Alex Steer, Global Chief Data Officer at Choreograph
Data sits at the heart of most major M&A deals happening in the agency world at the moment. Omnicom’s impending merger with Interpublic Group is motivated in part by the prospect of combining IPG-owned identity solution Acxiom with Omnicom’s digital commerce business Flywheel. Flywheel itself, Omnicom’s largest acquisition, provides access to a wealth of transaction data. French group Publicis Groupe, through its purchases of Epsilon and Lotame, has bought itself an ID base which it says covers 91 percent of adult internet users. And WPP’s most notable acquisition so far this year is data collaboration business InfoSum.
But one of these deals is not like the others. Acxiom, Epsilon, and Lotame all have their own banks of data, pinned to unique identifiers, which marketers can use for targeting and measurement. By acquiring these businesses, Publicis and IPG have essentially bought massive banks of data tied back to individuals, as well as the various data tools and capabilities each of those businesses owns.
InfoSum doesn’t run this sort of database. Instead, it enables multiple parties to effectively combine their own databases and take action based on shared insights (again, powering targeting and measurement), without data changing hands. Strong data capabilities, yes. Big bank of audience profiles, no.
But Alex Steer, global chief data officer at WPP’s data and technology business Choreograph, says WPP’s media arm GroupM is charting a different course from its competitors when it comes to data, one which doesn’t require ownership of an Epsilon-like database. And while competitors may claim that WPP has bought itself “a commodity” and “an empty shelf”, Steer says InfoSum is a key component of a data strategy which WPP believes will unlock a much wider range of data sets for its clients.
Keeping data where it is
There are really three key market dynamics informing GroupM’s data strategy, according to Steer. First, there is an ever-growing amount of data out there in the world which is valuable to marketers in deciding “where to go and how to grow”. Secondly, that data is increasingly fragmented, split more than ever across a range of different organisations. And finally, the organisations which control that data generally don’t want to share it, whether that’s due to privacy or competitive concerns.
“Our view is that the traditional approach that the industry has taken, which is about centralising data and focusing on certain types of data — particularly data that’s easy to join up based on individual identifiers like cookies or emails — is quite a limited approach,” said Steer.
There are therefore two key challenges for agencies to solve, according to Steer. And both of these, in GroupM’s eyes, require a shift from an ID-led approach to an AI-enabled approach.
The first problem is figuring out how to get more value out of data while moving it less. GroupM’s answer to this is ‘federated learning’, (a phrase you might be familiar with from Google’s Federated Learning of Cohorts). Federated learning is “an umbrella name for a set of AI techniques that allow us to extract the information from different data sets without combining or compromising the data sets themselves,” says Steer. So in essence, shifting to federated learning means leaning on tools which facilitate this process — like InfoSum, for example.
Translating across data sets
The second key challenge is getting value out of a wider range of data, rather than just whichever data sets can most easily be joined up to a specific identifier.
AI’s role in solving this issue is perhaps a little less obvious (since many in the industry are already familiar with clean rooms at least at a conceptual level). But Steer says the key thing to understand is that AI tools are good at understanding relationships between completely different types of data — something which is apparent whenever you use an AI tool like ChatGPT or Google Gemini.
“Over the last few years, we’ve all got used to the idea that AI is really good at translating between different things,” said Steer. “If you ask one of the multimodal AI models to show you a picture of a happy frog, it will show you a picture of a happy frog. The principle is that because AI models are trained to understand the world, they understand the relationships between different types of data points. You can give it words and get pictures, you can give it video and get code.”
“It’s the same sort of principle with data,” added Steer. “We reckon that only around 10 to 20 percent of data that’s useful for decisions that we make in marketing can be resolved back to an individual ID about an individual person. Most of the information that businesses own comes in different types and sits in different structures. You might have data that’s organised by store, or by postcode, or by keyword. AI is really good at translating between those things, and at learning and spotting patterns.”
So GroupM is training AI models on the relationships between different data points, and also different types of data points, creating a multimodal model for advertising-relevant data. “If you can do it for words about frogs and pictures of frogs, you can do it for data about stores, consumers, campaigns, brands — all those sorts of things,” said Steer.
A question of focus
IDs aren’t completely irrelevant in this world. Steer says WPP is training models on around 350 different data sources, and some of that data is organised around consumer IDs. But they’re just one part of a much broader spectrum of marketing-friendly data.
Not everyone agrees with this vision. Publicis Groupe CEO Arthur Sadoun, when asked directly about WPP’s acquisition of InfoSum on the company’s most recent earnings call, said that a clean room is “an empty shelf” if you don’t have an identity graph to feed into it. “If you don’t power a clean room with identity, you can’t find new sources of growth,” he said. “If you don’t add proprietary data that will bring to your clients something that they don’t have, they won’t find new sources of growth. They won’t be able to connect every individual to the entire media ecosystem.”
Maybe the big difference between the two groups’ strategies is a question of focus. WPP doesn’t completely discount the value of proprietary ID-linked data for an agency, and Publicis would likely acknowledge that there’s plenty of valuable marketing data out there in the world which can’t easily be tied back to an ID. Perhaps in an ideal world, they would pursue both at the same time — maintain a massive ID graph, and feed that into a much larger pool of data which is brought together through clean room technology and “translated” by AI.
But with limited resources, agencies have to place their bets. WPP believes that in the long run, its time and effort is better spent investing in technologies and training AI models which aren’t reliant on IDs, which it says will always be limited. “You will never build a database that’s big enough,” said Steer. “The world’s best data sits with a lot of the world’s best businesses, what’s really important is to be able to connect them. We’ve not just bought pipes and plumbing, we’ve invested in a platform and we’re investing in AI and data capabilities that allow us to make those connections, and to make them fast.”
Steer said he’s increasingly seeing this belief reflected among clients. “I think there’s a growing acknowledgement that just knowing who people are and where to find them is not enough in order to make high quality predictions about how to serve great creative, how to deliver great experiences with media, and how to drive effectiveness,” said Steer. “You’ve got to think bigger than that. And I think a lot of advertisers are getting very sophisticated and very ambitious about understanding what AI models can do in terms of spotting patterns across different data sets, and making predictions that don’t have to tie back to an email address or a cookie.”
An “accelerator of ambition”
The question remains (and was raised by Sadoun): what does WPP get from owning InfoSum, rather than simply working with it?
Steer said that for a start, InfoSum was an attractive acquisition target in its own right, regardless of GroupM’s own wider strategy. “It has a lot of direct client relationships, a lot of organisations use it and trust it to enable data collaboration with each other in ways that have nothing to do with media, and nothing to do with GroupM,” said Steer. “They’ve been a great partner over the years, and we intend to keep it operating under its own identity and with its own go-to-market, working directly with clients who want to be able to do data collaborations.”
But when it comes to GroupM’s own data strategy, Steer says owning InfoSum will help accelerate its roadmap. “InfoSum being part of WPP allows us to work much more closely with the incredible team of people there,” said Steer. “It allows us to have direct input into that roadmap, and to really use it as an accelerator of the ambition of what we’re trying to develop. Really simply, it lets us do things much, much faster, and we’ve already started to see that with clients.”
Follow VideoWeek on LinkedIn.
