Attribution
Data-driven attribution
Also called algorithmic attribution.
Data-driven attribution is a model that estimates credit from observed conversion paths. It compares paths with and without a touch to infer each touch's contribution.
How it is measured
The model needs many conversions and paths. It builds from observed sequences and estimates how each touch changes the likelihood of conversion. The output is a credit share per channel.
Check the volume. Platforms often require a minimum number of conversions in a window for the model to run. Below that, the tool falls back to a rule-based model without making it obvious.
Worked example
A wholesale tile supplier has 2,400 conversions over 90 days across paid search, email, display, and organic. A data-driven model shifts 18 percent of last-click credit from paid search to display and email.
The supplier tests it by cutting display spend for a region for four weeks. Orders there fall by 6 percent, which is about what the model implied. A smaller supplier with 90 conversions has too little data for any such result.
How it differs
Data-driven attribution learns weights from paths. Multi-touch attribution is the broader class, which also includes fixed rules like linear. Data-driven is one kind of multi-touch.
Common errors
Using it with too few conversions. Treating its output as proven cause. Not testing with a holdout. Forgetting it is a black box in many tools. Changing models and comparing across them.
In practice
Check your conversion volume first. If it is low, use a simple rule-based model and be honest about it. If high, validate the model's credit with a spend experiment.