Skip to content

Latest commit

 

History

History
197 lines (151 loc) · 8.93 KB

2021-09-20-minutes.md

File metadata and controls

197 lines (151 loc) · 8.93 KB
Raw

Attribution Reporting API

Sep 20, 2021

Meet link: https://meet.google.com/jnn-rhxv-nsy

Previous meetings: https://github.com/WICG/conversion-measurement-api/tree/main/meetings

Use Google meet “Raise hand” for queuing.

If you can’t edit the doc during the meeting, try refreshing as permissions may have been updated after you loaded the page.

Agenda

Add agenda topics here! Chair: John Delaney

Scribe volunteer: Maud Nalpas

Attendees — please sign yourself in!

  • Brian May (dstillery)
  • Andrew Pascoe (NextRoll)
  • Moshe Lehrer (Neustar)
  • Brendan Riordan-Butterworth (eyeo GmbH)
  • Swati Lal (Yahoo)
  • Abdellah Lamrani Alaoui (Scibids)
  • Alex Turner (Google Chrome)
  • Maud Nalpas (Google Chrome)
  • Chris Evans (NextRoll)
  • Lorenzo Hernandez (NextRoll)
  • Taejin In (dstillery)
  • Badih Ghazi (Google Research)
  • Phil Lee (Google Chrome)
  • Erik Anderson (Microsoft)
  • Betul Durak (Microsoft)
  • Aloïs Bissuel (Criteo)
  • Mariana Raykova (Google)
  • Bill Landers (Xandr)
  • Angelina Eng (IAB & IAB Tech Lab)
  • Daniel Rojas (Google Chrome)
  • Aleksei Gorbushin (Walmart)
  • Hélène Maestripieri (Google Chrome)
  • Martin Pal (Google)
  • Aditya Desai (Google)

Notes

Issue 212: Consider the case of the same priority in event and navigation

  • JD John Delaney: Prio of event and navigation sources. Currently there’s a prioritization mechanism where you can pick which impression will be attributed to a conversion; e.g. you can prio clicks over views. What should be the default? Right now by default the latest impression is prioritized. This issue suggests changing the default, by prioritizing clicks over views.

  • AE Angelina Eng Most advertisers are used to the last event Using the example of DV360: last-event reporting and assisted/unassisted reporting, that allows for advertisers to see what sites have contributed to that conversion

  • JD Not familiar with that use case, atm we choose to not generate multiple reports. You could have a last click reporting origin and an assisted one; but you can only register one impression for a click

  • AE: Assisted and unassisted reporting help change some decisions for shifting budgets; having different views help. Most people have been trained to use reports this way (or event adding own model).

  • JD: [Action item] Let’s file an issue for assisted and unassisted cases. The case described provides value. Back to the basic use case: I think it’s reasonable to prio clicks over views. Any concern?

  • BM Brian May: ? connectivity issues. Click, because it is deterministic, is the better default.

  • AE: Are they specs for specific latency? How long can a conversion be attributed? Is there some flexibility?

  • JD: Right now, the latency is determined at impression time. That’s modifiable. But you can’t edit that at conversion time.

Data deletion numbers https://groups.google.com/a/chromium.org/g/attribution-reporting-api-dev/c/5Ppe0cL-l1Y/m/kPATnUbwCAAJ

  • JD: Chrome published numbers around browser clearing attribution data. With cookie-based measurement you get reports at conversion time since joining is done on the server. If the user deletes data, conversions are not lost. With AR reports are kept around for some time, so if the user clears browsing data after conversion, reports won’t be sent. That’s the case for 6.5% of reports right now. Additionally, 16% of sources (converted OR not) are lost right now. More context: issue 201. Y!Japan talks about report loss in comparison with cookie-based measurement. Another reason for losses was network losses.

  • AE: Will there be an analysis to tell us what % of reports are lost? I.e. How many people clear their cookies/data?

  • JD: Think that’s possible but unsure. Cookie deletion isn’t an exact mapping of that deletion here.

  • AL (Abdellah Lamrani): Do you have a target for this? I.e. max 5% of deleted attributions? E.g. could you send reports when the user wants to delete their data?

  • JD: Not a ton of flexibility here. Sending reports when the user wants to delete their data is tough since users intended to delete that data. Another option: lowering the delay in the API => smaller windows. Send more reports before deletion typically happens.

  • AL: OK, so we’d need to find the right tradeoff... Shorter reporting windows sounds like an acceptable tradeoff.

  • JD: 5% loss is quite a bit for that industry, so we’re keeping our eyes on this.

  • Lorenzo Hernandez (LH): Any plans to re-enable the OT?

  • JD: OT expires Sept 14th but currently there’s a request to extend the OT to Chrome 94: https://groups.google.com/a/chromium.org/g/blink-dev/c/DmsUL3KHqMk/m/J2v3I_aEBAAJ. Join the mailing list for updates on origin trials: https://groups.google.com/a/chromium.org/g/blink-dev

  • AE: Creating a task force. IAB members didn’t understand the OT, nor how to contribute. Meeting kickoff on 30th with IAB members to educate them. There’s a lot of education to do.

Presentation of the advantages and drawbacks of each conversion measurement API for campaign optimization (asking for feedback)

Slides: Campaign Performance and Conversion measurement APIs #170

Presenter: Abdellah Lamrani, data scientist at Scibids

  • JD: Questions: Slide 3: you talked about 5% of noise makes it hard to measure diff types of conversion. You can recover in aggregate, IIUC this isn’t enough, you need the conversion type of the event level?

  • AL: yes, we need event-level precision. For post views, the additional restrictions (one reporting window, one bit): if there were an earlier reporting window, does that help? Or is the event-level API slower than expected?

  • AL: Short window would help… but daily data in general is really important, this should be the target. [Action item] File an issue to talk about the restrictions on views.

  • AE: You have a point; if there was a delay for 2 days and advertisers spending a lot of money. Once advertisers run a campaign, they look at it multiple times a day to troubleshoot any issues, including on the first days. Almost real-time data on a daily basis to figure out your pacing and optimization is important.

  • JD: What would you need the event-level APi to do that can’t be done with aggregate API? event-level is more for ML optimization. Can you help us understand where the gaps are?

  • AE: Small businesses has little budget and spreading dollars to try to learn quickly. Large advertisers spend lots of money and concerned about inefficiencies if data is not available on the same day. Businesses need to know what’s performing, especially during the first couple of days.

Issue 96

#96

  • JD: We’re looking at resolving this issue. Right now the reports go to the reporting origin; but publishers may want their own notion of performance and measure perf themselves. A duplicate report sent to publishers would be useful; however the trigger data will be meaningless for them, but publishers can recover conversion counts. This is still valuable. And can become even more so in case publisher and reporting origin agree on an encoding. Should publishers receive a report by default? The issue with this is bandwidth. So the publisher may need to opt-in. E.g. at the anchor level (specific ad) or at the page level. The ladder is better for the browser.

  • AE: From the agency side, and we did not provide Attribution Reporting for all publishers. Most of the time, advertisers are very sensitive about what reports are sent and to whom. Not all publishers are equal in their eyes.

  • JD: That was discussed, on both PCM and AR GH. Local maxima: there needs to be an agreement.

  • AE: Right now: this is set via Reporting profiles: certain vendors will get impressions and clicks, others would get specific campaigns, etc. Publisher has an aggregation tool that pulls from these APIs and has some dashboard. So they can compare their numbers with advertisers’.

  • JD: Is this being mediated by the ad server?

  • AE: Yes.

  • JD: So there’s a way to configure this today. [Action item] Going forward, let’s propose different options to configure this.

  • AE: AdOps team was the one that set up all the campaigns and platforms; so if there’s a way for the ad servers to feed that info into the browser, that’s a way. A better way than finding someone on the advertiser’s side to set it up.

  • JD: Yes, we’re trying to make config delegatable to third-parties. But there’s some friction here -- you want to be in control of whether or not to receive reports.