Constrain the Risk Problem

The arithmetic that forces the question

Card lending is a thin-margin business where credit losses dominate the outcome. Eric Glyman states the problem directly: "Let's say it's a 1% margin business. If I give money to 100 people and I'm 99% right, I've made zero dollars. You need to be more than 99% right." Growth compounds the exposure every month as balances build against thirty-day repayment terms, and a new company has no loss history of its own, what Glyman calls "a terrible cold start problem": you must be extremely accurate before you have any basis for accuracy.¹

Pick a segment where losses are structurally small

Rather than try to out-model the problem, Ramp chose a segment where the problem is smaller by construction. Corporate cards in the Amex class run loss rates around 0.1 percent, against 5 percent or more for risky consumer populations. "It was a very different place, and so you could constrain the problem quite a bit."¹

The second constraint reuses a trick from Glyman's earlier company, Paribus, which connected to Gmail to see receipts and verify claims directly rather than inferring them. Ramp connects to a customer's bank account and sees how much money is actually there, converting an inference problem into an observation problem for a large share of applicants and letting the company "effectively box out a lot of the early risk questions."¹

What the constraint buys

The payoff is attention. With risk contained, the founders could spend their thinking on the question they actually cared about: "Could you have a card that had some software that would help you spend less, or software that would help you get a receipt and close your transaction faster?"¹ Choosing a market where the existential problem is structurally small is not timidity, it is how a founder frees up the scarce resource, attention and engineering time, for the differentiated question.

This corroborates a related point Glyman makes elsewhere about underwriting: even great machine-learning models only move approval rates from the low nineties to the mid nineties, so Ramp "deliberately chose corporate credit because losses are structurally tiny." The two statements are the same argument from two directions: underwriting was never going to be where Ramp wins, so the company picked a segment where it did not have to be.²