Skip to main content

On which date should my events be attributed?

Attribution is central to Webmarketer. It determines which traffic source is responsible for an event, a state or a statistic.

To keep analysis consistent, an event is not attributed on the date it happens, but on the date the interaction took place. This article covers that concept and explains why it is crucial to work this way for a consistent analysis of your performance.

Attribution date​

When attributing an event, a state or a statistic to an interaction, the attribution date is the date the interaction took place, and not the date the event, state or statistic was recorded.

Counter-intuitive as it may seem, this is in fact necessary to guarantee correct attribution of the data, and to make it possible to compare the performance of the various traffic sources.

Consider the following example​

Say you want to measure the ROAS of a Google Ads campaign that ran for 24 hours. On February 10 you launch the campaign, spend €15 and generate 30 clicks to your website, but no purchase happens that same day. On February 15 (5 days later) one of the visitors comes back to your website through organic search and makes a €100 purchase.

With a linear model, we could simply attribute those €100 on February 15, 50% to Google Ads and 50% to organic search.

But that would give a distorted view of the Google Ads campaign's performance, because looking at February 15 alone we have:

  • €0 of ad spend for Google Ads (the campaign only ran for a day, so the €15 were spent on February 10)
  • €50 of revenue for Google Ads (the €100 purchase being 50% attributed to Google Ads)

Which gives us a ROAS of €50 / €0 = infinite for Google Ads.

warning

That figure is obviously wrong, because in our case it is the ad spend incurred on Google Ads that made the visitor — who eventually purchased on February 15 — discover our brand.

Likewise, looking at February 10 we have:

  • €15 of ad spend for Google Ads
  • €0 of revenue for Google Ads

Hence a ROAS of €0 / €15 = 0

warning

That analysis is wrong too: attributing this way does not let you match the ad spend with the revenue of the Google Ads campaign when you only look at a portion of the time period.

An illustrated example​

To illustrate the problem more clearly, here is a more visual example:

Fixing the analysis​

Let us now revisit our example, this time attributing on the interaction date rather than on the purchase date:

We now have €50 attributed on February 10 and €50 attributed on February 15.

Looking at February 10 we have:

  • €15 of ad spend for Google Ads
  • €50 of revenue for Google Ads

Hence a ROAS of €50 / €15 = 3.33

Looking at February 15 we have:

  • €0 of ad spend for organic search
  • €50 of revenue for organic search

Hence a ROAS of €50 / €0 = infinite (since we have no ad spend for organic search)

That analysis is far more consistent, and aligns revenue with the dates the ad spend was incurred.

Troubleshooting​

The number of events in the performance explorer doesn't match​

When using the performance explorer, it is normal not to find the number of events that happened over the period you are looking at. As explained on this page, events are attributed on the interaction date and not on the date they were recorded, so an event that happened on a given day of the observed period may be partially attributed to another day, outside that period.