Library · Subscriptions · How to run this
Predictive LTV is not observed cash
Predictive LTV is a model. Actual LTV is settled cash on the cohort. Gap % = (predicted − actual) ÷ actual. Type predicted LTV $, actual LTV $, and cohort size.
Pair: Predictive vs actual LTV on Subscriptions. Bid ratio: cash LTV:CAC. Flow clock: flow vs rebill.
What you type
Predicted LTV $ — model LTV per buyer for the cohort (Klaviyo / CRM export).
Actual LTV $ — observed settled LTV per buyer on that same cohort window.
Cohort — buyer count in that cohort.
Gap per buyer = predicted − actual. Gap % = gap ÷ actual. Cohort gap $ = gap × cohort. Bid and forecast on observed cash when the gap is material — do not invent a predictive lift chip.
What not to do
Do not invent a gap percent. Do not mix billed remaining cycles into actual LTV. Do not compare predicted LTV from one window to actual from another. Do not treat predictive LTV as contribution margin.
Related: Subscriptions, cash LTV:CAC, flow vs rebill clock, send fatigue cost.
Procedure
- Type predicted and actual LTV $ for the same cohort.
- Type cohort buyer count.
- Forecast on observed cash when gap % is material.
Failure modes
- Inventing a gap percent.
- Mixing billed remaining cycles into actual LTV.
- Treating predictive LTV as contribution margin.
Worked example
None on this page yet. When one lands it will be a real case: what broke, the number, and what changed. Not a composite.
Glossary
- Predictive vs actual LTV gap %
- (Predicted LTV − actual LTV) ÷ actual LTV. Cohort gap $ = gap per buyer × cohort.
Work with us — We run this job on your accounts. Services
FAQ
- Is this cash LTV:CAC?
- Related ratio. [Cash LTV:CAC](/manuals/ltv-cac) divides surviving-cycle cash by CAC. This page compares a predictive LTV figure to observed cohort LTV on the same buyers.
- Is this Klaviyo flow vs rebill clock?
- Related CRM clock. [Flow vs rebill clock](/manuals/klaviyo-flow-vs-rebill-clock) is send timing vs MIT day. This page is predicted vs actual LTV dollars.
- What gap percent should I invent?
- None. Type predicted and actual LTV from the same cohort export.