Attribution and attribution models in Metrica
Contents
Share
Attribution in the Metrica glossary is a way to distribute the contribution of different sources to a conversion and to the visits that follow. The From, UTM, and Openstat tags are URL parameters Metrica handles separately: they are absent from the “URL parameters” report and, by default, land in special reports. That glossary page does not spell out the five standard UTM names and their roles.
From 20 May 2026, Metrica and Yandex Direct disabled the models “First click,” “Last significant click,” “Last click from Direct,” and “Last click from Direct, cross-device.” The date is the help’s wording, checked on 25 October 2026. The help promises to move saved reports onto the current models itself. Four models remain for a report: first click cross-device, last click, last significant click cross-device, and automatic attribution. Impression attribution on this page is placed in Metrica for display advertising and is not one of the four models.
The attribution window is visits with no break longer than 90 days. After such a break the history ends and starts again. The user does not change the window. Cross-device for the first-click and last-significant models follows device history through Crypta.
The same four visits, four splits
The help’s example: search, then Direct, then an ad, then a direct visit with an order.
| Model | Where the order goes in the example |
|---|---|
| Last click | Each visit keeps its own source; the direct visit stays direct |
| Last significant click, cross-device | The direct visit is assigned to the previous ad: direct visits are called insignificant in the example |
| First click, cross-device | All four visits are assigned to search |
| Automatic | The third and fourth visits are assigned to Direct |
“Last click” is for comparing Direct and Metrica and for a technical reading of internal transitions. “Last significant click” cross-device also fits a conversion inside a single visit. The page suggests “First click” cross-device for a delayed conversion and for the question of who brought new visitors. The page does not give the full list of “significant” sources: in the example, a direct visit is the one called insignificant.
The automatic model’s split in the example is not explained step by step. The page says it includes cross-device and households from anonymized data, but it does not disclose why the third visit went to Direct. Automatic attribution data is calculated starting 8 April 2023. Before that date the model is not built in the report. You turn it on with the switch “Attribution → Automatic attribution.” The page does not publish the makeup of a household or the cross-device rules.
Spend is matched to the campaign, not written into the visit
To see Direct spend you need to be the campaign owner, the owner’s representative, or an agency representative. File data and Direct campaign data do not supplement visits: UTM tags are matched to visits that already have those tags, and Direct campaign data is linked to that campaign. Which clicks and conversions fall into the selected period depends on the attribution model.
The example on the expenses page refers to the models “Last click from Direct” and “First click.” The models page as of 20 May 2026 calls those two disabled, along with “Last significant click” and “Last click from Direct, cross-device.” For the list of models that still work, use the four models above, not the example on the expenses page.
In Direct’s Report Wizard, CR is the share of goal visits among all visits, CPA is the average spend per goal visit, and “Conversions” is how many goal visits the ads brought. The report is built as period, goals, attribution model, then metrics. CR and CPA need Metrica goals. Revenue, profit, and ROI appear when a conversion value has been passed. How to divide spend by leads outside this model is in the article on CPA. Changing the model changes which visits fall into the period, and CR and CPA with them. These fields are for Direct goal visits, not the “per completion” and “per visitor” CPAs from Metrica’s expenses report.
By default Metrica assigns a goal to the visit in which it happened. If a person arrived from an ad without a goal and bought later from search, the conversion goes to search. Through yclid you can assign the conversion to Direct by hand, including when it happened off the site. The file has to be passed so that no more than 21 days pass between the visit with that tag and processing. Under the “last click” model, some of those visits are not visible in the report, because the ad click was not the last one.
CTR does not pick an attribution model: clicks over impressions do not decide which source gets the order. Attribution also does not move the share of clicks with no buyer — that is counted separately, as in budget loss. Robot share of a slice is not an attribution model: what the percent means is in the article on robot share.