Google’s data-driven attribution solution does not solve the effectiveness puzzle
Claiming that the new emphasis on data-driven attribution is a deliberate privacy-forward action is a nice spin
In the latest of its string of data developments, Google has just announced it is all but ditching last-click attribution
This is an attempt – Google says – to provide more accurate and privacy-centric measurement, by moving away from relying on the last user interaction and towards machine learning to look at the conversion pipeline.
According to the company blog, the last-click attribution approach falls short of advertisers’ needs and new privacy standards.
Instead, the latest approach using machine-learning to better understand which mix of marketing touchpoints in the user journey led to conversion, removing the need for minimum data requirements and adding support for a broader range of conversion types, including offline conversions.
It will become the default attribution method for Google Ads – and is already available for Search, Shopping, Display and YouTube ads – using an algorithm to pull all relevant data across the entire user journey, in an attempt to navigate measurement in a post-cookie world.
So how should we be reacting to this latest development?
Google’s hand was forced over privacy
At first glance, this move appears to conveniently hide uncertainty. Google doesn’t address the question: how can it suddenly drop minimum data requirements without accepting that there is now far more uncertainty in digital attribution?
Despite being lauded as a privacy-centric move, we can just as easily take the view that this change isn’t being introduced to protect privacy; instead it’s being forced because moves to protect privacy have already been made.
Digital attribution has become inadequate (on its own) and a switch to the data-driven attribution model by default is a more practical way to adapt than continuing to focus on last-click.
Claiming that the new emphasis on data-driven attribution is a deliberate privacy-forward action is a nice spin.
Digital attribution alone is not enough
There are two real key points in the announcement. The first is the recognition that digital attribution alone is inadequate for judging marketing effectiveness. It’s encouraging to see in the announcement that Google is combining digital attribution with holdback experiments.
Advertising effectiveness needs to be judged with multiple methodologies in order to enable meaningful innovation in the marketing mix and continual growth in performance.
These methodologies will include:
Digital attribution – Despite its limitations, ad platform attribution data is integral to driving automated bidding and optimisation and plays a role in the effectiveness mix. It underpins day-to-day channel management, and is very readily available for continuous reporting.
Controlled experiments – Automation has made manual testing less prominent in campaign optimisation, but a continuous programme of testing – including holdback experiments – is now vital for calibrating and complementing digital attribution.
Causal effect analysis – Modelling can be used to prove the effect on a KPI of the introduction of a new channel, campaign or tactic. Studies can be run quickly and repeated regularly, fitting into a test & learn framework alongside controlled experimentation more broadly.
Marketing mix modelling – MMM attributes performance to individual components of the marketing mix, including competitor activity and other market factors, and it can be informed by digital attribution measures and the results of experimentation These are big studies and sophisticated models that inform strategy and media planning.
Google is reacting to industry’s course correction
The second real key point in the announcement is about the broadening of conversion types available for data-driven attribution. Alongside the need for a varied effectiveness ecosystem, the activation of first-party data is a key strategy in today’s competitive digital advertising.
Being able to contribute first-party data to data-driven attribution makes great sense, although again we’re seeing a reaction to our industry’s privacy-led course correction. Being able to incorporate offline conversions isn’t itself a privacy-forward action.
Hidden between the lines of Google’s announcement, we can see they’re aware of needing more than just a new emphasis on a data-driven model for digital attribution.
They’re spinning it as a privacy-forward action, but – especially with the removal of minimum data requirements – we can see it as an acceptance of where digital advertising is today.
Incorporating the results of holdback experiments, and broadening conversion types, are signs of this acceptance too, and these changes reflect my own recommendations for activating first-party data, and taking a more nuanced approach to judging digital advertising effectiveness.
Kevin Joyner is director of data solutions at Croud