What Will Your Customers Be Worth? Predicted, Then Back-Tested

How it works

Each customer's transaction log is reduced to the three quantities the buy-till-you-die models read: how many repeat purchases they made, how far into their history the last one fell, and how long they have been observed. BG/NBD is then fitted by maximum likelihood — it treats each customer as buying at their own latent rate while active and as able to churn permanently after any purchase, which is what lets it distinguish a customer who is quiet because they buy rarely from one who has gone. Gamma-Gamma is fitted the same way to the monetary side, blending each customer's own average transaction value with the population's. Expected future purchases multiplied by expected transaction value gives predicted lifetime value over a horizon taken from the data's own span. Every run is back-tested: the model is refit on the first 70% of the history and its predictions for the remaining 30% are compared against what actually happened, and that comparison is reported prominently — including when it goes badly.

Use it on a non-contractual business (retail, e-commerce, marketplaces, anywhere customers can lapse silently) when you need a forward-looking number: what to pay to acquire a customer, who to spend retention budget on, or how much future revenue the current base represents.

Not for subscription or contractual businesses where churn is observed directly rather than inferred from silence — use survival analysis there. Not for pre-aggregated per-customer data, and not when you need segments to target rather than a forecast to budget against: RFM segmentation is the right tool for that and complements this one. Not a substitute for margin analysis — the value here is gross.

What data do you need?

A raw transactions table, one row per purchase. For example, an order export:

Minimum 100 rows · Best with 2,000-200,000 transactions across 500+ customers spanning 12-36 months, with a healthy share of repeat buyers

What's in the report?

Standard-library analysis: forecast what your customers are worth from here, straight from a raw transactions table (one row per order). The industry-standard buy-till-you-die model — BG/NBD for how often each customer buys and how likely they are to have quietly churned, Gamma-Gamma for what a transaction of theirs is worth — gives every customer an expected number of future purchases, a probability they are still active, an expected transaction value, and a predicted lifetime value over a stated horizon. Every run is back-tested: part of your own history is held out, the model is refit without it, and predicted purchases are compared against what actually happened, so you can see whether the forecast is worth trusting before you act on it.