Using predicted LTV in bids: disclosure checklist for the UA team

Predicted LTV models need known failure modes before scale. Disclosure checklist — not a vendor model card.
Bidding on predicted LTV without saying so turns a model error into a “channel” failure.
Disclosure checklist
- Observed vs predicted — State whether the bid target uses observed
af_revenue(events overview) or a forecast. - SKAN modelled rows — If the SKAN overview modelled metrics (null-CV fill) enter the target, say so explicitly.
- Training scope — Write geo, platform, and genre mix the model saw.
- Failure modes — List known breaks (ATT opt-in shock, new geo, new creative hook).
- Change control — No silent definition swap mid A/B.
Facts vs limits
| Fact | Limit |
|---|---|
| Observed revenue and SKAN modelled metrics are documented as distinct | Full vendor model cards are often private |
| Checklist is governance | Not a validation of any pLTV vendor |
CSV.


