An attribution model is the rule that decides how credit for a conversion is split across the marketing touchpoints that preceded it, from giving everything to the last click to spreading it with a statistical, data-driven model. The model you pick changes which channel looks best, but none of them prove causation.
What is an attribution model?
A buyer might see a paid social ad, later search for you organically, then convert from an email. An attribution model answers “how much of that $900 order does each touch get?” The answer feeds channel ROAS, CAC by channel and, ultimately, budget.
Channel credit=Σ over conversions (Conversion value × Model weight for that channel’s touchpoint)
- Model weight
- Share of credit the model assigns to each touchpoint; weights sum to 1 per conversion
What are the main types of attribution models?
| Model | Rule | Paid social | Organic search | |
|---|---|---|---|---|
| Last click | 100% to the final touch | $0 | $0 | $900 |
| First click | 100% to the first touch | $900 | $0 | $0 |
| Linear | Equal split | $300 | $300 | $300 |
| Position-based | 40% first, 40% last, 20% middle | $360 | $180 | $360 |
| Data-driven | Statistical model of paths that convert vs. those that do not | Varies | Varies | Varies |
In April 2023 Google announced that first click, linear, time decay and position-based models were going away across Google Ads and Google Analytics 4, while data-driven, last click and external attribution would not be affected; conversion actions still on a retired model were switched to data-driven from September 2023.1Source 1 · Google Ads Help, 2023First click, linear, time decay, and position-based attribution models are going awaysupport.google.com Rules-based models are still easy to compute yourself from path data when you need them for comparison.
Does attribution measure what marketing caused?
No. Every attribution model, data-driven included, distributes credit among observed touchpoints; it cannot see the buyers who would have converted with no ads at all. In a comparison against 15 randomized experiments at Facebook, Gordon and colleagues found observational methods often failed to reproduce the experimental effects.2Source 2 · Gordon, Zettelmeyer, Bhargava & Chapsky, Marketing Science, 2019A Comparison of Approaches to Advertising Measurement: Evidence from Big Field Experiments at Facebookkellogg.northwestern.edu Use incrementality tests and media mix modeling to check what attribution tells you.
Common mistakes
- Comparing channels under different models (platform view-through in one, last click in another).
- Switching models without restating history, which creates fake trend breaks.
- Broken tracking masquerading as model choice: inconsistent UTM parameters distort every model equally.
How to compare attribution models in Kimo
Kimo builds a touchpoint table from GA4, ad platforms and your CRM (HubSpot, Salesforce), then computes several models side by side in Explore. The Command center shows the spread between models per channel: a wide spread is a sign to run a lift test before moving budget.
Frequently asked questions
What is the best attribution model?
What is data-driven attribution?
Why do ad platforms report more conversions than my analytics?
Sources
2 references- First click, linear, time decay, and position-based attribution models are going away (opens in a new tab)Google Ads Help2023support.google.com
Rules-based models removed in Google Ads and GA4; data-driven, last click and external unaffected.
- A Comparison of Approaches to Advertising Measurement: Evidence from Big Field Experiments at Facebook (opens in a new tab)Gordon, Zettelmeyer, Bhargava & Chapsky, Marketing Science2019kellogg.northwestern.edu
Observational methods often fail to match randomized experiments.
External sources were accessed at the time of writing. Kimo product details, customers and figures in examples are illustrative unless a source is cited.
