This document is an explainer for a potential new web platform feature which allows for measuring and reporting ad click conversions.
(Name probably needs bikeshedding)
See the explainer on aggregate measurement for a potential extension on top of this.
Table of Contents
- Glossary
- Motivation
- Prior Art
- Overview
- Sample Usage
- Privacy Considerations
- Open Questions
-
Publisher: Page that shows ads, sells ad slots
-
Advertiser: Purchaser of ad slots, conversions happen on advertiser sites
-
Conversion: The completion of a meaningful (advertiser specified) user action on the advertiser's web site by a user who has previously interacted with an ad from that advertiser.
-
Event-level data: Data that can be tied back to a specific low-level event; not aggregated
-
Click-through-conversion (CTC): A conversion attributed to an ad click
Currently, the web ad industry measures conversions via identifiers they can associate across sites. These identifiers tie information about which ads were clicked to information about activity on the advertiser's site (the conversion). This allows advertisers to measure ROI, and for the entire ads ecosystem to understand how well ads perform.
Since the ads industry today uses common identifiers across advertiser and publisher sites to track conversions, these common identifiers can be used to enable other forms of cross-site tracking.
This doesn’t have to be the case, though, especially in cases where identifiers such as third-party cookies are either unavailable or undesirable. A new API surface can be added to the web platform to satisfy this use case without them, in a way that provides better privacy to users.
This API alone will not be able to support all conversion measurement use cases, such as view conversions, or even click conversion reporting with richer / more accurate conversion data. We envision this API as one of potentially many new APIs that will seek to reproduce valid advertising use cases in the web platform in a privacy preserving way. In particular, we think this API could be extended by using server side aggregation to provide richer data, which we are continuing to explore.
There is an alternative Ad Click Attribution draft spec in the WICG. See this WebKit blog post for more details.
Brave has published and implemented an Ads Confirmation Protocol.
An attribution source is an anchor tag with special attributes:
<a attributiondestination="[eTLD+1]" attributionsourceeventid="[64-bit unsigned integer]" attributionexpiry=[unsigned long long] attributionreportto="[origin]">
Attribution source attributes:
-
attributiondestination: an origin whose eTLD+1 is where attribution will be triggered for this source. -
attributionsourceeventid: A DOMString encoding a 64-bit unsigned integer which represents the event-level data associated with this source. This will be limited to 64 bits of information but the value can vary for browsers that want a higher level of privacy. -
attributionexpiry: (optional) expiry in milliseconds for when the source should be deleted. Default is 30 days, with a maximum value of 30 days. The maximum expiry can also vary between browsers. -
attributionreportto: (optional) the desired endpoint that the attribution report for this source should go to. Default is the top level origin of the page.
Clicking on an anchor tag that specifies these attributes will create a new attribution source event that will be handled according to Handling an attribution source event
An attribution source can be registered for navigations initiated by window.open().
A source is registered in a call to window.open() by providing attribution source attributes within the features parameter:
window.open("https://dest.example", "_blank",
"noopener,attributionsourceeventid=1234,attributiondestination=https://dest.example,\
attributionreportto=https://reporter.example,attributionexpiry=604800000");)
At the time window.open() is invoked with these attributes specified in features, if the associated window has a transient activation, an attribution source event will be created and handled following Handling an attribution source event.
An attribution source event will be logged to storage if the resulting document being
navigated to ends up sharing the an eTLD+1 with the attributionDestination origin. Concretely, this logs <attributionsourceeventid,
attributiondestination, attributionreportto, attributionexpiry> to a new
browser storage area.
When a source is logged for <attributionreportto,
attributiondestination>, existing sources matching this pair will be
looked up in storage. If the matching sources have been triggered at
least once (i.e. have scheduled a report), they will be removed from
browser storage and will not be eligible for further reporting. Any
pending reports for these sources will still be sent.
An attribution source will be eligible for reporting if any page on the
attributiondestination eTLD+1 (advertiser site) triggers attribution for the
associated reporting origin.
In order to prevent arbitrary third parties from registering sources without the publisher’s knowledge, the Attribution Reporting API will need to be enabled in child contexts by a new Feature Policy:
<iframe src="https://advertiser.test" allow="attribution-reporting 'src'">
<a … id="..." attributionreportto="https://ad-tech.com"></a>
</iframe>
The API will be enabled by default in the top-level context and in same-origin children. Any script running in these contexts can declare a source with any reporting origin. Publishers who wish to explicitly disable the API for all parties can do so via an HTTP header.
Without a Feature Policy, a top-level document and cooperating iframe could recreate this functionality. This is possible by using postMessage to send the attributionsourceeventid, attributionreportto, attributiondestination values to the top level document who can then wrap the iframe in an anchor tag (with some additional complexities behind handling clicks on the iframe). Using Feature Policy prevents the need for these hacks. This is inline with the classification of powerful features as discussed on this issue.
This API will use a mechanism for triggering an attribution source similar to the Ad Click Attribution Proposal.
Attribution can only be triggered for a source on a page whose eTLD+1 matches the eTLD+1 of the site provided in attributiondestination. A source can be triggered through an HTTP GET to
its attributionreportto origin, that redirects to a .well-known
location.
This redirect is useful, because this mechanism enables the reporting origin to make server-side decisions about when attribution reports should trigger.
Note that .well-known is only used to register a path that the browser will understand; it shouldn't point to any actual resource, since the request will be cancelled internally.
Triggering attribution requires the attribution-reporting Feature Policy to be enabled in the context the request is made. As described in Publisher Controls for Attribution Source Declaration, this Feature Policy will be enabled by default in the top-level context and in same-origin children, but disabled in cross-origin children.
Today, conversion pixels are frequently used to register conversions on advertiser pages. These can be repurposed to trigger attribution in this API:
<img src="https://ad-tech.test/conversiontracker"/>
https://ad-tech.test/conversiontracker can be redirected to https://ad-tech.test/.well-known/attribution-reporting/trigger-attribution
to trigger attribution for all matching sources.
The browser will treat redirects to a URL of the form:
https://<attributionreportto>/.well-known/attribution-reporting/trigger-attribution[?data=<data>]
as a special request, where optional data associated with the event that triggered attribution is stored in a query parameter.
When the special redirect is detected, the browser will schedule an attribution report as detailed in Trigger attribution algorithm.
The attributionsourceeventid will be limited to 64 bits of information to enable
uniquely identifying an ad click.
The advertiser-side data must therefore be limited quite strictly, by limiting the amount of data and by applying noise to the data. Our strawman initial proposal is to allow 3 bits of trigger data, with 5% noise applied — that is, with 5% chance, we send a random 3 bits, and the other 95% of the time we send the real trigger data. See privacy considerations for more information, including speculative thoughts on the hard question of going farther and adding noise to whether or not the attribution report is even sent. In any case, noise values should be allowed to vary between browsers.
Disclaimer: Adding or removing a single bit of data has large trade-offs in terms of user privacy and usability to advertisers:
- Less bits is more private but less usable to advertisers
- More bits is less private but more usable to advertisers. Browsers should concretely evaluate the trade-offs from these two perspectives before setting a limit. As such, this number is subject to change based on community feedback. Our encoding scheme should also support fractions of bits, as it’s possible to limit data to values from 0-5 (~2.6 bits of information)
When the browser receives a attribution trigger redirect on a URL matching
the attributiondestination eTLD+1, it looks up all sources in storage that
match <attributionreportto, attributiondestination>.
The most recent matching source is given a credit of value 100. All other matching sources are given a credit of value 0.
For each matching source, schedule a report. To schedule a report,
the browser will store
{attributionreportto, attributiondestination eTLD+1, attributionsourceeventid, decoded trigger-data, credit} for the source.
Scheduled reports will be sent as detailed in Sending scheduled reports.
Each source is only allowed to schedule a maximum of three reports (see Triggering attribution multiple times for the same source). Once reports are scheduled, the browser will delete all sources that have scheduled three reports.
If multiple sources were clicked and associated with a single attribution trigger, send reports for all of them.
To provide additional utility, the browser can choose to provide additional annotations to each of these reports, attributing credits for the triggering redirect to them individually. Attribution models allow for more sophisticated, accurate measurement.
The default attribution model will be last-click attribution, giving the last-clicked source for a given trigger redirect all of the credit.
To remain flexible, the browser sends an credit of value 0 to 100 for all reports associated with a single trigger. This represents the percent of attribution a source received. The sum of credits across a set of reports for one trigger should equal 100.
There are many possible alternatives to this, like providing a choice of rules-based attribution models. However, it isn’t clear the benefits outweigh the additional complexity. Additionally, models other than last-click potentially leak more cross-site information if sources are clicked across different sites.
Many ad clicks end up converting multiple times, for instance if a user goes through a checkout and a purchase flow. To support this in a privacy preserving way, we need to make sure that triggering a source multiple times does not leak too much data.
One possible solution, outlined in this document, is for browsers to specify a maximum number of reports that can be sent for a single source. In this document our initial proposal is 3.
Note that triggering attribution for the same source multiple times does not refresh the reporting windows (see Sending Scheduled Reports).
Note that from a usability perspective, it is important that all reports for the same source are allowed the same amount of data. Otherwise, it becomes quite difficult for advertisers to efficiently use the space of possible data values.
After an attribution source is successfully registered, a schedule of reporting windows and deadlines associated with that source begins. The time between the click and expiry can be split into multiple reporting windows, at the end of which the browser will send scheduled reports for that source.
Each reporting window has a deadline, and only reports created before that deadline can be sent in that window. An example of deadlines and windows a browser could choose are:
2 days minus 1 hour: Reports will be sent 2 days from source registration time
7 days minus 1 hour: Reports will be sent 7 days from source registration time
attributionexpiry: Reports will be sent attributionexpiry
milliseconds plus one hour from source registration time
If attributionexpiry occurs before the 7 day window deadline it will be used as the next reporting window. For example, if attributionexpiry is 3 days, there will be two deadlines, 2 days minus one hour and attributionexpiry. If attributionexpiry is before the 2 day deadline, the 2 day deadline will still be used.
When a report is scheduled, it will be delayed until the next applicable reporting window for the associated source. Once the window has finished, the report will be sent out of band.
If there are multiple reports for a source scheduled within the same window, the reports will be sent at the same time but in a random order.
The report may be sent at a later date if the browser was not running when the window finished. In this case, reports will be sent on startup. The browser may also decide to delay some of these reports for a short random time on startup, so that they cannot be joined together easily by a given reporting origin.
Note that to improve utility, it might be possible to randomly send reports throughout each reporting window.
To send a report, the browser will make a non-credentialed (i.e. without session cookies) secure HTTP POST request to:
https://attributionreportto/.well-known/attribution-reporting/report-attribution
The report data is included in the request body as a JSON object with the following keys:
-
source_event_id: 64 bit event id set on the attribution source -
trigger_data: 3 bit data set in the attribution trigger redirect -
credit: integer in range [0, 100], denotes the percentage of credit this source received for the given trigger. If a trigger only had one matching source, this will be 100.
The advertiser site’s eTLD+1 will be added as the Referrer. Note that it might be useful to advertise which data limits were used in the report, but it isn’t included here.
It also may be beneficial to send reports as JSON instead of in the report URL. JSON reports could allow this API to leverage the Reporting API in the future should it be desirable.
The source event id and trigger data should be in a way that is amenable to any privacy level a browser would want to choose (i.e. the number of distinct data states supported).
The input values will be 64 bit integers which the browser will interpret modulo its maximum data value chosen by the browser. The browser will take the input and performs the equivalent of:
function getData(input, max_value) {
return input % max_value;
}
The benefit of this method over using a fixed bit mask is that it allows browsers to implement max_values that aren’t multiples of 2. That is, browers can choose a "fractional" bit limit if they want to.
publisher.com wants to show ads on their site, so they contract out to
ad-tech.com. ad-tech.com's script in the main document creates a
cross-origin iframe to host the third party advertisement for
toasters.com, and sets ad-tech.com to be an allowed reporting origin.
Within the iframe, toasters.com code annotates their anchor tags to use
the ad-tech.com reporting origin, and uses a source event id value that allows
ad-tech.com to identify the ad click (12345678)
<iframe src="https://ad-tech-3p.test/show-some-ad" allow="attribution-reporting ‘src’ (https://ad-tech.com)">
...
<a
href="https://toasters.com/purchase"
attributiondestination="https://toasters.com"
attributionsourceeventid="12345678"
attributionreportto="https://ad-tech.com"
attributionexpiry=604800000>
...
</iframe>
A user clicks on the ad and this opens a window that lands on a URL to
toasters.com/purchase. An attribution source is logged to browser storage
since the landing page matches the attributiondestination eTLD+1. The following data is
stored:
{
attributionsourceeventid: 12345678,
attributiondestination: https://toasters.com,
attributionreportto: https://ad-tech.com,
attributionexpiry: <now() + 604800>
}
2 days later, the user buys something on toasters.com. toasters.com
triggers attribution on the few different ad-tech companies it buys
ads on, including ad-tech.com, by adding conversion pixels:
<img src="https://ad-tech.com/trigger-attribution?model=toastmaster3000&price=$49.99&..." />
ad-tech.com receives this request, and decides to trigger attribution
on toasters.com. They must compress all of the data into
3 bits, so ad-tech.com chooses to encode the value as “2" (e.g. some
bucketed version of the purchase value). They respond with a 302
redirect to:
https://ad-tech.com/.well-known/attribution-reporting/trigger-attribution?data=2
The browser sees this request, and schedules a report to be
sent. The report is associated with the 7 day deadline as the
2 day deadline has passed. Roughly 5 days later, ad-tech.com receives
the following HTTP POST:
URL:
https://ad-tech.com/.well-known/attribution-reporting/report-attribution
body:
{
"source_event_id": "12345678",
"trigger_data": "2",
"credit": 100
}
The main privacy goal of the API is to make linking identity between two different top-level sites difficult. This happens when either a request or a Javascript environment has two user IDs from two different sites simultaneously.
In this API, the 64-bit source ID can encode a user ID from the publisher’s top level site, but the low entropy, noisy trigger data could only encode a small part of a user ID from the advertiser’s top-level site. The source ID and the trigger data are never exposed to a Javascript environment together, and the request that includes both of them is sent without credentials and at a different time from either event, so the request adds little new information linkable to these events.
While this API does allow you to learn "which ad clicks converted", it isn’t enough to link the user's identity on the publisher's and advertiser's side, unless there is serious abuse of the API, i.e. abusers are using error correcting codes and many clicks to slowly and probabilistically learn advertiser IDs associated with publisher ones. We explore some mitigations to this attack below.
Trigger data, e.g. advertiser side data, is extremely important for critical use cases like reporting the purchase value of a conversion. However, too much advertiser side data could be used to link advertiser identity with publisher identity.
Mitigations against this are to provide only coarse information (only a few bits at a time), and introduce some noise to the trigger data. Even sophisticated attackers will therefore need to invoke the API many times (through many clicks) to join identity between sites with high confidence.
Note that this noise still allows for aggregate measurement of bucket sizes with an unbiased estimator. See generic approaches of dealing with Randomized response for a starting point.
By bucketing reports within a small number reporting deadlines, it becomes harder to associate a report with the identity of the user on the advertiser’s site via timing side channels.
Reports within the same reporting window occur within an anonymity set with all others during that time period. For example, if we didn’t bucket reports, the reports (which contain publisher ids) could be easily joined up with the advertiser’s first party information via correlating timestamps.
Note that the delay windows / deadlines chosen represent a trade-off with utility, since it becomes harder to properly assign credit to a click if the time from click to conversion is not known. That is, time-to-conversion is an important signal for proper attribution. Browsers should make sure that this trade-off is concretely evaluated for both privacy and utility before deciding on a delay.
If the advertiser is allowed to cycle through many possible
attributionreportto origins (via injecting many <img> tags on the
page), then the publisher and advertiser don’t necessarily have to agree
apriori on what origin to use, and which origin actually ends up getting
used reveals some extra information.
To prevent abuse, it makes sense for browsers to add limits here, potentially on a per-page load or per-reporting epoch basis.
Attribution source data and attribution reports in browser storage should be clearable using existing “clear browsing data" functionality offered by browsers.
To limit the amount of user identity leakage between a <publisher, advertiser> pair, the browser should throttle the amount of total information sent through this API in a given time period for a user. The browser should set a maximum number of attribution reports per <publisher, advertiser, user> tuple per time period. If this threshold is hit, the browser will disable the API for the rest of the time period for that user.
The longer the cooldown windows are, the harder it is to abuse the API and join identity. Ideally report thresholds should be low enough to avoid leaking too much sensitive information, with cooldown windows as long as practically possible.
It’s an open question what specific limits are possible here.
Another mitigation on joining identity across publisher and advertiser sites is to limit the number of reports for any given <publisher, advertiser> pair until the advertiser clears their site data. This could occur via the Clear-Site-Data header or by explicit user action.
To prevent linking across deletions, we might need to introduce new options to the Clear-Site-Data header to only clear data after the page has unloaded.
Another way to add privacy to this system is to add noise not only to the reported trigger data value, but also to whether attribution was triggered in the first place. That is:
-
With some probability p, sources that are triggered will be dropped
-
With some probability q, sources that have not been triggered will be triggered, and given random
trigger-data.
The biggest problem with this scheme is that attribution trigger events are, in general, rare. Additionally, different advertisers can have wildly different attribution rates. These two facts make it very hard to pick a q that works reliably without drowning out the signal with noise. We’re still thinking of solutions here.
Additionally, sending reports for sources that never actually converted could have real monetary impact on advertisers that pay per attributed source. Tight bounds on error estimation will be crucial for correct billing in these cases.
An advertiser may want to send reports to multiple reporting partners at the same time for the same attributed source. This is very tricky to get right without revealing any extra information. Allowing different trigger data for different reporting endpoints makes things even more difficult.
This problem becomes a bit easier if reporting partners mutually trust each other, and can share reporting server-to-server.