The 30-Second Window: Why One Payment Processor Chose to Slow Down Checkout
Archi4 min read
Contents
The decision nobody wanted to make#
In early 2023, a mid-sized payment processor (call it the company) faced a choice. Its one-click checkout converted well, but chargebacks were climbing. The fraud team wanted to add a step. The product team said any extra step would cost more in lost sales than it saved in fraud. The CEO had to pick a side.
The company ran a test. For two weeks, half of new customers saw a 30-second identity check before their first purchase. The other half sailed through. The results were not close. The friction group lost about 4 percent of completed checkouts. But chargebacks in that group fell by 22 percent. When you price chargebacks at the full cost (the refund, the fee, the ops time, and the risk of losing a merchant account), the math flipped. The company kept the step.
The interesting part is not that fraud exists. It is that the cheapest place to stop it was not where the company had been looking.
Where the money actually goes#
Most payment fraud discussions focus on the stolen card. That is the wrong unit of analysis. The real cost lands on the merchant, and it comes in three waves.
First, the chargeback itself. The merchant refunds the sale, pays a fee (often $15 to $25), and may lose the goods. Second, the dispute process. Someone has to gather evidence, respond to the bank, and track the case. That is labor. Third, and least visible, the risk of losing the ability to accept cards at all. A merchant that crosses a chargeback threshold (often 1 percent of transactions) can be placed in a monitoring program or dropped. For a small business, that is an existential event.
The company in this story was not the merchant. It was the processor in the middle. But processors feel the same pressure. Too many fraudulent transactions and the card networks fine them. So they have a direct interest in stopping fraud before it settles.
The usual answer is machine learning scoring. Build a model, score every transaction, block the bad ones. That works, up to a point. The problem is that fraudsters adapt. A model trained on last month's patterns misses this month's. And every false positive (a good customer blocked) is a real cost.
The slow lane as a filter#
The 30-second check was not a CAPTCHA. It was a step that asked the customer to confirm a detail the bank already knew, like the last four digits of a phone number on file or a one-time code sent to the cardholder's device. It added friction, but only for first-time buyers. Repeat customers skipped it.
That design choice matters. The company did not slow down everyone. It slowed down the group most likely to contain fraud: new accounts with no history. And it accepted a small conversion loss in exchange for a large fraud reduction.
The product team hated it at first. They had spent a year optimizing the checkout flow, shaving seconds off every step. Adding 30 seconds felt like a betrayal of the roadmap. But the data won. The company kept the step and later added a second layer: a risk score that could skip the check for customers whose device and behavior looked normal.
What other companies get wrong#
The common mistake is to treat fraud prevention as a pure technology problem. Buy a better model, add more data, block more transactions. That approach has diminishing returns. Models are good at spotting patterns, but they cannot ask a customer to do something. A model can flag a transaction as risky. It cannot make the customer prove they are real.
A second mistake is to optimize for conversion at all costs. That works until the chargebacks arrive. Then the merchant is stuck: the checkout is fast, but the fraud rate is high, and the processor is threatening to drop them. By then, the fix is expensive.
The company in this story made the decision early, before the fraud rate became a crisis. That is the harder path. It means accepting a visible cost (slower checkout) for an invisible benefit (fewer chargebacks). Most teams do not have the stomach for it until they are forced.
The takeaway for anyone building a checkout#
If you run a payment flow, ask a simple question: where is the cheapest place to stop fraud? It is rarely at the end, in the dispute process. It is usually at the start, before the transaction settles. That might mean a check, a delay, or a limit on new accounts. None of those are popular. All of them are cheaper than a chargeback.
The company's 30-second window is not a universal answer. Different businesses have different fraud profiles. A digital goods seller with instant delivery faces different risks than a furniture store that ships in two weeks. But the principle holds. Friction, applied in the right place and only to the right people, is a tool. The companies that treat it as a failure of design will keep paying for it later.
The processor kept the step for a year, then replaced it with a smarter version that skipped the check for most customers. The fraud rate stayed low. The conversion rate recovered. The decision to slow down, even briefly, made the faster system possible.
More articles