I've spent the bulk of my career in a peculiar cross-section between finance and tech—Fintech, for short. It was an area in vogue in the 2010s: I ended up at Square and Affirm, but I could have also joined everything from banking (Chime) to trading (Robinhood) to online payments (Stripe).

Most of these companies' stated missions were, in some approximation, to "democratize access" to financial services. The popular perception is that accredited investors and wealthy people make use of complex financial products that just aren't available to everybody else, so the notion of technology leveling the playing field is the righteous, noble motivation. This takes the form of eliminating fees on trading (Robinhood), broadening credit card acceptance (Square), or issuing loans quickly to more applicants (SoFi).

It's worth asking, though: how much does improved access increase common prosperity? When everyone can buy the same financial product, does that automatically translate to better outcomes for everyone as well? The scaling from technology tends to solve problems of distribution, but it may not eliminate the underlying dynamics around scarcity.

Now, tech's disruption playbook is simple. In economic terms, tech uses software, automation, and nowadays AI to massively increase supply, which drives down cost—and sometimes that cost is further subsidized by venture capital or free cash flows from other business lines. When the product or service is about as good but a whole lot cheaper, it attracts demand, in the form of a growing userbase. Webvan, from the dotcom era and bust, is an example of what happens when demand never materializes to match the scale. Ridesharing proved that demand can successfully scale with supply and expand entire markets. Currently, AI providers face so much demand that they're husbanding compute resources to keep their systems stable. These are examples of distribution actually creating more access, letting more people experience what was previously too expensive or exclusive.

The anti-pattern is when technology shows up with additional supply, but the nature of the demand is relative and positional: the value is in its scarcity, which creates zero-sum, comparative mechanics. Simply increasing supply shifts that value elsewhere. Twitter first introduced verified checkmarks to identify real personas, during a time when bots and celebrity impersonators ran amok. When the service went private, the previously opaque verification process became an $8/month subscription. While the change did mint many more verified users, the move also devalued the status, to the point where trustworthiness, previously associated with the blue checkmark, is now informally gleaned from a combination of likes, followers, and posting activity.

So let's talk Fintech.

Given my background at Affirm, I find Buy Now Pay Later (BNPL) lending to be a fascinating case study. The pitch is to broaden access to consumer credit by issuing easily approved loans for single purchases that can be paid back quickly. The most popular variant is the "pay in 4" payment plan, which essentially splits up a single payment amount into four smaller, weekly installments. There are plenty of criticisms of the business model: from accusations of predatory lending, to making it too easy to overspend and end up in debt, to getting hit with unexpected fees despite advertisements claiming "0% interest."

Yet, if we reframe credit as supply, then broadening access indiscriminately is an unsustainable business model: naïvely issuing credit to riskier borrowers doesn't make them more creditworthy, but rather just invites more people not to pay back their loans and go into default. The innovation here is to limit the amount of credit issued, doling out $10s and $100s of credit as opposed to credit cards' limits in the $1,000s, along with payback periods in weeks instead of months. This setup allows BNPL providers more leniency in whom they choose to lend to, which has led to their current popularity.

Credit rationing, though, is still happening; it's just typically enforced after the loan is issued. It initially seems like democratization, but much like credit cards, the punitive processes trigger on the back end: missed payments incur fees, accrue interest, and damage credit scores.

Affirm's main differentiator, the one that most critics of BNPLs miss when they lump all providers into a singular group of credit issuers, is where we rationed credit. We didn't charge late fees or tack on interest from missed payments, so all the decisioning had to be made up-front, during the initial phases of the signup/sign-in process before providing credit. To do this well, we generally avoided blanket approvals; instead, our teams used machine learning to underwrite loans, making intelligent guesses on our users' ability and propensity to pay back their debts.

Not that we haven't tried the industry-standard approach. Back in 2018, we were scaling up a new lending product in the mobile app, bootstrapping it with the underwriting model copied from our main product. The team came back, week after week, reporting punishing losses and painful default rates, as we had no good way to determine which new users were good borrowers. It took months of analyzing approvals data to customize this product's underwriting model to acceptable levels, and I appreciated the unglamorous grind for slow, incremental improvements in approvals.

Squint, and the dynamics that drove Twitter's verified statuses are what underlie BNPL's broader lending capabilities. The initial move was to make an intrinsic service—the verification, the loan itself—available to more people. But in doing so, the product pushed a subtle signal—Twitter's label for trustworthy users, borrowers' likelihood of paying off their debts in full—onto another layer. For Twitter, this was muddying the waters on who to pay attention to; for loan providers, gathering and understanding the signal is existential.

We often tell ourselves in tech, and Fintech in particular, that we level the playing field. But while we can claim the moral high ground, we don't recognize that access alone is sometimes insufficient, or that negative second-order effects like onerous late fees undo much of the benefits, or that the dynamics shift scarcity. We shouldn't be surprised, then, when the roots of the problem aren't solved by slathering on technology.