What is an attribution model?
Before performing a target action, users can come to a website via different sources. A user might first find out about a site for the first time from an ad, for example, then return to the same site clicking through from search engine results, and later type the address into the browser and make a purchase. Attribution models describe the contributions of different traffic sources to conversions and subsequent visits to a site (sessions).
Attribution models help understand which channels initially brought those users who eventually converted, and which sources motivated users already familiar with the site to make a purchase.
Why do both Yandex.Direct and Yandex.Metrica need attribution models?
Statistics in Yandex.Direct and Yandex.Metrica serve different purposes. Yandex.Direct reports show you in detail how changes in campaign settings affect the effectiveness of your ad placements. In Yandex.Metrica, however, the focus on the behavior of users who came to the site via advertising, and comparing the returns from different traffic channels. So, attribution models in Yandex.Direct enable you to assess the impact of ad campaigns in detail, whereas in Yandex.Metrica, they allow you to compare the effectiveness of different sources and analyze site users' paths to conversion.
How do attribution models work in Yandex.Metrica?
Let's first take a look at how Yandex.Metrica builds reports with different attribution models. If the last non-direct click attribution model is selected in the report, the system will take all the sessions for the report’s set time period, find the last non-direct click for each of them (it may be an advertising system, search engine, site link or another source), and then show the number of sessions for which each of the sources served as the last non-direct click. The two other attribution models work in similar ways.
Example: A user clicked on an ad in Yandex.Direct on June 5, then returned to the site via Yandex Search results and made a purchase. The user then clicked through again via an ad in Yandex.Direct on June 7, and then returned to the site again on June 10 from a link bookmarked in their browser and once again made a conversion. Yandex.Metrica counts four sessions:
June 5, 1st session, source → Yandex.Direct.
June 5, 2nd session, conversion, source → Search.
June 7, 3rd session, source → Yandex.Direct.
June 10, 4th session, conversion, source → Direct traffic (click-throughs from bookmarks are counted as direct traffic).
Here's how Yandex.Metrica recorded the source of each session:
Table 1: Information about traffic sources in the Yandex.Metrica database
To make a report on traffic sources with a specified attribution model, we count the number of sessions from the first table in which each source provided the last click, last non-direct click or first click.
Here's the data shown in the traffic source report for each attribution model:
Table 2: Data shown in the Sources report based on different attribution models
The first difference from Yandex.Direct: clicks and sessions
Attribution models in Yandex.Direct work in a similar way, except that Yandex.Direct reports are based on clicks rather than sessions. After all, it's Yandex.Metrica that records sessions, while Yandex.Direct only registers clicks on ads.
The exception to this rule is the conversion rate indicator in Yandex.Direct reports. It uses conversion data from Yandex.Metrica.
Let's take a look at how the difference in each system's accounting units affects the data in statistics. Using the same example:
June 5, 1st session, source → Yandex.Direct.
June 5, 2nd session, conversion, source → Search.
June 7, 3rd session, source → Yandex.Direct.
June 10, 4th session, conversion, source → Direct traffic
Yandex.Metrica shows four sessions, while Yandex.Direct records two clicks – on June 5 and 7.
According to the last click model, Yandex.Direct reports will show two clicks, because for each click Yandex.Direct was the last source. And in the Yandex.Direct: Summary report in Yandex.Metrica, there will be two sessions, because Yandex.Direct was the last source for only two sessions (see the Last click column in Table 1).
According to the last non-direct click model in Yandex.Direct, there are two clicks and one conversion (the one that occurred on June 10). The June 5th conversion is not counted in this attribution model in Yandex.Direct, because the last non-direct click was from search results. In Yandex.Metrica, the Yandex.Direct: Summary report will show three sessions where the last non-direct click is attributed to Yandex.Direct. At the same time, one of these sessions, on June 10, is a conversion. The conversion rate in Yandex.Direct will be 1 conversion/2 clicks*100%=50%, while in Yandex.Metrica it will be 1 conversion/3 sessions total*100%=33%.
Using the first click model in Yandex.Direct, we get two clicks and two conversions. Yandex.Metrica will attribute four sessions to Yandex.Direct (because for each of the four sessions, the first source was Yandex.Direct), and two conversions. In Yandex.Direct, the conversion rate will be 2 conversions/2 clicks*100%=100%, while in Yandex.Metrica the conversion rate will be 2 conversions/4 sessions total*100%=50%.
Here you can see the differences between Yandex.Direct statistics and the statistics in the Yandex.Direct: Summary report in Yandex.Metrica:
The table compares the Yandex.Direct: Summary report data, since it does not include the traffic and conversions that the Yandex.Direct anti-fraud considered invalid, so the data in it is closer to the figures in the Yandex.Direct statistics.
The second difference from Yandex.Direct is in the conversion dates
In Yandex.Metrica, reports record conversions the same day that they actually occur. Yandex.Direct, being more concerned with showing the return on advertising rather than site users' behavior, links conversions to the date of the click that provided it.
According to the first click attribution model, this will be the first click on a Yandex.Direct ad that brought a person to the site. In the last non-direct click model, it is the last click from Yandex.Direct if there were no other significant traffic sources after it, e.g. clicks from search results, site links, or from another advertising system. With the last click model, it is the last click from Yandex.Direct if the converted session started there.
Let's take the last two sessions from our example:
June 7, 3rd session, source → Yandex.Direct.
June 10, 4th session, conversion, source → Direct traffic
This is how all types of sources for these two sessions are recorded in the Yandex.Metrica database:
In the "Sources, summary" report in Yandex.Metrica, set June 7-10 as the period to capture both sessions, and select the last click attribution model. In the Yandex.Direct line, we see one session and zero conversions. But with the last non-direct click and first click models, two sessions and one conversion are attributed to Yandex.Direct. If we specify the report period of June 8 to 10, without the June 7 traffic from Yandex.Direct, and select the first click or last non-direct click attribution models, we get one session from Yandex.Direct and one conversion. And according to the last click model, there is one conversion where the source is direct traffic.
Yandex.Direct recorded one click, not two sessions. For June 7-10, we see one conversion according to the last non-direct click and first click models. And for June 8-10, we don't see any clicks or conversions, because the click from Yandex.Direct occurred on June 7. This means there's nothing to link the conversion to, so it doesn’t appear in the report. So when you are assessing an ad campaign’s impact, you should check for the presence of conversions in a period that includes the campaign’s start date.
When to use which model
In Yandex.Direct, attribution models are needed for detailed analyses of ad campaigns’ impact. First click attribution shows you which campaigns and ads brought the users to your site who converted later. The last non-direct click model shows which ad materials were decisive in stimulating users to make a purchase. And the last click model shows the conversions that occurred almost immediately after the user clicked an ad.
In contract, attribution models in Yandex.Metrica help you compare the returns from different sources of traffic to your site: advertising systems, search engines, mailing lists, social networks and links on other sites.
Still have questions? Get in touch with us via the feedback form. We'll be glad to help you!