Measurement
Part of How to build affiliate marketing measurement that stands up
Seven affiliate marketing attribution methods compared for UK teams
Compare seven affiliate marketing attribution methods, from last click to incrementality testing, and pick the model your programme can evidence.
What to take away
- The most common mistake is choosing a model out of habit: last click credits the final touch and starves discovery.
- Seven methods are worth comparing, from last click and first click through to incrementality testing.
- First click, linear, time decay and position based sit in between, and each suits a different question.
- Match the method to your commercial question, your tracking capability and your reporting cycle.
- Consent rules decide what data you can collect, so document the model, window and exclusions, then review each quarter.
Seven affiliate marketing attribution methods
Last click
Last click gives all credit to the final tracked interaction before purchase. The weakness is that it rewards closing partners and ignores assist roles: a cashback site may take credit for a sale a content publisher influenced earlier. Use it when your programme is small, or when contracts pay on the final click.
First click
First click credits the first tracked touch. It suits brands testing discovery channels, because it shows which partners introduce new audiences. It can overpay upper-funnel partners that never influence the decision, so pair it with a closing metric.
Linear and time decay
Linear splits credit evenly across tracked touches. Time decay weights touches closer to the sale. Both need a defined lookback window, otherwise counts drift. Time decay suits shorter consideration cycles.
Position based
Position based assigns set shares to the first and last touches, with the remainder spread across the middle. It acknowledges discovery and closing roles. The split is a judgement call, so document the percentages and check them against margin.
Data driven
Data driven models use your own conversion data to weight touches. They need volume and clean tagging, and the model can be opaque to partners. If you cannot explain it to a publisher, expect disputes.
Incrementality testing
Incrementality testing compares outcomes with and without a partner, often through holdout groups or geo tests. It answers whether the partner caused the sale rather than merely appearing near it. That is stronger evidence for budget decisions, but it needs enough traffic to reach a view.
What the rules require before you model
Consent sits underneath every method. Cookies and similar technologies used for attribution fall within the UK GDPR, so you need a lawful basis and clear information for users. The UK General Data Protection Regulation sets the framework your tracking must follow.
The ICO's register of fee payers lets you confirm a partner's data protection registration before you share audience data.
On the money side, how you treat network fees and partner costs affects reported return. Costs passed to customers have specific VAT treatment, explained in GOV.UK's guidance on costs or disbursements. Getting this wrong distorts the numbers your model produces.
Choosing between the methods
Start with the decision the model must support. If you are setting commission rates, last click or position based may be enough. If you are defending budget, incrementality testing gives stronger evidence.
Keep the definition stable for at least a quarter so trends mean something. Record the lookback window, deduplication rules and exclusions. Compare the output against finance data before you act on it. The measurement and reporting guide covers the metrics and reporting cycles in more detail.
Glossary
- Lookback window: the period during which a touch can earn credit.
- Deduplication: removing duplicate conversions from the same user journey.
- Holdout: a group excluded from a partner's activity to test its impact.
- Assist: a touch that contributes to a sale without closing it.
Common questions
Which attribution method is best for a small affiliate programme?
Last click or position based is usually enough at low volume, because data driven models need more signal.
Can I use more than one method at the same time?
Yes. Many teams report a closing model for payouts and an assist or incrementality view for planning.
How often should I review the model?
Quarterly suits most programmes, or sooner if you change networks, tracking or commission structures.
Does attribution affect publisher contracts?
The model decides what partners are paid, so it is a commercial term. If you change it mid-term, check the contract and give reasonable notice. A reporting dashboard makes those changes visible to both sides.
