Concepts

How to read risk bands

What each band means, why the numbers are bands and not percentages, and how to act on them.

The four bands

Why bands, not percentages

You will sometimes see a probability in the customer detail page, labelled "p = …". Treat it as an internal ranking signal, not a real-world probability.

Until the model is calibrated against real repeat-purchase history, the raw probability is not a reliable forecast. That is why the merchant-facing product shows bands, and why the pre-release notice appears on every screen.

Bands are stable and comparable across customers. "Very high" always outranks "high". That is what you need to prioritise outreach — not a decimal you have to interpret.

How to act on each band

What makes a score go up

The score is driven by recency, frequency, monetary value, and how those have trended. A customer whose gap between orders is stretching, whose order count is falling, or who has stopped buying entirely will move up the bands.

A single large order does not make someone loyal, and a single missed month does not make them lost. The model looks at the whole pattern.

← Getting started with ChurnWhisper Managing digest recipients →