Analytics Reference

Attribution Models Reference

Every attribution model is wrong, and choosing one is really choosing which mistake you can live with. I use this table to set expectations before a reporting change, because the model swap will move every channel's numbers and someone will assume performance changed.

Models compared
ModelCredit assignmentBiases towardGood for
Last clickAll credit to the final touchBrand search and retargetingSimple, short sales cycles
First clickAll credit to the first touchTop of funnel channelsUnderstanding discovery
LinearEqual credit to every touchChannels with high touch volumeLong considered purchases
Time decayMore credit to recent touchesClosing channelsLong cycles with a clear close
Position based40 percent first, 40 percent last, 20 percent middleDiscovery and closing togetherBalanced default
Data drivenModelled from observed conversion pathsWhatever the data supportsAccounts with enough volume
Beyond deterministic attribution
MethodWhat it answersRequirement
Incrementality testWould this conversion have happened anywayHoldout group and patience
Media mix modellingHow channels contribute at an aggregate levelYears of spend and outcome data
Self reported attributionHow did you hear about us, in the buyer's wordsOne form field
Modelled conversionsEstimates for unobservable usersPlatform provided, consent dependent

Notes

Adding a self reported attribution field to your lead form is the highest value measurement change available to most B2B teams, and it takes an afternoon. It captures podcast mentions, word of mouth and community referrals that no tracking system can see, and it consistently contradicts the platform reported picture in useful ways.

Changing your attribution model is not a performance change, but it will look like one in every dashboard. Announce it, restate a prior period under both models, and give stakeholders the comparison before they find the discrepancy themselves.

Data driven attribution needs enough conversion volume to model paths meaningfully. Below that threshold it behaves erratically and a position based model is more stable and easier to explain, which matters more than theoretical accuracy when the output is a budget decision.

Attribution capabilities, default models and consent requirements change regularly across Google, Meta and analytics platforms, including the retirement of several legacy models. Verify against the current platform documentation before rebuilding reporting.

Frequently asked questions

Which attribution model should I use?

Data driven if your account has the volume for it, position based if not. More important than the choice is applying one model consistently and never comparing numbers across models.

Why do my platform numbers add up to more than my actual sales?

Because each platform claims credit for conversions it touched, and touches overlap. Reconcile in one place, usually your CRM, and treat platform reported conversions as optimisation signals rather than accounting.

Is incrementality testing worth the effort?

For any channel taking a large share of budget, yes. A geographic or audience holdout answers the only question that matters, which is what would have happened without the spend.

Official sources

Platforms change specifications without notice. Check the primary documentation before a launch that depends on an exact value.

Read next

You might also like

Ranked by how closely each page overlaps with this one, using a similarity model over the whole library.

More Analytics Reference