Journal · August 2025
When day-one LTV tells you to buy the wrong users
First-day revenue looks decisive in a UA dashboard. It is often an incentive, a shared family plan, or a refund that has not landed yet.
Bidding systems love day-one value because it arrives in time to train. Humans love it because it feels like proof. In hybrid IAP apps especially, day-one spikes are frequently coin bundles sold next to a large install incentive, or a family organiser paying for seats that will never open the app again.
We ask Ledger Bench seats to plot day-1 net revenue beside day-30 and day-90 contribution for the same cohort, after refunds. If the rank order of networks flips between those columns, day-one is not a proxy. It is a different product.
Refunds have a calendar
Store refund windows mean Tuesday’s “whale” can be Friday’s reversal. If your LTV tile books revenue on purchase and never revisits it, you will scale the creative that attracts refund-prone buyers. In Thailand, rail-specific reversal timing makes this worse: carrier billing and some e-wallets do not reverse on the same clock as the stores.
A practical brake
Do not let day-one value raise budgets by itself. Require a second gate: day-7 retained payer rate, or day-14 net after refunds. The gate can be cheap to compute. The point is that UA cannot claim “the model said so” when the model was fed a number that expires.
If you want the longer treatment, the predictive-tail module in Cohort LTV Mastery spends an afternoon on exactly this failure mode — including when a cautious tail is still worth fitting, and when you should simply wait.