Finance.

The framework

1. 01

   A single churn number is a survival curve with the shape rubbed out.

   Every lifetime value figure rests on one churn assumption, and that assumption is almost always a single monthly percentage standing in for a curve. Real cohorts don't decay at a steady rate: they lose a chunk early, as the people who bought for the wrong reason work that out, then flatten into a durable tail. Divide by a blended monthly rate and you shorten the tail and steepen the front, which is wrong in both directions at once. Plot the survivors instead — of the hundred who joined in March, how many were still here at month three, at six, at twelve — and read the horizon off the shape. Then notice what moves the shape: raise price and churn rises with it, so a plan that raises price and holds churn flat is a plan arguing with itself. Carry the loaded acquisition cost you built on Day 127 straight across; today's work is the other side of the calculation, the side that decays.

2. 02

   Cap the horizon at a future you can defend.

   A lifetime value with no end date is a perpetuity, and a perpetuity is a claim that this business will still exist, unchanged, in fifteen years. Cap it. Pick a horizon you can defend from your own data — twenty-four months, thirty-six at the outside, and never longer than you've actually been trading — then discount the cash inside it at your real cost of capital, because a dollar arriving in year three isn't a dollar today. The shorter horizon usually halves the figure. That isn't pessimism. It is the removal of years you invented.

3. 03

   The average is where decay goes to hide.

   A blended lifetime value averages the customers you won when the market was young and cheap with the ones you are buying now, and the blend keeps looking healthy long after the recent cohorts have stopped being. Split the base by the month people joined and plot retention and revenue for each one. What you want is the newest cohorts tracking at or above the older ones. What you usually find is a slow fade that has never been named, because the average kept clearing the threshold and gave everyone a reason not to look underneath it.

4. 04

   Move the number to finance, and re-test it every time spend steps up.

   Give the calculation to whoever owns the accounts rather than whoever owns the budget. When the team requesting the spend also sets the assumptions that justify it, the assumptions come back generous, and no dishonesty is required; the optimism is structural. Then re-run the whole thing at every step change in spending, because the cheap channels saturate first. Quadruple the budget and acquisition cost rises while the customers it buys churn faster than the ones you already had. A ratio that held at fifty thousand a month isn't evidence about two hundred thousand a month.