Direct answer: An attribution model is a rule for assigning credit, not proof of causality. Choose the model based on the decision: first-touch for discovery, last-touch for conversion capture, multi-touch for journey reporting and experiments or incrementality methods for causal questions.
Key takeaways
- Credit assignment and causality are different.
- One model should not answer every business question.
- Publish the lookback window, included channels and identity rules.
- Compare model outputs with pipeline and experiments.
Common attribution models
| Model | Useful lens | Main limitation |
|---|---|---|
| First touch | Discovery source | Ignores later influence |
| Last touch | Conversion capture | Undervalues discovery |
| Linear | Journey participation | Equal credit is arbitrary |
| Position based | Opening and closing | Weights are assumptions |
| Data driven | Modelled contribution | Depends on data and platform logic |
Choose by decision
Use first touch for acquisition discovery, last touch for handoff efficiency, cohort multi-touch for journey analysis, and controlled experiments when asking what caused incremental outcomes.
Document the rules
State window, identity, direct-traffic treatment, offline matching, channel inclusion and revenue source. Without these, model comparisons are meaningless.
Avoid false certainty
Dark social, consent gaps, cross-device use and offline influence remain. Use models as decision lenses and triangulate.
Experience, sources and limitations
This guide reflects GrowthSparx's practical work across acquisition, analytics, CRM and conversion operations. Platform interfaces and requirements change, so verify official documentation before implementation. The framework does not promise perfect attribution; it makes evidence, controls and uncertainty visible.