Google’s announcement on phasing out the support for third-party cookies within the next two years has led to a lot of speculation around its impact on the ad tech ecosystem. An obvious question to raise is how it will impact the ability to perform effect and attribution measurement at scale?

The Why

It has always been the holy grail of advertising to measure and document the efficiency of online advertising. Naturally, advertisers want to understand how to spend advertising dollars for maximum effect. And naturally, vendors will clamour to persuade advertisers that it’s only their inventory or data that leads directly to the said grail.

However, most digital attribution methods have either been relying solely on third-party cookies and/or last-click attribution. In our opinion, this has resulted in an incomplete picture of which parts of your advertising truly drive your business outcomes.

There is a huge need for a new paradigm that will ensure both privacy and best-in-class measurement of real effect. In this article, AudienceProject outlines how we approach this complex and very important issue. It may sound weird, but we actually don’t need to know which individuals have been exposed to which ads. Let’s explain.

The Problem

Relying on third-party cookies has (almost) since its invention been a deeply flawed method. By “tagging” users exposed to a specific advertisement with a third-party cookie (cause), it was (at least in theory) possible to measure which of the users subsequently purchasing the product (effect) could be attributed to the ads in question.

It seems like the logical method to use – but in reality, less so. Third-party cookies have been notoriously bad at measuring long term effects across devices and platforms for quite some time due to the ongoing evolution of the internet, which has presented a number of challenges – the ‘highlights’ are summarised below:

Firstly, advertisers have a long tail of choice. More and more digital media channels have, over time, become available for advertisers:

  • Programmatic display advertising
  • Search / AdWords
  • Social media (Facebook, Instagram, Snapchat, Twitter etc.)
  • Direct IO advertisement and sponsorships
  • Digital video (YouTube etc.)
  • Etc.

Tracking exposure to ads cross-channel in order to attribute effect by third-party cookies is increasingly challenging due to different platforms using different sets of identifiers and due to the rise of the walled gardens (yes, we are looking at you, US tech giants!).

Secondly, there has been an explosion of different connected devices which consumers use, with the same person often owning multiple devices:

  • Smartphone web-browser
  • Smartphone in-app browser
  • Smartphone native app
  • Laptop web-browser
  • Tablet web-browser
  • Tablet in-app browser
  • Tablet native app
  • Smart TV
  • Gaming console
  • IoT
  • Etc.

Each of the above operates with a different set of identifiers (if an identifier is even present) and again, we see the rise of walled gardens and more and more devices operating with proprietary or no ID systems at all.

Thirdly, there is the rise of regulation like ITP, GDPR and ePrivacy consent, ad blockers and many other similar problems hampering traditional cookie-based attribution at scale.

Fourthly, even if we still ignore all of the above, the average ad tech third-party cookie has a half-life of twelve days. That’s it. Twelve days after a cookie’s inception, 50% of the cookies from the same cohort are dead.

Fifthly, don’t even get me started about why last-click attribution is even worse. It rarely rewards anyone except those who manage to (often smartly) place themselves as gatekeepers of the last click, which is why we will ignore that option entirely.

The ad tech industry at large has unfortunately been trying to solve the challenges above with even more tracking technology like fingerprinting and by adding even more cookie-based identifier systems. Resources have increasingly been invested in creating “new” technologies that would allow the industry to maintain the “old” ability to track individuals across as many online interactions as possible.

Privacy has not exactly been the first item on the agenda, which is also why the industry at large is experiencing a privacy backlash today.

In my opinion, the industry just kept doing so because it’s an easy sell. Everyone can relate to the story about tagging users and then counting your “beans” (the exposed) once they arrive at whatever conversion page you are interested in. The reality is, however, that the “counting third-party cookies” approach has been impossible for many years, as described above.

So, the question remains: how do we successfully tie cause and effect together, while respecting user privacy at the same time?

The Solution

Let’s examine the task at hand from a first principle perspective. To estimate the relation between cause and effect – to attribute effect to a campaign – we actually don’t need to know which individuals have been exposed to which ads. Whenever someone converts, be it through an online visit, download, signup or purchase, we just need to be able to describe the conversion probability within groups (exposed versus non-exposed). We don’t need detailed information about individuals to succeed.

This is the crux of the issue; we do not need to track individual behaviour at scale, tied to an identifier like the third-party cookie, to estimate the conversion probability within groups.

Rather, estimating with a high degree of accuracy, the conversion probability within groups (exposed versus non-exposed), is what we really need to be doing.

Instead of using cookies, we can use geography groups. Geography is a universal common denominator for almost all types of cause and effect. We can, with a high degree of accuracy, attribute both cause and effect to geography as long as we don’t try to work with hyper-localisation. Most ad exposures (cause) can, to a certain extent, be controlled on a city/county level. Likewise, almost all conversions (effect) can be tied to a city/county level geography.

Using geography, we can define what we refer to as exclusion zones. Exclusion zones are geographical areas that resemble the rest of the country in demographic composition and which have very little movement of residents outside of the zone on a daily basis. When executing a campaign, we carefully ensure that no ads are shown to residents within the exclusion zone. The exclusion zone(s) essentially becomes our control group when we want to measure the outcome of the campaign.

By comparing the relative uplift on our conversion metrics (effect) across the inclusion and exclusion zones, we can very accurately estimate the attribution effect of a particular campaign.

Figure A:
Figure A illustrates the principle. We can measure and quantify the uplift generated by the campaign in the inclusion zones by comparing it to the performance delivered from the exclusion zones.

There are many advantages to this model compared to the regular third-party cookies based models:

  1. The approach enables the long sought after ability to tie offline conversions from bricks and mortar shops to online campaigns. As long as it is known which geographical area a shop serves, sales numbers can be added to the model as easily as online conversions or orders.
  2. The model is extremely privacy safe. You don’t need to know which devices or actual individuals were exposed to content online. We operate with exposure and conversion probability within large geographical groups.
  3. The model delivers effective control towards seasonal effects. Our control groups will identify and accommodate naturally occurring upward or downward trends in sales due to holiday seasons, time of the month, etc. and ensure that such natural variations are not attributed to a campaign for better or worse.
  4. The model measures true uplift by not automatically assuming that correlation equals causation (which is kind of how the cookie-based last-touch attribution model works). Advertisers who excel in identifying and targeting existing customers only get attributed if they manage to increase sales/conversions within this group of customers. They cannot be attributed to existing sales due to the control group defining the base level of performance.

Building and maintaining such a methodological framework is no small task. It requires comprehensive analysis and statistical modelling to ensure proper zones are defined and respected at both campaign execution time as well as when measuring effects during and after the campaign. At the same time, integrations are needed on several ad tech platforms. It is definitely a lot harder than just dropping third-party cookies and conversion tags. We know because we have been building and maintaining such complex tools for almost a decade at AudienceProject. The upside is that even extremely complicated models can be highly optimised due to advances in technology in recent years.

I know from personal experience how challenging it is to run old-school call center driven exit polls on election nights since I spent many years as part of a team designing and executing such complicated projects before founding AudienceProject. The complexity is staggering. Today? Our systems in AudienceProject run thousands of campaigns on a daily basis, and each campaign incorporates a much higher degree of computational and methodological complexity than any exit poll I ever have taken part in. And they do it in real-time and at a low cost. Through automation and standardisation, great complexity and the usual cost tied to such complexity can be made very manageable.

We, as an industry, really have no excuse for not stepping up to the plate to deliver better (i.e. way more accurate) measurement while at the same time ensuring the privacy of users. We have the methodology today, and we have the technology available to deliver today. This is why we applaud Google’s recent move.

The future is today.

Real-world case

In a case study made in cooperation with the Danish full-service media agency Orchestra and an entertainment company, we show how Orchestra managed to measure the true effect of a large campaign by the entertainment company with AudienceProject’s cookie-less attribution modelling. This allowed Orchestra to document how many conversions could be attributed to the campaign and what the Return Of Advertising Spend (ROAS) was of the campaign. Learn more.

This article is part 2 of 3 in our series: Life beyond third-party cookies

About the series:

Google decision to phase out support for third-party cookies in Chrome has led to concerns on how the ad-tech ecosystem will be influenced. In this series of articles, AudienceProject addresses those concerns and explains why Google has made the right move.