no stake in what comes next. The consumer has, in effect, carried their financial data out past the walls of the institution that holds it and handed it to a stranger.
Whatever accountability lived inside that banking relationship doesn’t travel with the data. The bank isn’t answerable for how the data gets interpreted, or whether the resulting decision is fair, because as far as the bank is concerned, nothing happened.
Cash flow data is more accurate and more comprehensive, and those two traits are extremely valuable! However, it doesn’t solve the accountability problem. In fact, it doesn’t advance it at all past the point that the Retail Credit Company left it in 1901, when it decided to make consumers the product and lenders (and life insurance companies) the customer.
A First-Party Credit Score
Here’s a different idea.
And, as far as I can tell, an unprecedented one.
What if the company that already has the relationship with the consumer — that already holds the account and generates the data — built a credit score itself, out of its own first-hand experiences working with that consumer, and then let the consumer see it, improve it, and carry it to other lenders?
Call it a first-party credit score.
It sounds like fintech science fiction, but it’s not. It’s a real idea that is, at this moment, being tested in the market by one of the largest consumer fintech companies in the U.S.: Block.
They call it the Cash App Score.
I recently had the opportunity to chat with Juan Hernandez, Head of Credit and Underwriting at Block about the Cash App Score. As he explained to me in our conversation, the initial version of the model was designed to underwrite Cash App’s small-dollar lending product (Borrow). The initial thinking was that the model would combine Block’s proprietary first-party data with traditional credit data. However, when Juan and his team tested that version of the model against a version that utilized just Block first-party data — no traditional credit data — they got back a result that isn’t supposed to happen: Higher approval rates, higher conversion, and lower losses, all at once. The first-party view of the customer, the actual texture of their financial life on the platform, turned out to carry essentially all of the predictive signal.
The Cash App Score is that internal underwriting model, generalized beyond Block’s own lending products, surfaced, in-app, to Cash App customers, and, soon, made available to other lenders.
The data that drives the score is a reflection of customers’ holistic engagement with the Block ecosystem: Cash App and Afterpay loan repayments, Cash App deposit account inflows and outflows (including bill pay and Cash App Pay transactions), Cash App debit card spending, Cash App savings and investing activity, and Cash App peer-to-peer payment patterns. Across tens of millions of monthly active users, those signals resolve into something like a financial fingerprint; a high-frequency read on how a person actually earns and manages money.
And the reported results, on Block’s own book, are impressive: Roughly 70% of active Borrow customers have a traditional FICO score below 580, and they repay within the economics the product is built on. Block says the technology behind the score approves 38% more customers than conventional scores at the same loss rate.
The Cash App Score is rolling out to customers in the app with a small set of legible, do-this-not-that suggestions, and it refreshes as often as weekly rather than the bureau’s monthly cycle. When Block piloted the customer-facing version in its mobile app, around 55% of users came back to check it within 30 days, and, of those who took an action in response to the suggestions, 70% returned within 30 days.
Now let’s revisit our three structural constraints.
#1: Proxies, not fundamentals.
This is the same win cash flow underwriting delivers, and for the same reason: Real income and real spending in place of inquiries and utilization tricks.
The Cash App Score isn’t guessing at ability to repay; it’s reading it. With no proxies underneath, there’s almost nothing to game. The whole cottage industry of credit score optimization has nothing to grab onto when the thing being measured is whether your paycheck shows up.
#2: Data is now an asset.
Block isn’t furnishing its data into a shared pool for the credit bureaus to monetize and for its competitors to use. Nor is it simply opening up consumer-permissioned access to a subset of its own data. It’s building a product on top of all of its first-party data and selling access to it.
This inverts the incentives completely.
A furnisher wants to contribute the minimum required data with the minimal amount of effort and accountability. Just enough to not violate the FCRA. Same thing for data providers under open banking. They will follow their customers’ instructions for routing covered data to third parties, but they won’t usually go further than that (and, in the case of certain large banks, they will go much less far, unless they get paid).
A company selling a score wants the underlying data to be as complete, as clean, and as current as possible, and it wants to build real analytical expertise in turning it into an accurate prediction of risk in a way that doesn’t just help its partner lenders make better risk decisions but also help its customers get more access to more financial products.
The same economics that made furnishing decay and open banking contentious are the ones that make a first-party score compound.
#3: A loss of accountability.
This is the one where a first-party score has the opportunity to really differentiate itself.
The traditional credit bureaus have no relationship with consumers because they are, in no meaningful sense, their customer. They’re the product.
Cash App does have a relationship with consumers, and that relationship boxes it in from both directions, manufacturing the very accountability the FCRA had to try to artificially manufacture.
The company can’t treat consumers purely as the product. Picture Cash App quietly spinning up a side business slicing its customers’ financial lives into non-FCRA marketing segmentation products, the way the credit bureaus long have. Its customers would (rightly) revolt, because they are right there, in an active relationship with the company, watching what it does. The relationship makes moves unthinkable that, for an entity with no relationship, are simply good business.
But it also can’t treat its customers’ behavior as irrelevant to anything beyond its own business objectives. Because Block is selling the score to other lenders, the value of that score depends on it being genuinely predictive for them, which means Block has to care about data quality issues that it might otherwise be tempted to tolerate. It can’t, for example, let first-party fraud run in its own portfolio even where it could absorb the loss, because sloppiness there degrades the asset it sells to everyone else. The lenders buying the score — and their loss rates — become the enforcement mechanism.
Block’s recently-announced partnership with Nova Credit proves the point.
On an integration level, this partnership makes a ton of sense as it will allow Block to quickly get the Cash App Score plugged into lenders’ existing workflows through the Nova Credit platform. However, purely in terms of accountability, when a customer who’s permissioned their score to a lender gets declined, they’re not going to turn to Nova Credit for an explanation or advice on what to do next. They’re going to open up Cash App, because that’s where the score lives and where they can see how it’s changed over time and what they need to do to make it go up.
The artificial accountability the FCRA was built to force is largely redundant when the incentives are already aligned. The relationship does the job the regulation was invented to compel.
Realigning Incentives
For a century, the deal was that your financial reputation would be assembled about you, traded around you, and shown to you only after something had already gone wrong.
A score you can see, improve, own, and carry — held by a company with an actual reason to get it right — is a fundamentally different arrangement.
The eye-catching numbers, the higher approval rates and lower losses, are real, but the most compelling part of this story is that after more than a hundred years, the company holding and evaluating your financial data might also be sitting on your side of the table.
Not out of altruism or regulatory obligation, but because, for once, the incentives are aligned.
About Sponsored Deep Dives
Sponsored Deep Dives are essays sponsored by a very-carefully-curated list of companies (selected by me), in which I write about topics of mutual interest to me, the sponsoring company, and (most importantly) you, the audience. If you have any questions or feedback on these sponsored deep dives, please DM me on Twitter or LinkedIn.
Today’s Sponsored Deep Dive was brought to you by Block.
Block, Inc. builds technology to increase access to the global economy. Each of our brands unlocks different aspects of the economy for more people.
