Small sample, strong signal: how to bet early without fooling yourself
River Plate and Newell's Old Boys both passed on Lionel Messi. Not because of his talent. Because of a monthly medical bill.
Within minutes of watching the same 13-year-old play, Barcelona's sporting director told his coaches: "Sign him. Don't even think about it." There was no 12-month track record. Just a strong early signal and the guts to price it.
Barcelona paid for the treatment, made an exception so his father could live with him, and wrote the deal on a napkin "regardless of any dissenting opinions."
Every growth team has a River Plate in the room
The one who sees the monthly cost and wants the 12-month LTV before approving anything.
You do not need to wait that long. RevenueCat's data shows that for products with 2,000 or more paid subscribers, over 75% get 12-month LTV predictions within 10% of the real outcome, and the signal shows up in the first 7 to 30 days. More in predicted LTV: fund and kill channels off the early signal.
Bet early, but do not call it proven
There is a trap on the other side. A first cohort that lights you up, at a sample too small to be significant, is not a result. It is a lead. Calling it proven is how teams scale flukes. See when to stop an A/B test.
The answer is to hold both ideas at once: act on the early signal, and treat the action as a test.
The move
- Put a predicted LTV on the promising cohort, fit to your own data, not a benchmark.
- Size the bet to the uncertainty. A small sample earns a small, capped budget increase, not the whole account.
- Write down what would prove it wrong, and when you will check.
- Build the test that proves it: a larger sample, a holdout or a second cohort.
- Scale or kill on that result. Worst case, you were wrong and learned in weeks, not a year.
Bottom line
Predict it. Bet on it. Then prove it.
Eight Ballon d'Or awards later, nobody remembers the dissenting opinions.
FAQ
Can I make budget decisions on a small early cohort?
Yes, if you treat them as bets, not proof. Use a predicted LTV, cap the budget at risk, and set up a test that confirms or rejects the signal quickly.
How early can I trust an LTV prediction?
For subscription products with enough volume, much of the signal appears in the first 7 to 30 days. RevenueCat reports over 75% of products with 2,000+ paid subscribers get 12-month predictions within 10% of the actual outcome.
What is the risk of acting on early data?
Scaling a result that was just noise. Small samples swing a lot. Capping the budget and confirming with a larger test limits that risk.
What is the risk of waiting for 12-month LTV?
Losing a year. By the time the full number arrives, the budget decisions that mattered have already been made without it.