Skip to main content

Preview — Pro guide

You are seeing a portion of this guide. Sign in and upgrade to unlock the full article, quizzes, and interview answers.

Attribution Modeling: Last-Touch, Multi-Touch, Shapley, MMM, and Incrementality

Marketing and growth interviews ask how you credit conversions to channels — last-click is biased, multi-touch rules are arbitrary, and MMM is slow. This guide covers Markov attribution, Shapley-Owen values, geo holdouts for ground-truth incrementality, and how Uber-style teams pair fast digital attribution with slower marketing mix models.

38 min read 2 sections 1 interview questions
AttributionMulti-Touch AttributionShapley ValueMarketing Mix ModelIncrementalityLast-Click BiasGeo ExperimentsGrowth AnalyticsCausal InferenceFacebook AdsGoogle AdsROAS

The Attribution Problem Is a Credit Assignment Problem

Users touch many ads, emails, and organic surfaces before converting. Attribution allocates 100% credit across touches — a zero-sum accounting exercise that is inherently not causal unless you bolt on experiments.

Interviewers want you to separate descriptive attribution (explaining historical paths) from incremental lift (what would not have happened without the channel). Confusing them ships bad budget.

This distinction matters because budget allocation is a control decision, not a reporting dashboard. A channel can look great in descriptive attribution simply because it appears late in journeys with already-high intent (for example, branded search). If you optimize spend on that score alone, you often cannibalize upper-funnel demand generation and then wonder why quarter-over-quarter growth softens. Strong interview answers call out this feedback loop and propose periodic incrementality experiments to anchor tactical MTA outputs.

IMPORTANT

Premium content locked

This guide is premium content. Upgrade to Pro to unlock the full guide, quizzes, and interview Q&A.