CAC: the full-cost fraction
The numerator is everything acquisition consumed; the denominator is a declared anchor; the clock is the attribution window in force. The three lines most often omitted, and what omitting them does:
| Omission | Effect |
|---|---|
| Affiliate cost | Paid channels look expensive next to an "organic" line that is actually CPA-funded — budget drifts toward the hidden invoice; see the precedence rule |
| Welcome-bonus cost (consumed) | Channels attracting bonus-hungry traffic look cheap at acquisition and expensive nowhere — the cost surfaces in cohort value instead, blamed on "retention" |
| Payment cost of early deposits | Small systematic understatement everywhere; material in markets with expensive local rails |
Compute it per channel and per market — the blended CAC hides exactly the comparisons the number exists to enable.
LTV: the censored fraction
Everything the cohort analysis page says applies with money attached: anchor declared, windows player-relative, filters versioned, censoring handled. The additions specific to value: the revenue base is contribution (net of bonus, payment and tax lines — the same bridge as NGR decomposition), whale treatment is decided and kept (a handful of accounts can own a cohort's average), and projected LTV is a labelled model output, never silently blended with observed value. A useful internal convention: write LTV with its horizon, always — a number without its clock is not a number.
Payback: the policy fraction
Payback divides observed cumulative contribution by full-cost CAC and asks when the fraction crosses one. Two design points carry the honesty:
- The window is a company policy, reviewed like one. It encodes cost of capital and risk appetite; it is set by finance and leadership, not inherited from a dashboard default. Different products and markets may legitimately carry different windows — declared, not accidental.
- Verdicts respect maturation asymmetry. Channels mature at different speeds; a uniform early verdict systematically rewards fast-shallow cohorts. The guard is a standing rule: no channel verdict before its cohort completes the declared horizon, with FTD2SD as the sanctioned early signal in between.
The review that uses all three
The monthly acquisition review reads one table: per channel and market — full-cost CAC, observed contribution at declared horizons, payback status against policy, and the incrementality multiplier from the experiment registry discounting the attributed volumes. Four columns, every basis labelled, no platform-reported ROAS in the room. Operators who run this table for two quarters report the same experience: the arguments stop being about whose number is right and start being about what to do — which was the point of the numbers all along.
Continue reading: The measurement stack — the cluster overview. LTV and media buy — the glossary definitions underneath.