Align ARA and PCM on last-click/last-touch attribution?

[johnwilander]

You write this in the click-through case:

Multiple sources for the same trigger (Multi-touch)

If multiple sources were clicked and associated with a single attribution trigger, send reports for the one with the highest priority. If no priority is specified, the browser performs last-touch.

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.

I recently started working on support for wildcard triggering events and unlinkable tokens (a fraud prevention measure) on the click destination side in PCM issue 88. To not have to request multiple tokens to be signed for every wildcard trigger, I started exploring last-click attribution, or last-touch in your parlance.

I talked to @johnivdel at this week's Privacy CG call and promised to discuss with you on getting aligned. As you point out in the quoted text, last-click/last-touch reduces complexity significantly. What are your thoughts? Maybe you already have meeting notes that back up "it isn’t clear the benefits outweigh the additional complexity"? Thanks!

[csharrison]

Thanks for filing John. We are experimenting with a more complicated set-up that is simpler than full multi-touch but more complex than last-touch: one that allows each source event to be labeled with a priority:
https://wicg.github.io/conversion-measurement-api/#element-attrdef-a-attributionsourcepriority

This allows some more complicated schemes like first-touch, or probabilistic linear. However, it seems like in general it aligns with your requirement that you want to just attribute one source to one trigger. One important use-case for priority which has come up is the ability to have different priorities across clicks and views. I know PCM doesn't currently support views, but if this comes up in the future it will be important to dedupe properly. We've heard that it is common for clicks to be higher priority.

[johnwilander]

Thanks for filing John. We are experimenting with a more complicated set-up that is simpler than full multi-touch but more complex than last-touch: one that allows each source event to be labeled with a priority:
https://wicg.github.io/conversion-measurement-api/#element-attrdef-a-attributionsourcepriority

@johnivdel told me about this click source priority functionality in ARA and I see how it can be used to pick attribution model.

This allows some more complicated schemes like first-touch, or probabilistic linear. However, it seems like in general it aligns with your requirement that you want to just attribute one source to one trigger. One important use-case for priority which has come up is the ability to have different priorities across clicks and views. I know PCM doesn't currently support views, but if this comes up in the future it will be important to dedupe properly. We've heard that it is common for clicks to be higher priority.

Yeah, the important piece here that I'd like to get alignment on is that one triggering event results in one click source getting attribution, at least attribution with full fidelity (in our case a signed unlinkable token from the merchant). Thanks!

[csharrison]

Hey John sorry for the delay on this.

Yeah, the important piece here that I'd like to get alignment on is that one triggering event results in one click source getting attribution, at least attribution with full fidelity (in our case a signed unlinkable token from the merchant). Thanks!

For now I think we are aligned, although I wanted to mention that it is a future goal to allow more complicated multi-touch use-cases. I don't think we have a concrete design right now, but you could imagine something like our worklet proposal with a small change where an entire list of attributionSourceContexts are submitted as input.

With a complex attribution model like that, I would be very concerned if we needed to submit tokens for each impression, so I am not extremely concerned with short term plans that limit a report to sending two tokens. I think even if we supported a complex multi-touch attribution mode, we might be able to tolerate last-touch tokens.

[csharrison]

Closing for now, will re-open if things change on our end.

See https://github.com/WICG/attribution-reporting-api/blob/main/report_verification.md for more information on the proposal.