What is an attribution model?
An attribution model is the rule that decides which marketing touchpoints get credit for a conversion when a customer interacted with several before buying. Common models assign all credit to the first touch, all to the last, or split it across every touch by position or by recency. The model is a choice, not a measurement.
Worked example
One customer, four touches, a $400 purchase.
| Day 1: organic search on a blog post | First touch |
|---|---|
| Day 4: paid social ad | Middle |
| Day 9: newsletter link | Middle |
| Day 11: direct visit, then purchase | Last touch |
| First-touch credit | Organic search $400 |
| Last-touch credit | Direct $400 |
| Linear credit | $100 to each of the four |
| Position-based (40/20/40) | Organic $160, social $80, newsletter $80, direct $160 |
The same purchase justifies four completely different budget decisions Nothing about the customer changed between those rows. The model did. This is why attribution debates are really budget debates wearing a measurement costume.
Compared
The models, and what each one is for Every model is a defensible answer to a different question. The failure is not picking the wrong one, it is switching between them inside a single report.
| Model | Credit goes to | Best for | Blind spot |
|---|---|---|---|
| First touch | The first interaction | Judging what creates awareness | Ignores everything that closed the deal |
| Last touch | The final interaction | Simple, and the default nearly everywhere | Over-credits branded search and direct traffic |
| Last non-direct | The last identifiable source | Stops direct traffic eating the credit | Still ignores the earlier touches |
| Linear | Every touch equally | Long considered purchases | Treats a stray click like a demo call |
| Time decay | Recent touches get more | Short sales cycles | Systematically undervalues awareness work |
| Position based | 40% first, 40% last, 20% between | A reasonable default when you need one number | The weights are conventional, not measured |
Why last touch keeps winning, and why it lies
Last touch is the default in almost every analytics tool because it needs no configuration and no cross-session identity: whatever brought the converting visit gets the credit.
Its distortion is systematic and always in the same direction. Direct traffic and branded search sit closest to the purchase, so they collect credit for demand that something else created. Content marketing, display and every awareness channel look worse than they are, budget moves toward the bottom of the funnel, and eighteen months later there is no demand left to capture.
The honest reading is that last touch tells you which channel closes, not which channel works. Those are different questions and both are worth funding.
Live demo
See it on real data. The live demo is a fully populated dashboard on demo traffic, with no signup and no card.
Open the live demoWhat limits attribution in practice
Cross-device journeys
Research on a phone, purchase on a laptop. Without a login there is no way to join them, so the phone touches are simply absent from every model.
Cookie lifetimes
Safari and Firefox cap first-party cookie storage, so long consideration cycles lose their early touches regardless of which model you pick.
Dark traffic
Clicks from apps, private browsing and untagged emails arrive with no referrer and land in direct. They are real touches that no model can see.
Cookieless windows
Cookieless analytics resolves identity within a day by design, so multi-day multi-touch attribution needs cookie mode or your own user id. Within a session, campaign to conversion is exact.
A practical position
Most teams under fifty people do not need a multi-touch model. They need three things that are easier and more honest:
- Consistent UTM tagging, so the touches you can see are labelled correctly.
- Last non-direct attribution as the standing default, stated openly as a simplification.
- A self-reported source field on signup, which routinely surfaces channels analytics never sees, such as a podcast or a colleague's recommendation.
mrkr reports campaign, source and medium against goals and revenue on a last non-direct basis, which is the model that matches what a cookieless tool can honestly claim to know. Where you need longer windows, per-site cookie mode or your own user id extends them, and cookie mode requires a consent banner for that site.
The rule that survives every model change: pick one, write it into the metric definition, and do not switch it mid-quarter to win an argument.
FAQ
Which attribution model should I use?
Last non-direct is the sensible default for most teams: it is simple, it stops direct traffic absorbing credit, and everyone can explain it. Move to a position-based or time-decay model only when the sales cycle is long enough that early touches are genuinely being missed.
Can you do attribution without cookies?
Within a single session, yes and precisely: the campaign that brought the visit is attached to any conversion in that visit. Across days it needs persistent identity, which means either an optional cookie mode with consent, or your own user id sent from your backend.
Why does direct traffic get so much credit?
Because a lot of it is not really direct. Untagged email links, app clicks, private browsing and stripped referrers all fall into direct, and last-touch models then credit it for conversions other channels created. Last non-direct attribution and consistent UTM tagging are the two fixes.
Start with one line of HTML.Your first visitors show up within a minute.