9.3

Attribution Models

Rules that assign credit for a conversion across the touchpoints that preceded it.

No single originator; data-driven attribution uses Shapley value (1953), Markov-chain from Anderl et al. (2016)

What it does

Distributes conversion credit across touchpoints — last click, linear, time decay, data-driven.

When it breaks

Observational, not causal. Benchmarked against experiments it systematically inflates cheap and search channels. It cannot substitute for an incrementality test, and it constantly is.

Case

Lewis, Rao and Reiley ran a placebo-ad experiment on Yahoo!: users shown an unrelated charity ad were compared with unexposed users, and the observational method attributed a large effect to an ad that could not have caused anything. The cause is activity bias — active people both see ads and convert — which last-click and most multi-touch models cannot remove.

Lewis, Rao & Reiley — correlated online behaviours ↗

Diagram — not yet drawn

One conversion path with credit distributed four ways under four models, and beside it the experimental answer, which none of the four reaches.

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…