“We’ve solved attribution using machine learning – all you need to do is click a button and using our proprietary algorithms, we will automatically allocate your media budget at the point of maximum efficiency.”
This is effectively Google’s pitch for data-driven attribution.
Advertising being a somewhat murky business, however, this has been met with some skepticism. But in view, almost all brands spending on Adwords should at the very least review the analysis.
What’s the history?
Data-driven attribution is not a new offering from Google, in fact is has been kicking around since 2013 in one form or another, yet you can sense a recently renewed confidence in the product from the Googlers’ closest to it.
It is now available to everyone in Adwords as a reporting feature (i.e. just look at the data) and as a model to optimise to (slightly scarier).
What is wrong with the traditional models?
Last Click – this model is incredibly limited. It works in some very specific cases like flash sales but otherwise credits the ‘finisher’ 100%. It can be manipulated easily through unscrupulous retargeting activity. It’s essentially like giving Carlos Alberto all the credit for the greatest team goal of all time: http://www.telegraph.co.uk/football/2016/10/25/the-35-separate-moments-that-make-carlos-albertos-goal-for-brazi/, while in fact, there were 35 component parts – take 1-2 of them away and there’s no goal.
First Click – this model reverses all of the above – while the conversion/goal may not have happened without the first touch-point/tackle, it’s pretty unlikely this first touch solely led to the goal.
Position based (i.e. first and last upweighted) – this model ignores the all important nudges along the way.
Linear – This model is my current favourite as everyone involved gets equal credit and you can use other means to determine the channel’s influence, not just the single variable of time.
Data-Driven, in theory, should be the best.
Is Data-Driven better?
With data-driven attribution, the algorithm analyses, predicts and acts in real time based on prior data-points and then learns and refines as it goes. The AI examines conversion paths and looks for paths where conversion is higher.
If it sees a common touch point in these higher converting paths it automatically attributes a higher share of the conversion. This is how it prioritises and bids.
It seems like a no-brainer, right?
The problem?
It’s Google’s algorithm and people don’t trust media owners to tell them where to invest their budget. It’s a clear moral hazard.
But sometimes it gets to the point where the performance uplift is so great that it’s a competitive disadvantage not to use it. While we’re not there yet, it could very well happen a lot sooner than you think.
