Google, the end of last-click measurement, and the future of attribution
Is last-click measurement dead?
That would appear to be the logical conclusion if you read Google VP Vidhya Srinivasan’s blog post last week. She’s the general manager of buying, analytics, and measurement for Google Ads, and her blog post was a massive shot across the bow of last-click measurement.
Of course, mobile attribution continues to be last-click. Apple’s SKAdNetwork is last click. Facebook is last click. App installs driven by Google are still … last click. With a few small exceptions, the entire mobile measurement industry is last click.
(None of which guarantees that last click is necessarily a great, amazing, foolproof model for all time and all purposes, of course.)
But despite the fact that last-click attribution is basically the default today in the wider marketing industry and to an even greater extent pretty much the only methodology that the average user acquisition specialist uses, the mobile user acquisition industry hasn’t really responded to Google’s post.
So let’s do that.
Google’s last-click argument: privacy, effectiveness, data
Privacy requires change, Srinivasan says:
In the face of a changing privacy landscape, marketers need new measurement approaches that meet their objectives and put users first …
Last click isn’t effective, according to her post:
As the industry continues to evolve, last-click attribution will increasingly fall short of advertisers’ needs …
The answer is data-driven attribution:
Advertisers around the world have seen better results by switching to data-driven attribution.
None of the above is particularly controversial. Despite the fact that Google has oddly contrasted last-click with data-driven attribution (is last click somehow not data-driven?) there’s probably not a lot of argument to the contentions that privacy requires changes to marketing measurement, that last-click attribution has challenges, and that there are potentially better models.
The new attribution model will become “the default attribution model for all new conversion actions in Google Ads,” Srinivasan says.
The question is: which model? What is it, really?
Google’s term “data-driven attribution” is odd: all attribution is data-driven. It’s not one of the recognized attribution models, nor is it clearly an incrementality-driven solution:
- First click attribution
- Last click attribution
- Multi-touch attribution
- Linear attribution
- Time-decay attribution
While not 100% obvious, Google’s new “data-driven attribution” appears to be a sort of mix of multi-touch attribution and modeled attribution. More touches in more places, with some AI/machine learning thrown in to make sense of it all when you don’t have cookies or IDFAs or AAIDs.
“It’s really blurry … wasn’t deterministic ‘data driven’ as well?” asks growth consultant Thomas Petit. “Modelled attribution would be more descriptive. It’s basically extrapolated attribution [with] more inputs.”
And clearly, given what Google itself is doing in mobile user acquisition measurement, this is much more about brand ads driving web visits, retail purchases, or other types of activity. Right now it’s not really about mobile app install measurement.
“For mobile first businesses (or mobile and web), the claim of death of last click attribution is premature,” says Paul Bowen, general manager of AlgoLift. “SKAN and MMP attribution are both last touch and deterministic models. Any other attribution model (e.g. MMM or data-driven attribution) are probabilistic and should be built on top of these attribution methodologies given that they are deterministic.”
Privacy, data, and measurement
It’s also not hard to argue that multi-touch attribution can easily be much more privacy-invasive than last click.
Last click is literally looking at one point in time, and a deliberate action taken by a person to access a resource. Multi-touch attribution by virtue of being multi-touch needs to look at more activity, from direct action like clicks to indirect activity such as ad views. In fact if you were going to achieve full marketing measurement of every single action or view on every iota of marketing effort, you’d require full awareness of what people are doing.
Talk about an invasion of privacy …
Of course, this is not what Google is suggesting. And modeling and incrementality are ways of reducing that privacy overhead.
It’s interesting that Apple’s answer to mobile attribution is SKAdNetwork. That’s privacy-safe by hiding individual action in aggregations of activity. It is also last-click, though via “loser postbacks” introduced in iOS 14.6, Apple is still providing context around a more complex user or customer journey.
From my post back in May:
Most postbacks are “winner” postbacks: a notification that your ad network’s ad impression was successful. It generated a click and the click resulted in an app install. With iOS 14.6 and SKAdNetwork 3.0, however, we got a little gift from Apple: postbacks to up to five other ad networks whose ads were seen but not clicked on, or, at least were not responsible for the app install.
So modeling is not the only possible solution when you need to solve for both privacy and marketing measurement.
And, of course, it’s not exactly simple, either.
“When I entered the market five years ago, it was more or less clear how to attribute, how to measure events,” Nakusi Games user acquisition lead Vladimir Ilchenko told me recently. “Everything should become better and better and easier and easier for marketers … but nowadays I understand that it’s vice versa.”
That’s one challenge. Another is that a solution needs to be broader than just one company or ad partner, even one as large as Google.
A solution for brands needs to be comprehensive. Not of all customer or user activity — which breaks privacy protocols — but of all channels. In other words, not just what happens in the Google ecosystem, not just what happens on Apple devices, and not just what happens in the Facebook ecosystem.
The attribution solution we need
Look: Google is on to something. We need more than last-click attribution for a number of different reasons, including both privacy and the fact that the journey which leads to action is much more complex than a single click.
Even if it is the last one before a purchase or important event.
And we need that, eventually, in mobile user acquisition as well as other areas of marketing measurement. Modeling is part of the answer here. But it needs to be based on privacy-safe datasets from multiple places, channels, funnel levels, and partners.
From Singular’s chief growth officer, Ron Konigsberg, on the future of mobile attribution:
That future involves building out varying views of reality and integrating them intelligently into a single source of truth. At Singular, we’re looking at marketing performance from known and aggregated spend data, from deterministic last-click measurement, from probabilistic aggregated results data, from first-party data, and from other sources. All of those have their unique perspective on what is actually happening as marketers market, whether putting dollars to work or investing in organic promotion. Each of them has value.
But then they also need to coalesce into a single source of truth to provide a simpler modeled view of reality.
That’s the attribution solution we need. It’s data-driven (thanks, Google), it’s privacy-safe (thanks, Apple), it’s comprehensive (thanks, all ad partners everywhere), and it’s going to be the best way to meld aggregate and granular and deterministic and probabilistic data together to understand the impact of marketing spend and the best options for future allocation.