Cohort LTV modeling
Predicting 12-month customer value from the first 30 days of behavior.
4 min read•Updated April 7, 2026
Cohort LTV tells you what a new customer acquired today is actually worth over their lifetime. Before you've waited a year to find out.
The approach
We train a per-brand model on your historical cohorts. Inputs: first-order characteristics (AOV, SKU, channel, discount used), first-30-day behavior (repeat rate, support tickets, refund). Output: expected 12-month net revenue per customer.
The accuracy
On brands with 12+ months of history and 1000+ customers, median MAE is 8-14%. Meaning the model is usually within $10-$20 on a $100 LTV prediction.
How to use it
Two operational moves:
- Adjust your max CAC: set your payback target (3 months, 6 months, etc.) and derive max CAC from predicted LTV per channel.
- Flag at-risk cohorts early: if a cohort's 30-day behavior predicts below-threshold LTV, you know to cut spend on that source within the first month, not after 6.