The five models, compared
| Model | How credit is assigned | Best fit | Watch out for |
|---|---|---|---|
| First-touch | 100% to the first touch in the journey | Understanding what drives initial awareness | Ignores everything that closed the sale |
| Last-touch | 100% to the last touch before conversion | Short sales cycles, single-channel journeys | Overcredits bottom-funnel/branded channels |
| Linear | Equal credit across every touch | A simple, defensible multi-touch baseline | Treats a footer-link click the same as a demo request |
| Time-decay | More credit to touches closer to conversion | Longer sales cycles where recency matters | Still somewhat arbitrary about the decay rate |
| Data-driven | Statistical/ML weighting from your own conversion data | High-volume accounts wanting the most accurate split | Needs real conversion volume; a black box to explain to stakeholders |
Types of attribution models: the broader landscape
Beyond the five above, most platforms group models into two families: single-touch models (first-touch, last-touch) that assign all credit to one interaction, and multi-touch models (linear, time-decay, position-based, data-driven) that spread credit across the journey. See our dedicated multi-touch attribution guide for the multi-touch family specifically.
How to choose
hypothesis, from implementation experience rather than a controlled study: the model choice matters less than most teams assume, and identity resolution/event collection quality matters more. A sophisticated data-driven model fed by broken identity resolution will still produce numbers nobody trusts. Fix collection first (see the attribution pillar guide), then pick the simplest model that matches your sales cycle length.
Linear attribution model, in practice
The linear model is worth calling out separately because it is the most common starting point for teams moving off last-click: it requires no statistical modeling, it is easy to explain to stakeholders, and it corrects the worst distortion of last-click (ignoring the top of funnel entirely) without introducing a black box.