Upcoming: Gaming & Advertising Guide, Learn More>

Google’s Latest Privacy Sandbox Proposal Seeks to Fix TURTLEDOVE’s Problems

Tim Cross  10 November, 2020

Google this morning introduced a new proposal to its Privacy Sandbox: the Augury API.

Augury is designed two complement TURTLEDOVE, a framework designed to enable cookie-free interest-based targeting on Google’s Chrome browser. Google engineer Ardian Poernomo, who posted the proposal to Github, says Augury will tackle issues around brand safety enforcement, data leakage from ad tech companies, publishers and advertisers, and auction integrity which face TURTLEDOVE.

The release is significant in that it should help clear some of the technical hurdles facing TURTLEDOVE. But it also suggests that Google is putting all its eggs in the TURTLEDOVE basket. TURTLEDOVE is still just a proposal, rather than a working technical specification, and it’s possible that an alternative solution could emerge within the sandbox. But it seems that Google is prioritising fixing potential issues for TURTLEDOVE, rather than working on entirely new frameworks to handle interest-based targeting.

Easing Pressure on the Browser

TURTLEDOVE was first proposed back in February, as a basic framework for how advertisers might be able to target users based on their interests, without actually being able to see data on that user’s browsing activity.

A full explanation for how TURTLEDOVE might work can be found here. But a key component is that data on users’ browsing activity is stored on the browser itself, rather than in a remote database. Users are placed into interest groups, like ‘cars’, ‘travel’, or ‘football’, based on their browsing activity. And in the TURTLEDOVE frameworks, ad auctions are also run within the browser.

But Poernomo says containing all this activity within the browser poses four significant problems:

  1. TURTLEDOVE uses machine learning models to inform bidding within auctions, but doing this within browsers would be complicated and expensive.
  2. Enabling brand safety restrictions, both for advertisers and for publishers, is difficult and expensive to do within the browser.
  3. Publishers’, advertisers’, ad tech providers’ confidential information and intellectual property could be leaked, since sensitive information is shared with the browser.
  4. It is important for ad tech providers to be assured of the credibility of submitted bids and auction clearing prices, but it’s difficult to give these assurances when the auction is performed on the browser.

Augury aims to fix these issues by tweaking how interest-based bids are made within the TURTLEDOVE framework.

In the basic TURTLEDOVE framework, users are placed in interest groups based on their browsing behaviour. So if a user visits a bike ecommerce store, they might be placed in the ‘bikes’ interest group. And at the same time, the bike ecommerce store would send an advert and some bidding instructions to the user’s browser, where they are stored.

When that same user visits another site which runs ads, the browser sends out a bid request for contextually relevant ads. It receives bids back through the supply-side platform,  and then weighs them up against any interest-based ads stored on the browser.

Going back to our example, after visiting the bike ecommerce store, the user might then read an article about football on a sports website. So the browser would ask SSPs it works with for contextually relevant bids, and might get an ad and a bid for a sports streaming service. It would then decide how much the bike ecommerce store would be willing to bid for the impression, based on the contextual information it has available and the bidding instructions it received from the bike store. Whichever ad has the higher bid would be shown.

With the Augury API, these interest-based bids are moved outside the browser. Instead, demand-side platforms still submit bids for contextually relevant ads. But they also submit bids for other ads contingent on whether the user belongs to a certain interest group (which are called ‘augury candidates’).

Returning to our example from before, the users visits The Guardian after the bike shop. The browser still sends out bid requests, and receives the bid from the sports streaming service. But this time, it might also receive a bid which says “If this user belongs to the ‘bike’ interest group, bid £X for the impression.” The browser might also get a bid which says “If this user belongs to the ‘car’ interest group, bid £Y for the impression”.

The browser then chooses the winner. The browser knows the user doesn’t belong to the ‘car’ interest group, so it discounts this bid. It knows the user does belong to the ‘bike’ interest group, so it weighs up the bike ad’s bid against the streaming service ad’s bid, and the ad with the winning bid is displayed.

Latency Issues to be Ironed Out

The final auction is still run on the browser. But the browser no longer has to process bids for interest-based ads, as these are handled remotely.

Moving this function outside of the browser fixes the four problems described above according to Poernomo. Bidding, brand safety controls, and assurances around bid credibility can be handled within existing ad tech infrastructure, in much the same way they are currently. And since less data is shared with the browser, the data leakage problem described above is eliminated.

But there are still issues to be ironed out.

The main problem is that when DSPs submit augury candidates, they know very little about the user. This means they could belong to any possible interest group. But Poernomo says it would be infeasible for DSPs to send bids for every possible interest group, since there could be millions of different interest groups.

To help solve this problem, Augury proposes that DSPs would predict which interest groups a user is likely to belong to, based on the contextual information they receive in the bid request.

But even so, DSPs would send a significant volume of bid requests through SSPs, which would be passed on to the browser. And Poernomo says there would be an increase in latency as a result.

Nonetheless, Poernomo says Augury should ease adoption of the TURTLEDOVE standard. And if TURTLEDOVE really is the future of programmatic advertising, any developments which ease its adoption should be welcome news to the industry.

Go to Top