What is an attribution model?
Attribution means assigning credit. An example of a typical customer journey:
- On Monday, a customer clicks on a Google Ads ad and looks at a product.
- On Wednesday, she searches for the brand name and returns via an organic search result.
- On Friday, she clicks on a link in the newsletter and buys.
Which channel brought in the sale? Under the “last click” model, the newsletter alone. Under a “first click” model, the ad alone. Other models split the value across all three. Each of these answers is a rule, not the truth. What matters is that you know which rule your report applies.
Rule-based models
Rule-based models distribute the value according to a fixed pattern:
- Last click: the last touchpoint before the conversion receives everything.
- First click: the first touchpoint receives everything.
- Linear: all touchpoints receive an equal share.
- Time decay: touchpoints shortly before the conversion receive more than early ones.
- Position-based: the first and last touchpoints receive the largest share, and the others split the rest.
Google removed the first click, linear, time decay and position-based models from Google Analytics 4 in November 2023, and Google Ads no longer supports them either: it now offers only last click and data-driven attribution. However, you will still come across these rules in other tools and in affiliate networks.
Data-driven attribution in GA4 and Google Ads
Data-driven attribution has no fixed pattern. According to Google Analytics Help, a machine learning model compares paths that led to a conversion with paths that did not. From this, it estimates how much each touchpoint contributed to the result.
In Google Analytics 4, three models are available: data-driven, paid and organic last click, and Google paid channels last click. There are a few details you should know:
- Direct visits receive no credit as long as the path also contains other touchpoints.
- The lookback window, i.e. the period in which touchpoints are taken into account, is 30 days by default for acquisition key events such as the first visit, and 90 days for all others.
- Changing the reporting attribution model also applies retroactively. Figures in older reports can therefore change.
In Google Ads, data-driven attribution is the default for most conversion actions. It takes into account clicks and video engagements on Search, Shopping, YouTube, Display and Demand Gen. Every conversion action can use it, but for more accurate results Google recommends at least 200 conversions and 2,000 ad interactions within 30 days. Bidding strategies in Google Ads learn from the conversions distributed in this way.
Limits of attribution models
- Only measured touchpoints count: a billboard, a trade fair stand, a recommendation from a colleague or a phone call do not appear in any path.
- Gaps caused by consent and devices: if someone declines cookies or switches from their phone to a laptop, a path breaks into several incomplete parts.
- Each platform mainly sees itself: Google Ads only distributes conversions across Google ads. An affiliate network only sees its publishers. If you add up the successes reported by all platforms, the total can be higher than what was actually sold.
- Attribution is not impact: even a data-driven model estimates contributions from patterns. Whether a channel brings additional sales or only captures buyers who would have bought anyway can only be shown by controlled tests with a test group and a control group.
Attribution reports are therefore well suited to comparing channels and spotting trends. They are not suitable as an exact account of advertising success.
Attribution in practice: tracking switch and clean data
When several advertising channels claim the same sale, it gets expensive: without coordination, affiliate networks and advertising platforms each report the same order as their own success, and in the worst case you pay commission twice. A tracking switch determines which channel a sale is reported to when several channels were involved.
How well an attribution model works also depends on how completely touchpoints are recorded. Consent, browser restrictions and the measurement method all play a part. Our entry client-side tracking vs. server-side tracking explains the differences, and you can read more about implementation on our server-side tracking page.
For day-to-day work: choose one model as your standard, use it consistently in all reports and deliberately compare several perspectives when making budget decisions.
Do your channels report more sales than you actually have, or are you unsure which report to trust? Then get in touch with us or read more about our tracking switch and our server-side tracking!
