In this guide
- Rule-based models (first click, last click, linear, time decay, position-based) split credit by a fixed rule.
- Data-driven models let an algorithm decide from your own conversion paths, but need enough data.
- Incrementality and marketing mix modelling ask whether the ads caused the sale, which is the real question.
Attribution models are the rules for dividing the credit for a sale. They are simple to describe and surprisingly easy to misuse. This page explains each family with an example, says what it hides, and suggests which to use for which decision. It follows attribution tracking and is useful alongside marketing mix modelling.
One journey, five models
A customer clicks a paid social ad (day 1), reads a blog post from search (day 4), opens an e-mail (day 7) and buys after clicking a retargeting ad (day 9). The order is worth $100. Here is how each rule credits it:
| Model | Paid social | Search | Retargeting | |
|---|---|---|---|---|
| First click | $100 | $0 | $0 | $0 |
| Last click | $0 | $0 | $0 | $100 |
| Linear | $25 | $25 | $25 | $25 |
| Time decay | $10 | $20 | $30 | $40 |
| Position based (40/20/20/20) | $40 | $10 | $10 | $40 |
The numbers for time decay and position based are illustrations of the idea, not a standard split. Different tools use different weights.
Connected in minutes
Add your domain, install the pixel, connect your ad platforms and watch the first events land.
What each model is good for
First click
Shows what starts journeys. Good for judging awareness channels. Ignores everything after.
Last click
Shows what closes sales. Simple and common. Undervalues everything that built interest.
Linear
Credits every touch the same. Fair-looking, but rarely true.
Time decay
Gives more credit to recent touches. Suits short sales cycles.
Position based
Rewards the first and last touch. A compromise.
Data-driven
An algorithm learns how much each touch contributes from your paths. Needs volume and is a black box.
Beyond rules
- Incrementality tests compare a group that sees ads with one that does not. They answer whether the ads caused the sales.
- Marketing mix modelling uses totals over time (spend and sales by channel) to estimate the effect of each channel without individual tracking.
- Surveys ("how did you hear about us") catch what clicks miss.
Works with your stack
Grab a logo and throw it. These are the tools Open Tracked already talks to.
Which to use
| Decision | Useful view |
|---|---|
| Where to put the next dollar in one channel | The platform's data-driven or last-click view |
| Is a channel worth keeping at all | Incrementality test or mix modelling |
| What starts customer journeys | First click and survey answers |
| Overall efficiency | Blended ROAS and POAS |
Data quality comes first
A model cannot credit a touch it never saw. If Safari removed the click ID, or a blocker stopped the event, the journey looks shorter than it was. Collect events on your own subdomain, send them from the server, and keep cookies alive before debating models. See Safari ITP and server-side tracking. Open Tracked's analytics lets you compare channels on first-party data next to the platforms' own figures.
A worked example
A store runs paid social, search, e-mail and a podcast. Last-click attribution gives search and e-mail almost all the credit, and the podcast none. The team is tempted to cut the podcast. Before it does, it runs a simple incrementality test: it pauses the podcast in a few regions for six weeks and compares sales with regions where it keeps running. Sales in the paused regions are lower by a margin that more than covers the podcast's cost, and the new-customer count falls the most. Last click had hidden the channel that started the journeys. The test cost six weeks and a few regions, and it saved a profitable channel.
Not every question needs an experiment. For small budget moves inside a channel, the platform's numbers are fine. For large decisions, such as cutting a channel entirely, one good test beats any amount of model comparison.
Runs on your domain, anywhere
Your tracking runs on your own domain, so a visit from any country is handled the same way.
Frequently asked questions
What is an attribution model?
A rule for dividing the credit for a conversion between the touchpoints that came before it.
What is the difference between first click and last click?
First click credits the first touch. Last click credits the final one.
Is data-driven attribution better?
It can reflect your real paths, but needs enough data and is hard to inspect.
What is incrementality?
Whether ads caused sales that would not have happened otherwise, tested by comparing exposed and unexposed groups.
Which model should I use?
Compare several and choose by the decision you are making.
Do I need spreadsheets to read my results?
No. Once tracking is live, the Open Tracked analytics shows spend, revenue, ROAS and profit, and you can ask Ari, the built-in AI assistant, questions about your own numbers in plain words. Ari's allowance depends on your analytics plan.
Is Open Tracked different from Open Track or Open Space Track?
No. Open Tracked is the official name of our server-side tracking platform, and the website is opentracked.io. Many people search for it as Open Track, OpenTrack or Open Space Track. All of them lead to the same product: server-side Google Tag Manager hosting, first-party analytics and a CRM pipeline.
Fix your tracking at the source
Open Tracked hosts your server-side Tag Manager, keeps first-party cookies alive and shows what each ad is really worth.
The Open Tracked team
We build Open Tracked: server-side tracking hosting, analytics and the tools around them, for stores that spend on ads. Questions about this guide, or an idea for the next one?
Comments
Loading comments...
Guides · 5 min read
