Attribution Model

The specific rule assigning credit for a conversion across the touchpoints that preceded it.

The models and what each assumes

Same data, different answers
ModelCredit goes toAssumes
First touchThe first interactionDiscovery is what matters
Last touchThe final interactionConversion is what matters
LinearSplit evenlyAll touches are equal
Time decayWeighted toward recentInfluence fades
U-shapedFirst and last weightedDiscovery and close matter, middle less
Data-drivenFitted from observed pathsEnough volume and clean tracking

Source: Standard taxonomy

The rule outweighs the data

For a fixed dataset, changing the model can reorder your channel ranking completely. A comparison of channels under an unstated rule is not a finding about channels — it is a finding about the rule.

Choosing and holding one

  1. Pick from the decision. Budget allocation, channel exit and creative optimisation are different questions and may warrant different models.

  2. Document all four settings: lookback window, eligible touchpoints, credit rule, conversion definition.

  3. Hold it constant. Movement under a fixed rule means something; movement across rules means nothing.

  4. Where the decision is material, run a holdout test instead. The gap between attributed credit and measured effect calibrates how far to trust the model everywhere else.

RELATED TERMS

COMMON QUESTIONS

What are the common attribution models?
First touch, last touch, linear, time decay, U-shaped and position-based, plus data-driven models fitted from observed paths. Each distributes the same conversions differently.
Which attribution model is best?
The question is under-specified without the decision it supports. Every model encodes assumptions about how influence works, and none is discoverable from the data — which is why the rule must be documented and held constant.
Why do our attribution numbers change without campaign changes?
Usually the lookback window, which touchpoints are eligible, the credit rule, or the conversion definition moved. Undocumented model settings are the most common cause.
Does attribution measure causation?
No. It distributes credit for conversions that happened. Whether a channel caused conversions requires a holdout test, which is a different exercise entirely.

FURTHER READING

Learn how to apply this: RevOps 101: Revenue Operations Foundations

Definitions are the vocabulary. The courses are where you learn to operate it, with the interactive audit tools.

See the course