Your LTV Is a Guess Until It Isn't
Every UA meeting has the same ghost in the room: a Lifetime Value number on a slide that nobody built. Someone multiplied day-30 ARPU by a hunch and called it LTV. That number is not LTV. It is a hope with a decimal point, and it is the reason smart teams overpay for users they should have never bought.
Q: Why is my early LTV number almost always wrong?
Because you are projecting from a curve you have not seen bend yet. At day 7 you have almost no signal about month 6, and mini-game retention is famously front-loaded — the people who churn, churn fast, and the ones who stay are a different animal. A day-30 LTV extrapolated linearly will overshoot by 2–4x on a casual title. The number only earns the name once you have a real cohort curve past the inflection point, usually day 60–90, and a model that fits the decay instead of assuming it stops.
Q: What does a model that actually works look like?
Three pieces. First, fit each acquisition cohort's revenue to a decaying curve (power-law or exponential decay) rather than averaging. Second, compute LTV per source and per geo, because a blended number hides the cohort that is quietly losing you money. Third, use a payback window, not a lifetime — "we recover CPA in 22 days" is far more decision-useful than "LTV is $4.10" when you are deciding whether to scale a campaign tonight. Predictive LTV (pLTV) models using early behavior can shorten this, but they still need a calibrated holdout.
Q: How do I use LTV without lying to myself?
Report it with error bars and a date. A number stamped "D30 projected, ±35%, n=12k" is honest and usable; a bare "$4.10" is a trap. Tie every LTV you act on to the cohort and the channel it came from, and re-baseline monthly as the game's economy shifts. The teams that scale safely are the ones that treat LTV as a living range, not a trophy.
Q: Where does LTV:CPI actually break down?
At the long tail and at the mix. LTV:CPI above 1 means nothing if your payback is 120 days and your cash runs out in 60. And a blended LTV:CPI of 1.4 can hide three channels under 1.0 that you are subsidizing with one winner. Break it down by source, watch the payback horizon against your working capital, and kill the channels where LTV:CPI stays under 1 past the point you can afford to wait.
| What a real LTV model has | What the slide number lacks |
|---|---|
| Cohort curve fit past day 60–90 | Linear extrapolation from day 7 |
| Per-source and per-geo LTV | One blended number |
| Payback window, not just lifetime | "LTV is $X" with no horizon |
| Error bars + cohort + date | A bare decimal |
LTV is not a number you put on a slide. It is a discipline — fit the curve, split it by source, stamp it with doubt, and let the payback window, not the lifetime, drive the spend. Do that and the ghost leaves the room.