Some early PayPal folks gave me advice that went against all my instincts as an engineer:

When you’re starting out building a payments product, don’t build any customized tools to prevent fraud.

Instead, just have a human review every transaction. Do this until you’re seeing thousands of payments per day.

They said something that stuck with me: “You think you’re a payments company, but you’re actually going to be a fraud prevention company. The problem is, right now, you have no idea what fraud to look for.”

Like most engineers, I’m always excited to build a system. But they were telling me to just watch it myself.

And they were right. At both Stripe and Increase, I’ve caught things that I never would have anticipated.

I’d see ten PayPal verification deposits arriving in one account and think, “That’s strange.” I wouldn’t have built a rule to catch that.

Over time, patterns emerge. Non-US submissions need extra scrutiny. Florida is overrepresented in identity theft, so check it closely. A trucking company signed up and it looked like fraud; now we know to watch for that in the future.

Patterns like this became the basis for our automated systems later on. We’re feeding the neural net we already have. There’s no need to design it from scratch.

My advice: when you’re getting started, build the big red buttons so you can stop things fast if something looks wrong. And get ready to scale when the real patterns arrive.

But accept that you won’t know the system you actually need until later, and that’s fine. At the start, just watch everything yourself.