Havas' Mike Bregman on AI, black boxes and why the supply chain is under a microscope

Mike Bregman moves at the speed of data. Or at least, he’s building an ad buying operation at Havas Media Group designed to do so.

The chief activation officer joined us in The Current studio at Advertising Week New York this month — right before giving a talk on AI’s transformation of agencies — for a wide-ranging conversation on the state of programmatic. Bregman shared why pressure is mounting for the supply chain to deliver value, how marketers can harness omnichannel data points, his take on the “cookie-loss future” and an unpopular opinion about keeping AI in check.

You’re steeped in data. Is there a statistic that has meaningfully changed the way you approach your job?

When you look at trends, over 80% of CMOs have acknowledged that [AI] testing is happening right now, which is massive. That's the majority of the industry. They're all thinking about AI in one form or another. But when you look down the road, even 6 to 12 months, what percentage are going to keep those tests running? It's less than a third. And when you look at the start of 2025 and 2026, when money runs out on some of these tests, there's less than 10% that expect that the tests are going to scale in and of themselves. So budgets are tight, expenses are high. It's critically important that we get some of these activities to produce value for the clients.

Do you think we’ll be able to realize the dream of personalization sooner rather than later?

The industry is advancing at a rapid clip. It used to be A to B testing with two creatives — one that had a red background, one that had a blue. Now it's A to Z testing.

You can really create infinite iterations. You can push those into platforms much faster. I think what's interesting [is that] being a data analytics practitioner, I grew up in the industry where you had to run mathematical models and it was always a linear equation. A plus B plus C equals D. It’s not linear anymore. It’s really exponential growth.

It took data scientists a long time to piece all those different models together, but AI comes up with the premises for you. So you can actually look at the different types of segments you can build out. What are the different types of media partners I should work with? What are the right types of ad tech advancements I need to make?

And if you can layer those down and design a testing grid where you look at value versus effort, you can push some really interesting tests in a market.

What is your unpopular opinion? Something you feel really strongly about.

I have two. I’m not a fan of black boxes, and I know AI is in and of itself a black box. It's a mathematical formula that comes up with a solution for you. I like testing AI. So, when someone says I have a magical LLM — large language model — that comes up with a solution, that’s not enough.

You really have to know how the system works. So if it's dynamic creative optimization, how is it creating the template? How is it creating a structure for how you test those? How is it actually pushing the results in the market? What are the right types of parameters that it's looking at? So black boxes to me are a little bit of a faux pas.

The second part is transparency in terms of pricing. Our industry is a little bit delicate in terms of how media is priced. There's obviously the classic CPM, but then there's labor costs and data costs. The more transparent we can make the supply chain in terms of where the inventory is coming from, how the optimizations work, I think the better we’ll be as an agency.

The supply chain is coming up a lot lately. Is there increased pressure right now for everyone along the path to deliver value?

As an agent, we invest on behalf of our clients and it's really important that we show clients value wherever it's possible. Sometimes those are value-adds. They could be a measurement study, could be special types of inventory, it could be special deals that we negotiated on their behalf.

It could be new types of creativity or sponsorships. But clients want to know exactly where their dollars are going. So, we are very diligent with counting all of our beans. We want to make sure that we're good shepherds of our clients' dollars. And if something is not working, we very quickly shift.

And are you hearing that question from clients — are you feeling the scrutiny from clients?

Clients want transparency. They want agility. They want smarter thinking. So topics around optimization, they want to know something. Let’s say it’s an experiment with a new publisher and it’s not working. “How are you optimizing my dollars? Are you changing out the creative? Are you thinking about a new audience? Are you thinking about a different time of day? Is it a different part of the website?”

All of those are components that we’re always constantly trying to strive toward. If we don’t have the data, if we don’t have the right folks — hands-on-keyboard — working in the systems, it’s just not enough.

Marketers are uncertain where cookies are going, whether they’ll deprecate and when. How are you helping your clients navigate the privacy discussion and future-proof their strategies?

Over the last three years, we’ve done a multitude of “cookie-proof assessments,” as we call them. We had expected the cookies to be gone by now. Cookies are here to stay, but it doesn’t change the cookie-loss future that we’re all in. You can create an identity layer through a series of ecosystem partners.

We [also] expect that there will be privacy changes that come down the road that could be state specific. So KPA is going to change. Every state is approaching financial services and health data in nuanced ways. With enough identity layers and with enough security and privacy and with enough clean room technology, we’re confident that you can receive data that’s fully compliant, and HIPAA [Health Insurance Portability and Accountability Act] compliant for health care, for example.

You have a great title. Tell me, what does an activation chief do?

One of my big obligations is to take the data — some of that is internal, from our partners and tech companies — and infuse that into the media thinking.

Media investment as a traditional channel has been about putting dollars where they’re best spent. The data that’s available is now blowing up. So the question for my team is always “What are we receiving? How are we cataloging it? Are we using it in the best-possible way? Are we automating the tasks to get that data into the right format, [and] infuse [it] into the platforms? How do you take […] an omnichannel journey perspective?”

If we’re trying to capture a consumer that’s in-market for a car, or a home, or a box of cereal, how do we take all those data points [across platforms]?

You're making the case for omnichannel versus multichannel — not just adding on channels for the sake of it, but really thinking about it in a holistic way.

You want to hit the consumer wherever they are with an omnichannel journey. You want to bounce them around at the right points in time. So if you’re thinking about a website as a sales-capture mechanism, someone’s researching a new T-shirt that’s on the market, someone’s opening a basket, someone is purchasing — those are omnichannel touchpoints. It’s critical that you map out at what point you want to expose consumers to different parts of that research cycle.

Are you observing any surprising growth channels or strategies?

The ability to function faster is in and of itself a trend. When you look at streaming, when you look at different connected devices, whether it’s TV or it’s different types of digital inventory, the fact that a consumer is engaged — that they’re observing metrics like attention — for us, is critically important because we’re trying to capture meaningful moments and for us, meaningful identity, meaningful connection with consumers. That’s really the most important aspect, the most important signal.