User LTV analytics as a craft, not a KPI costume.

Four questions before any model

Who counts as a user — installer, account, payer, or household? When does the clock start — click, install, first open, or first successful payment? What money is net of store fees, VAT, refunds, and chargebacks? How long is “lifetime” when most of your history is still open?

If two people on your team answer those differently, you do not have an LTV problem. You have a vocabulary problem. The studio starts there.

Cohorts that respect different lives

Install-week cohorts are a default, not a law. First-purchase cohorts tell a different story for hybrid IAP apps. Plan-change cohorts matter after a price rise. Mixing them into one blended lifetime value is how Thai subscription products flatter last week’s media mix.

We also separate platforms. iOS and Android refund windows, payment rails, and organic quality are not the same in this market. Neither is TrueMoney versus card versus carrier billing.

Payback is a cash question; LTV is a contribution question

A cohort can pay back CAC and still be a poor lifetime bet if creative refresh, support load, or incentive cost eats the tail. Conversely, a slow payback on a high-margin subscription can be the right buy. User LTV analytics that collapse those into one ROAS tile are doing theatre.

Prediction with a visible fuse

We teach a cautious predictive layer: survival-style tails, simple parametric fits, and an explicit date after which you stop trusting the extrapolation. We do not teach you to ship a black-box scorer into production bidding. That work belongs to an internal data team with monitoring, not a six-week studio.