Media Buying Briefing: Stagwell’s chief AI officer explains why Palantir differentiates its data platform from competitors

Almost a year later, Kahan explained a bit about what Palantir can do — and is innovating in the privacy side of the data business — in an interview with Digiday. He was in New York to take part in the UN General Assembly’s health-related sessions as a board member of the Novartis Foundation. 

The following conversation has been edited for space and clarity. 

What is Stagwell building with Palantir? It seems different from the other holding companies. 

That’s really where I spend most of my time, between the two CEOs, so that’s a good place to start. Why Palantir? Because I believe personally in a platform-agnostic, AI-agnostic and world-class set of capabilities… It made perfect sense to me because I don’t want to be tied to any one model, any one platform, any one capability. I want to deliver what’s the best capabilities for our customers at any point in time. 

If you really think about Palantir, it’s a set of platform capabilities that is agnostic to the LLM or the cloud provider, and so I can run Stagwell’s targeting machine on any cloud I want, and I can take advantage of the latest capabilities in the world.

It’s not intended to be run by us. It’s actually intended eventually to be run by each of our customers. Any brand can hook it up directly to their supply chain system. It is not a CRM system like Epsilon is, it’s actually a data analytics platform that can pull data in from any disparate sources of data, structured and unstructured, make sense of it with an LLM model, and then be able to create impact in a marketing world… It knows what’s happening in the financials, and it puts the right ad out at the right time, not just based on what the customer wants, but what the business can deliver to make that customer happy.

Palantir has a workload that has been tried and tested in very difficult scenarios — the Ukraine war that’s occurring, all the way down to very small scenarios. And they can swap in and out any of these models, and they can run across platforms at scale. A real scenario right now: we’re building sovereign models with our customers on open weight models that come from Nvidia. Why would we do that? Because it does several things. It makes it it makes it less expensive for our customers. It protects their data, and it uses our data in a unique way for our customers’ directions.

With Palantir I can test any model against another model, real time in a real world scenario, and put the best model at the right cost factor for our customers directly in their hands. They’d never be able to afford to do that on their own. I can scale that across all of our customers.

Does the battle between AI models, such as Amazon cutting off Meta’s Muse, affect that ability?

These big companies are all going to push their models. The world I came from — and the world I still am in pretty heavily with Microsoft — is a world of of openness, where you can use the best models to solve the best scenarios. Don’t compete with your customers. Stagwell is not competing with OpenAI. We are not competing with Facebook. We’re not competing with Google. In fact, they are some of our biggest and brightest customers within the space. We will use their models when it’s appropriate for our customers — not just them, but others in those scenarios. 

That’s the beauty of the platform we’ve built. We are not proprietary in that sense. Some of our competitors like to be proprietary — we don’t do that. But we deliver incredible insights to our customers with the most privacy-compliant capabilities in the world using differential privacy, which is a set of capabilities that I donated from Microsoft to Harvard [specifically, to the university’s Institute of Quantitative Social Science] when I was at Microsoft. Why? Because I wanted to open them up to the real world, so that others could use them. And then we hired Harvard actually to implement it within the Palantir system directly.

What’s differential privacy? 

Most privacy in the world is done by stripping data out of the data, and so there are two problems with doing that. You lose meaning in the data, and then you can take that data and combine it with other data, and all of a sudden you can break the privacy. Differential privacy does something completely different. It actually adds noise to the data. So instead of having, say, five African Americans on a city block, maybe it has eight African Americans, and it does it at every level of the data set. The U.S. Supreme Court ruled it is the only version of data technology that they would allow to be used in the last census.

We treat privacy quite different than our competitors. We believe it’s a feature. Our customers should respect that we are delivering our data with the highest level of privacy possible in the world, and that’s where we operate.

How are you applying it? 

Anybody can use differential privacy, but the difference is that we will publish our algorithms, and Harvard actually certifies them, and they give us the code that we use in order to deliver the data. For example, in the United States, our data set alone is over 1 billion email addresses, 260 million individuals, 160 million households. And we have like 60 years of Harris Poll data. We don’t just understand what people do, but how they think. Because we have polling and survey level data that’s embedded in Stagwell. We think we take a different twist than the Epsilons of the world, which are really mostly behavioral data.

Color by numbers

As the midterm elections approach, there’s no doubt the Hispanic or Latino voter is key to any electoral victory. So Nielsen’s latest estimate of Hispanic audiences seems particularly relevant as candidates decide where to run their ads. According to Nielsen’s 2026 Hispanic Diverse Intelligence series, Hispanic audiences watched more than 2 trillion minutes of television in the first half of 2026. Here are some details: 

• Of that total, 1.2 trillion were streamed. Hispanic viewers delivered 19.1% of all U.S. streaming time— nearly one in every five streaming minutes — alongside 839 billion minutes of linear TV viewing.
• On linear TV, 42% of Hispanic viewing time went to Spanish-language content, despite English programming outnumbering Spanish content 9-1. 

Takeoff & landing

• Horizon Media’s specialist agency Horizon Commerce expanded its strategic partnership with commerce media firm Pacvue to boost its clients ability to connect retail media planning, activation and measurement.
• Digital agency Granite acquired B2B design agency Right Hat, which designs and builds brands for professional services firms and other B2B enterprises. The acquisition extends Granite’s U.S. reach into Chicago and Washington DC from its New York base. Terms were not disclosed.
• You know AI is getting serious in marketing when firms start acquiring agencies. AI growth firm Power Digital acquired Amazon marketplace agency Pirawna and TikTok Shop agency Uptickk to grow its roster of Amazon and TikTok shopping expertise into its platform. Terms were not disclosed.
• Account moves: WPP Media landed DoorDash’s global media remit, expanding on regional work the holding company network has done for the food delivery brand … Omnicom’s Initiative in Canada landed Moosehead Breweries’ media business … However, in Australia Initiative lost the media business of retailer Officeworks to corporate sibling PHD … Known landed Rivian‘s media business … Mediaplus won the North American media for entertainment restaurant brand Dave & Busters, alongside Gut, which won the creative business … Independent Moore won brand agency duties for the Alzheimer’s Association, overseeing strategy, creative and media.
• Personnel moves: Publicis’ Epsilon made three key hires, tapping Michael Chock to be chief growth officer, Ant McDonagh to be general manager, Epsilon Digital, Kathleen Dundas to be chief solutions officer … Monks tapped Nirvik Singh to be CEO of its EMEA Marketing Services business, which will lead him to step down from S4 Capital’s board as an independent non-executive director. 

Direct quote

“We are drowning in measurement… Instead of just measuring numbers, we need to measure gaps. If you’re measuring the gap, you can start actually allocating budgets to the pieces that really matter, instead of just chasing semi-generic industry metrics, which are useful but not actually strategic.” 

— Isabel Perry, global evp of strategy at Dept, speaking at Digiday Media’s AI Marketing Strategies event in New York on Sept. 24. 

Speed reading

• Speaking of the event, Digiday’s editors put together a look at the impact the AI bubble is having on standards across the marketing discipline, as discussed in the event’s town hall. 
• I wrote about 24 Seven consolidating the pitching efforts of the three agencies it purchased last year: SketchDeck, Markacy and Futureman. 
• Alyssa Mercante covered all the news coming out of the Made On YouTube event last week, which showcased AI agents, live “gamification,” and new testing tools.