What Is Multi-Touch Attribution? Models Compared
Linear, U-shaped, time-decay: multi-touch attribution models explained, with an honest note on when the complexity pays off.
Multi-touch attribution splits conversion credit across every channel a customer touched — the podcast that introduced, the newsletter that nurtured, the search that closed. Intellectually it is obviously more correct than crowning one touch. Operationally it is a sophistication tax most companies pay too early.
The standard models
| Model | Credit split | Bias |
|---|---|---|
| Linear | Equal across all touches | Flatters middle noise |
| U-shaped | 40/40 first & last, 20 middle | The diplomatic compromise |
| Time-decay | More to recent touches | Closer-friendly |
| Data-driven | Modeled weights | Opaque; needs volume |
Note what the table implies: the 'right' split is a modeling choice, not a discovery. Different models reorder your channel ranking from the same data — which is why attribution debates are eternal.
The honest prerequisites
- Volume: splitting credit three ways across 30 monthly conversions yields decimals, not decisions. Below hundreds of conversions/month, model choice is astrology.
- Touch data quality: multi-touch amplifies tagging gaps — every untagged touchpoint (dark traffic) silently redistributes credit to the tagged ones.
- A genuine multi-channel mix: if 80% of customers touch two channels, first-touch plus a glance at journeys answers everything multi-touch would, for free.
The startup-honest recommendation
Use first-touch for budget allocation, read individual journey timelines for the qualitative middle (the timeline is multi-touch attribution with the weights left to your judgment), and graduate to formal models when you have a marketing team large enough to argue about credit — that argument is the actual use case. The SaaS attribution guide frames the same advice around revenue.