Stripe's Real Business Is the Fraud Graph

Most people think Stripe sells payment processing the way any processor does — a card number in, an authorization out, a fee taken in between. Compare the headline rate to a competitor's and Stripe rarely wins on price alone: 2.9% plus 30 cents is not the cheapest number in the market.

That comparison misses what a merchant is actually paying for.

The Invisible Business

Every time a card is charged anywhere on Stripe, the transaction doesn't just get processed — it gets scored. Stripe's fraud system, Radar, has seen enough of the internet's card traffic that there's roughly a 92% chance a given card has already shown up somewhere on the network before. That single transaction becomes one more data point in a shared risk model protecting every other business on Stripe, not just the one that processed it.

A new payment processor starting from zero can copy Stripe's pricing in an afternoon. It cannot copy years of fraud patterns learned from millions of other merchants' transactions. That gap — not the rate card — is the actual product being sold.

Why The Rate Isn't The Moat

Radar's model doesn't just look at a single merchant's history. It scores every payment using hundreds of signals pulled from across the entire network — the card, the buyer, the device, the merchant, and how all of those have behaved everywhere else Stripe operates. When it catches a new fraud pattern at one business, it doesn't need to relearn that pattern from scratch at the next one; embeddings that capture similarity between entities on the network let it recognize the same pattern elsewhere automatically.

The result shows up as a number that's hard to argue with: Radar reduces fraud by an average of 32%. No merchant, however large, generates enough of its own transaction volume to train a model that good in isolation. It only works because the network is shared.

That's the part a switching cost comparison usually misses. A merchant evaluating a cheaper processor is really comparing two things: 2.9% plus 30 cents against a possibly lower number, and a fraud model trained on trillions of dollars of shared transaction data against a model with no history at all. The fee is negotiable. The fraud graph isn't something a competitor can quote a lower price on, because it doesn't exist anywhere else.

Seen this way, Stripe was never really selling a payment rail. It was selling the accumulated fraud experience of everyone else already on it — and every new merchant that joins makes that experience a little harder for anyone else to replicate.