Google shakes up ad metrics: last-click attribution axed for new AI model

A fundamental change to Google Ads is coming this week. Here’s what performance marketers need to know...

Google has announced this week that it will no longer use last-click attribution as the default model in its buy-side ad network Google Ads.  

The move means that the default attribution method for any conversion touchpoint will be handled by what Google calls ‘data-driven attribution’; its algorithmic solution that assigns credit to different impressions over time.

These conversion touch points could include:

  • A new product purchase page

  • An app install campaign

  • A display ad landing page  

Current conversion actions with last-click measurement will continue to attribute based on the final ad that drove a conversion. 

No more static rules to attribution

The new data-driven attribution solution is a live data model, so how the algorithm assigns credit will be different for every brand or campaign. Unlike the static rules of ‘last click’, the attribution scoring might change based on which sites or apps are contributing to conversions, or how consumer patterns change across browsers, apps and devices. 

What the advertisers say  

Lara Harter, Head of Online Marketing, DocMorris, says: "Data-driven attribution allows us to assign the right credit to every touchpoint. With automated bidding and data-driven attribution, we've seen an 18% reduction in cost of sales over last-click.

Marco Carola, Head of Online Acquisition, Crédit Agricole Italia, says: “Since we moved our search and display campaigns in Google Ads to data-driven attribution, we’ve seen an 8% increase in overall incremental conversions with an 8% lower cost per lead.

Writing in an official blog post, Vidhya Srinivasan, VP/GM Buying, Analytics and Measurement at Google Ads,  said the company hadn’t made data-driven attribution the default until now because in many situations it wouldn’t meet volume thresholds, smaller advertisers see fewer sales, downloads or other conversions, and the product needs data coming in to work.

“Unlike other models, data-driven attribution gives you more accurate results by analysing all of the relevant data about the marketing moments that led up to a conversion,” Srinivasan said. “Data-driven attribution in Google Ads takes multiple signals into account, including the ad format and the time between an ad interaction and the conversion.

“We also use results from holdback experiments to make our models more accurate and calibrate them to better reflect the true incremental value of your ads. And like all of our measurement solutions, we respect people's decisions about how their data should be used and have strict policies against covert techniques, like fingerprinting, that can compromise user privacy. 

Last-click measurement still available (but no longer default)

Despite the allure of AI-powered data, last-click remains popular as a metric, especially for small and medium-sized advertisers that don’t use outsourced measurement vendors or set aside testing budgets.

Advertisers still have the option to turn off data-driven attribution and choose one of Google’ five rules-based attribution methods:

  • Last-click

  • First-click

  • Linear (which credits every impression equally)

  • Time-decay (which credits by the duration between an impression and conversion) 

  • Position-based (40% credit each to the first and last impressions, and 20% spread over the rest)

More use = more conversions?

The real benefit to this change is scale. The more data feeding into AI, the better the decisions it makes on efficient conversions for ads.  The change to data-driven attribution by default consolidates more ad spend and in turn, more data­ into one channel.  Google now believes its modelling has now improved to the level where it can run data-driven attribution for any campaign type, large or small. 

“Because of how we’ve been improving and training our data-driven attribution models, we’ve eliminated [that] previously existing requirement,” Srinivasan added.

To improve further, Google is urging even more advertisers (and thus SMEs)  to use data-driven attribution because the quality of Google’s data modelling is tied to the quantity of impressions and conversions it sees.

Machine learning can compensate for gaps in data if advertisers can’t effectively track customers or conversions. This type of AI modelled data will be ever more important as third-party cookies are phased away over the next few years. 

“Shift from precision to prediction marketing”

René Plug, Chief Business Development Officer at 1plusX, says: "This move is further evidence of Google’s shift from precision to prediction marketing. It’s also an indication of how the industry will look post-third-party cookie – attribution will need to change and data modelling is really the only fair method if players want to continue to account for the full “conversion funnel of steps”. Moving forward, it will be important for media companies to prepare their first-party data and apply data modelling as a viable alternative to classic attribution mechanisms. Equally, the demand-side should embrace new prediction-based attribution models and partner with the sell-side, both Google and other key media partners.”

A holistic view to optimise future strategies 

Paula Gómez, Head of Data & Adtech at Making Science, says: “The Last Click model is used to analyze results, but it doesn't actually give a real picture in terms of what drove the results, making it difficult for businesses to optimize strategies properly. The Last Click model assigns the conversion to the last channel or campaign that intervened in the conversion process, but the reality is that these conversions are produced by the combination of several channels or campaigns. Because of this, it’s important to not only look at the last impact, but to distribute the results among all the impacts involved in the conversion path. Businesses can then use this holistic view to optimise future strategies. It’s likely that the main reason the data-driven model has not been used by default previously is because the algorithm wasn't developed enough to accurately attribute DDA with few data signals, but now it's more competent at that.”

“Education will be key”

Ollie Vaughan, Chief Media Officer, Tug, says: The downside of Google’s Last Click Attribution is that it provides an incredibly one-dimensional view of how your marketing is performing – ignoring channels that contributed to the final conversion earlier on in the path and focusing only on the channel that created the ‘last click'. But this is just one of the reasons why Google is phasing this model out.  With third-party cookies coming to an end, attribution will become harder, creating a need to evolve towards a more privacy-centric method. In theory, the vast number of ’signed in users’ Google possesses covers the privacy requirements, but for the modelled approach to work effectively, more data will be needed and the new approach will naturally provide this. This will strengthen Google’s position in the adtech landscape; not only for attribution, but also by potentially providing more data to effectively evolve to a new privacy centric version of other products like re-targeting.

“Savvy marketers, with sufficient budgets, will have stopped using ‘last click’ years ago, but marketers new to ‘data driven’ marketing may have the often complicated and difficult job of explaining the benefits of the new method to their CMOs or clients. Education will therefore be key – not only in terms of how the new model works, but for increasing people’s understanding of the fact that, in the coming years, attributing channel performance will no longer be based on precision or absolutes, but on models. KPIs and targets may need to be re-forecasted and, dependant on the data-driven results, media plans may be altered to allocate budget to the areas shown by the new model to be performing best.

"Increasing the height of the walled garden"

Marçal Serrate, Chief Technology Officer, Hybrid Theory, says: “With last-click attribution giving all the credit for a conversion to the ad clicked immediately before it, the model can restrict gaining a full understanding of the customer journey. However, last-click attribution is an industry standard that has been used for a long time, and people have always been aware of the potential problems with it. So it is notable that Google has decided to provide an alternative option now. What is so significant about this update is that the data-driven attribution model is only available to marketers using the Google Ads suite, while for others the demise of the third-party cookie makes it difficult to compete. As a result, this move could further increase the height of Google’s walled garden and tighten its grip on the open web. It will be interesting to see how this plays out.”