Stripe — Experimental Projects — Product Sense & Metrics

Role context: Data Scientist, Experimental Projects (zero-to-one product incubation) · Est. study time: 50 min · 5 practice questions

How to prepare for this role

This role is judged on whether you can turn thin, early evidence into a clear invest, iterate, or kill call, and on whether you picked a level of rigor that fits the decision.

Day to day you size new opportunities from Stripe's own payments data, define what success means for a prototype that has no baseline, design small tests with a handful of businesses, sit in on user calls, and write up what the team should build or test next. Some prototypes carry a model (a risk score, a classifier), so you'll also pick thresholds. Expect the interview to probe the same things: structuring an open "should we build this?" prompt, defining a metric for something brand new, designing and reading a test when n is tiny, SQL on payments-shaped data, and a past project where they'll ask how you decided how rigorous to be.

Where to spend your prep time

  • Matching the method to the stage: interviews, fake door, pilot, small A/B, holdback. Say why each fits.
  • Defining success and kill criteria before the data comes in.
  • Small-sample reasoning: clustering by business, wide intervals, deciding with expected value instead of a p-value ritual.
  • Selection bias in pilots and opt-in adoption, and how to get a causal read anyway.
  • Writing the recommendation down in plain words, with the uncertainty stated.

What Is Stripe Experimental Projects

Stripe is financial infrastructure for internet businesses. A business writes code against Stripe's APIs and gets payments, subscriptions, tax, payouts, fraud protection, money management and more. In 2025 businesses on Stripe moved about $1.9 trillion, up 34% on the year. Stripe earns mostly a cut of that volume (the take rate), plus software fees on products like Billing and Tax. So the company-wide score is total payment volume (TPV) and the net revenue it produces.

Experimental Projects is the in-house lab that tests what Stripe should build next. Engineers can get from an idea to a working prototype in days. The team puts it in front of real businesses, talks to them, reads the data, and decides whether the idea deserves to become a real product area. The current frontier gives a feel for the kind of bets involved: AI agents that buy things for people, stablecoin money movement, billing built for AI companies, agent-ready financial accounts.

Three structural facts shape every analysis here:

  1. The customer is a business that has to integrate code. Adoption needs engineering time on the customer's side. Getting from "interested" to "first live transaction" can take weeks, and it's where most pilots stall.
  2. There are very few units. A pilot has 10 to 40 businesses. Volume is heavy-tailed, so one big design partner can make a bet look huge or tiny. Buyer sessions run into the thousands, but they're clustered inside those few businesses.
  3. Money moves, so risk lags. Fraud and disputes show up 30 to 120 days after the payment. A new money product can look great on the dashboard and lose money two months later.

Put together, this is a 0-to-1 job inside a mature company. There's no baseline, samples are small, and the wrong answer costs a lot either way: a bad bet eats a team for six months, and a killed good bet never comes back. The data scientist's real output is decisions, made at the confidence the evidence allows.