What is an attribution model?

An attribution model does not measure reality, it divides it. The analytics tool stores a list of touchpoints tied to a user, and once that user converts it needs a rule to decide where the conversion is booked. That rule is the model.

It is worth separating from three things it gets confused with every day. Tracking is the capture of the data: tags, campaign parameters, consent. Without correct tracking, no model produces reliable figures. The lookback window defines how far back touchpoints are considered at all, which is why two properties with the same model and a different window will show different results. Conversion counting decides whether you record one conversion or every conversion per click.

You also have to distinguish where the model acts. In Google Analytics 4 the reporting model changes how channels look in the key event reports. In Google Ads the model additionally feeds automated bidding, because smart strategies learn from conversions exactly as the model presents them. The same order can appear with a different source on each platform without either one being broken.

The mechanism has four steps. First the user is identified across visits with whatever identifier is available: User-ID, Google Signals, first-party cookies or modelling. Then the interactions are sorted into a path. Next that path is trimmed by the lookback window, which in Google Analytics 4 defaults to 90 days for most key events, with options of 30 and 60 days, and to 30 days for acquisition events, with an option of 7. Finally the distribution rule is applied.

The rules fall into two families. Rules-based models assign credit according to a fixed position in the path: last click gives 100 % of the value to the final interaction. Data-driven models calculate the split from the account history, comparing paths that ended in a conversion with paths that did not, so the weight of each channel varies case by case.

The available catalogue is considerably shorter than older articles suggest. Google Analytics 4 offers three reporting models today: data-driven attribution, paid and organic last click (which ignores direct traffic unless the whole path is direct) and Google paid channels last click. Google Ads offers two: data-driven attribution and last click. The first click, linear, time decay and position-based models stopped being available in November 2023, and the conversion actions that used them were moved to data-driven attribution.

The decision that hangs on the model is where you put the money. Under last click, the channels that close the sale collect the credit and the ones that open the journey look poor. Cut budget on that reading and you cut exactly what creates the demand another channel later closes. An example, with invented figures for illustration: a 200 € sale that started with a video, continued through an article and ended in a brand search is booked entirely to the brand under last click, whereas a data-driven model would spread that value across all three.

There is a second effect that gets discussed less. In Google Ads the model feeds automated bidding, so switching it changes campaign behaviour and not only the look of the report.

And there is a documentation hygiene problem. Any dashboard, template or report that still says «linear» or «first click» describes something that no longer exists on these platforms; the figures it shows come from a different model even though the label stayed put. In Google Analytics 4 a change of reporting model applies to historical data as well, so two exports of the same period made on different dates need not agree. Before arguing about a channel decline, check which model was active for each export.

• Set one default model for the whole organisation and record the date of any change next to the reports that use it; without that note, a model change later reads as a performance change.
• Review the lookback window at the same time as the model: long purchase cycles with a short window discard the early contacts before anything is distributed.
• Always compare the same period with the same model on the same platform; a comparison between Analytics and Google Ads answers two different questions.
• Validate the tracking before blaming the model: consistent UTM tagging, working consent, no duplicated conversions.
• Keep a raw export of the conversion paths (BigQuery or CSV) so you can rebuild your own split when the platform retires the next model.
• Cross-check the reading against a source that does not depend on digital attribution: a how-did-you-hear field in the form, promo codes, or data from your CRM.

• Comparing two reports built with different models and presenting the difference as a drop in traffic or sales.
• Reusing old templates and decks that still offer first click, linear, time decay or position-based, models retired in November 2023.
• Judging demand-generation campaigns by last click and concluding they contribute nothing, when the model is built not to see them.
• Forgetting that the paid and organic last click model ignores direct traffic, which skews the reading for brands with heavy direct access.
• Switching the model in Google Ads in the middle of peak season and crediting the creative for a change in results that automated bidding caused.