How does user-initiated data clearing impact attribution reports? How does that compare with cookie-based measurement?
The Attribution Reporting API makes it possible to measure when an ad click leads to a conversion on an advertiser site, such as a sale or a sign-up. The API offers a privacy-preserving approach to measuring ad conversions. It doesn't rely on third-party cookies or mechanisms that can be used to identify individual users across sites. Instead, it works as follows: attribution reports that link an ad click with a conversion are generated and stored on the user's device; later, the browser sends these reports to a predefined endpoint.
Sites have experimented with the Attribution Reporting API in Chrome, via an origin trial, which has now concluded.
One of the insights provided by these early experiments is the impact of user-initiated data clearing, such as browser history clearing, on the data advertisers and ad tech companies get from the API. Aggregate Chrome statistics were recently published. This post addresses answering common questions on the impact of user-initiated data clearing on attribution reports.
How does user-initiated data clearing impact attribution reports?
And how does that compare with cookie-based measurement?
About the numbers
Multiple factors influence these numbers:
- Whether users clear history on sites they navigate to by clicking on an ad;
- Whether users convert on categories of sites where they tend to clear history—possibly shortly after converting. Ads on this type of content might experience higher rates of clearing and hence higher discrepancies with cookie-based measurement;
- Similarly, whether users clear history on sites they click ads to;
- Whether the attribution is configured to expire a long or short time after click
(
attributionexpiry); - Whether a long time actually elapses between click and conversion.
Chrome teams will continue to monitor and publish statistics on the mailing list for developers.
Data clearing occurring after a click and before a conversion
With cookies, some percentage of conversions aren't reported due to post-click pre-conversion data clearing. Because cookies have been cleared, there's no cookie to attach to a request at conversion time, thus that conversion can't be measured. The exact percentage of conversions that aren't reported due to user-initiated data clearing isn't known. It varies across ad tech companies (or advertisers).
With the Attribution Reporting API, some percentage of conversions aren't reported due to post-click pre-conversion data clearing. Observations of early experiments (origin trial) using the Attribution Reporting API have shown that about 16% of sources (click events) are deleted before conversion. A percentage of these sources lead to conversions, and the reports for these conversions won't be sent if the sources have been deleted.
Example
For a 10% conversion rate, assuming a total of 1000 clicks and without taking other errors into consideration:
- If users were never clearing any data: an ad tech company would observe 100 conversions.
- Considering data clearing:
- With cookies: data clearing occurring after a click and before a conversion causes some percentage of conversions to not be attributed to certain click events, since there is no cookie to map the click and the conversion. With our example, an ad tech company would observe less than 100 conversions.
- With Attribution Reporting: data clearing occurring after a click and before a conversion caused 1.6% of reports that signified a conversion to be cleared (16% of 10% = 1.6%) in the first origin trial. Without taking other errors into consideration, an ad tech company would receive 84 reports that signal a conversion, instead of 100.
Data clearing occurring after a conversion
With cookies, post-conversion data clearing has no effect, because conversions are reported immediately to the ad tech company.
With the Attribution Reporting API, post-conversion data clearing causes reports to be cleared in order to honor user choice—such as a user clearing browser history, or deleting site data. Because the API doesn't send the reports immediately but with a delay in order to protect user privacy, this means that the browser storage is already empty when the scheduled time comes from the browser to send the report to the predefined endpoint—typically an ad tech company.
Observations of early experiments (origin trial) using the Attribution Reporting API have shown that about 6.5% of reports have been deleted in this way.
What exact user actions can impact attribution reports?
As of Chrome 94—the stable Chrome version at the time of this writing—any of the following actions will clear stored click events and pending reports.
chrome://settings> Privacy and Security > Clear browsing data- Check the Browsing History checkbox
- and/or Check the Clear Cookies and other site data checkbox
- Click Clear data
chrome://settings> Privacy and Security > Cookies and other site data- Toggle Clear cookies and site data when you close all windows
- Or add a behavior under Always clear cookies when windows are closed
chrome://history:- Delete any individual entry
Site-scoped controls:
- site-scoped history controls
- Click the lock button in the URL bar, Navigate to Site Settings, Click Clear Data.
This list isn't exhaustive. These are common user actions that impact attribution data. Other user actions, such as uninstalling Chrome or running a system cleaner, would also impact measurement—whether it's based on the Attribution Reporting API or on cookies.
Does user-initiated data clearing account for any observed loss?
Organizations that have experimented with the API via an origin trial may have observed a discrepancy between cookie-based measurement and Attribution Reporting based measurement: less conversions may have been reported for Attribution Reporting. User-initiated data clearing may account for observed loss, but the discrepancy must be monitored in the future in order to answer this question with certainty.
Two elements are known to have played a role in the discrepancy between cookie-based measurement and Attribution Reporting based measurement in the past origin trial:
- User-initiated data clearing.
- Network errors. These are considered bugs and should be fixed in Chrome 94.
How do we balance privacy and utility in the Attribution Reporting API with respect to user-initiated data clearing?
During the testing phase of the API, the intent is to experiment with some of the parameters of the
API to observe the effect on reports sent balanced against identifiability of the user. One of these
parameters may be the reporting delay.
Ecosystem discussions are ongoing in the regular WICG
meetings
to explore these parameters.
Will aggregate reports be impacted in the same way?
We don't know for sure how of user-initiated data clearing will impact aggregate reports.
The statistics shared in this post are for event-level reports. Even though the reporting delays may be shorter for aggregate reports—a few hours, whereas event-level reports may be sent days or weeks after click—the impact of user-initiated data clearing may not be lower. For example, for certain categories of sites, people tend to clear the browsing data shortly after converting. Such user-initiated actions would still result in reports being cleared, even though they were scheduled to be sent soon.
Attribution Reporting: all resources
Get updates
- To be notified of status changes in the API, join the mailing list for developers.
- To closely follow all ongoing discussions on the API, click the Watch button on GitHub. This requires you to have or create a GitHub account.
- To get overall updates on the Privacy Sandbox, subscribe to the RSS feed [Progress in the Privacy Sandbox].