Predicted LTV (pLTV)

Predicted LTV, or pLTV, models how much a user is likely to be worth over their lifetime based on their earliest signals, so you can bid and optimize before the real number is ever known.

You do not have months to wait before deciding whether a user was worth acquiring. The auction wants your bid today, and the buying algorithms want to learn fast. Predicted LTV is how you give them an answer early: you take the first hours or days of a user's behavior, the actions they took right after installing, and you feed that into a model that estimates what they will eventually be worth. Then you optimize toward that prediction instead of waiting for the true lifetime value to slowly reveal itself.

This matters enormously on iOS, where SKAdNetwork gives you only a short, coarse measurement window and no clean view of long-term behavior per user. You often get one conversion value from a compressed early window and little else, so being able to turn those first signals into a value estimate is close to the only way to optimize toward revenue rather than toward raw installs. A user who spends a modest number of rupees in their first session might be forecast to become a high-value regular, and pLTV is what lets you bid up for people like them now.

But a prediction is a forecast, not a fact. It is only ever as good as the model and the data feeding it, and on iOS you are frequently predicting from the tiny, delayed SKAN signal, which makes the estimate shakier than it looks in a slide. Treat pLTV as a directional bet, not a settled truth, and keep checking it against your real cohort revenue as those cohorts mature. When the forecast and the actual money stop agreeing, trust the money and retrain.

Predict early so you can move early, but never fall in love with a number the future has not confirmed yet.