Almost a year after Stagwell surprised the industry with a Palantir partnership, chief AI officer John Kahan is making the case for what the challenger holding company believes is a different kind of data platform: one that is not only model- and cloud-agnostic but also treats privacy as a feature that can be published and certified.
Why Stagwell chose Palantir
Kahan, a 40-year data and analytics veteran whose career spans IBM and Microsoft and who chaired Stagwell’s board of advisors until last year, was initially reluctant to take the job. CEO Mark Penn talked him into it. The arrangement puts Palantir at the center of Stagwell’s Marketing Cloud.
Kahan says Palantir offers “platform capabilities that are agnostic to the LLM or the cloud provider”, so Stagwell can run its targeting machine on any cloud and swap in the latest models. The platform is not meant to be run by Stagwell alone. It is designed so client teams can eventually operate it and connect it directly to supply-chain systems. Kahan describes it as an analytics layer rather than a CRM: it pulls structured and unstructured data from disparate sources, makes sense of it with an LLM, and can time advertising around both customer demand and what the business can actually deliver.
Differential privacy, with Harvard’s certification
The sharper differentiator is how the platform handles privacy. Instead of stripping fields out of a data set—which Kahan argues destroys meaning and still leaves data vulnerable to re-identification when matched with other sources—differential privacy adds calibrated noise. Kahan gives a simplified example: a block with five people from a group might appear as eight at every level of the data set. He says the U.S. Supreme Court allowed only this technology to be used in the last census.
Stagwell says it will publish its algorithms, and Harvard’s Institute of Quantitative Social Science—where Kahan donated differential privacy capabilities while at Microsoft—certifies them and provides the code. The intent is to make privacy protections auditable, not a black box.
- Model-agnostic and cloud-agnostic, so clients are not locked into one LLM or platform.
- Designed to be run by client teams and connected to supply-chain systems.
- Uses differential privacy with published, Harvard-certified algorithms.
- Layers 60 years of Harris Poll attitudinal data on top of behavioral data.
What’s at stake for marketing leaders
The U.S. data set behind the platform is substantial: Stagwell says it spans more than 1 billion email addresses, 260 million individuals and 160 million households. Kahan’s argument is that scale alone is not the wedge—most holding companies have scale. The combination of openness, privacy certification and attitudinal data is the bet.
The interview arrives as AI model providers are jostling for position, with Amazon reportedly cutting off Meta’s Muse from its cloud. Kahan’s stance is that Stagwell should not compete with OpenAI, Facebook or Google; those companies are also customers, and the platform can use their models when appropriate. That open posture is a strategic choice—and a useful prompt for marketing and data teams to ask whether their own tools are truly agnostic or quietly tied to a single vendor.
Elsewhere in the briefing, agency and account movement continued: WPP Media landed DoorDash’s global media remit, Initiative Canada won Moosehead Breweries, Known picked up Rivian, and Mediaplus won Dave & Buster’s North American media alongside Gut’s creative win.
Source: Digiday
