ltv-prediction
Installation
SKILL.md
LTV & LT Prediction for the Internet Industry
Execution Modes (Choose First)
This skill has two execution modes. Pick the mode that matches how the user's data is available; both share the same fitting and reporting methodology.
| Mode | When to use | Data source | Go to |
|---|---|---|---|
| Mode A — Direct data | User pastes retention/LTV series, provides a raw event table, or wants payback / mature-cohort borrowing / ML user-level prediction | Data supplied by the user, or pulled ad hoc via ae-cli/SQL | Sections I–VII below (the primary methodology) |
| Mode B — ae-cli automated | User has a TE project reachable through ae-cli and wants automated data retrieval, or needs RFM / pay-tier / VIP stratified LTV |
ae-cli queries against the live project |
Section VIII (ae-cli edition) + references/ |
Mode selection rules:
- If the user has a TE project + ae-cli access and asks for prediction → Mode B for data retrieval, then reuse Mode A's fitting/extrapolation (Sections I–V) on the retrieved series.
- If the user directly provides data (pasted series, event table, no TE project) → Mode A.
- Stratified LTV (RFM / pay-tier / VIP) is a Mode B capability (B3). Payback analysis, mature-cohort decay borrowing, multi-function comparison (Power/Log/Exp), and ML user-level prediction are Mode A capabilities.
- The two modes are complementary, not exclusive: it is normal to use Mode B to fetch the cohort LTV series, then Mode A to fit multiple functions, compute payback, and borrow a mature cohort's decay rate.