NBD-Dirichlet

A stochastic model that predicts, from category structure alone, every brand's expected penetration, purchase frequency and buyer overlap.

Goodhardt, Ehrenberg & Chatfield (1984)

What it does

Predicts what every brand's share, repeat rate and buyer overlap should be in a category, given only its size. Its real use is the residual: a brand that departs from the norm is doing something that actually needs explaining.

When it breaks

Assumptions fail with few players, high involvement, or one-off purchases. It also tells you nothing about why a deviation exists.

Case

Sharp and Sharp benchmarked Australia's Fly Buys — then the country's largest loyalty programme — against Dirichlet norms and found only weak, patchy excess loyalty rather than the step-change claimed. The method became the template for auditing loyalty schemes against a model-predicted baseline.

Byron Sharp — do loyalty programs increase loyalty? ↗

Diagram — not yet drawn

Two bars per brand — predicted vs actual repeat rate — with the gap shaded. Everything near zero except one brand.

In the wild

Unvetted · not part of the tier assessment

What has been written about this tool in the last twelve months. Machine-retrieved and unchecked — everything above this line was checked.

Loading…