Traditional LTV calculation takes average revenue per account, multiplies by average retention length, and produces a number that describes the average customer — which is to say, almost no actual customer. The value of a predictive LTV model is that it produces an estimate for each specific account, updated as new behavioral data comes in, that reflects how that particular customer is likely to behave rather than how the median customer has behaved historically.
The Inputs That Matter
LTV models trained purely on billing data have limited predictive power. The strongest features tend to be behavioral: time-to-first-value (how quickly the customer reached their first meaningful outcome), feature adoption breadth, support ticket frequency and resolution pattern, expansion event history, and product engagement velocity over time. These behavioral features predict future revenue better than plan tier or company size because they reflect how deeply the product is embedded in the customer's workflow.
How LTV Changes Acquisition Decisions
When you have account-level LTV predictions, you can segment your acquisition channels by predicted LTV rather than just by acquisition cost. A channel that delivers customers at a higher CAC but with 2× the predicted LTV is a better channel, not a worse one. Most acquisition teams don't have this data because they haven't connected LTV predictions to their attribution models — doing so changes which bets are worth making.