‘We should own it’: Hershey’s programmatic chief on AI agents and media mix modeling

Hershey’s has been working with companies like Mutinex and Tracer.tech to speed up media measurement reporting and, in turn, their media investment decisions. The confectioner spent a year developing a system of several AI agents that provide media modeling for campaigns over the course of weeks, not months.

Vinny Rinaldi, vp of media and marketing technology at Hershey’s, said the new approach has enabled the company to prioritize “relevance over reach,” elevate planning and strategic objectives above the cheap thrill of a low CPM, and link media strategy closer to business strategy. It’s also helped steer investment decisions, such as a recent lean into Reddit, and will be informing the advertiser’s approach to the current TV upfront season.

In this conversation, Rinaldi explained how the strategy had altered its investment choices, Hershey’s thinking around agentic media buying and planning, and where tech should be stopped.

This conversation has been lightly edited for conciseness and clarity. 

Why has Hershey’s adopted a relevance over reach approach?

It’s not one or the other. It’s a compound statement. You still want to garner reach, but being relevant to the right reach is the most important thing for brands. You can gamify any system to go after tonnage for a lower cost. It doesn’t mean you’re standing out. 

[Relevance over reach] is a shift in how you think about your measurement ecosystem, [asking] a more fundamental question, not: how many people saw this? But really, who was this truly right for right now? And did it change your behavior?

So, has the measurement framework prompted you to alter any spending plans?

We went into a couple unchartered territories that we weren’t used to, one of the major ones being Reddit [particularly for Reese’s]. Reddit is a very scalable platform, but it’s really relevant to each consumer base. Each subreddit, how you interact with those consumers and show up as a brand, needs to fit. 

What we learned with Reddit quickly is that they’re moving more units per dollar spent than any other channel. So we said: “OK, that was a good bet — now, how do we move and do more of it and become more relevant within the platform?”

There’s a lot of traditional channels that are still being purchased. We’re not taking a big swing out of television because we’ve really diversified our portfolio mix quite a bit. We’re about 72% digital and 28% traditional [channels]. So we’ve already been on that journey.

But as you look at the upfront acquisition, where am I spending my dollars… and if I were to pull out of old school cable, where would I put it? This machine is allowing us to start to not only say where I would put it, but also forecast the revenue impact.

Do you expect to start looking at building out buying or planning agents?

Short answer, yes. I think planning is the most important thing that gets overlooked.

As an industry we’re super focused on the buying aspect. If you only focus on buying, you’re only focused on reach and cost measures. If you’re focused on relevance, you hone in on [your]  communication strategy, what [you’re] saying to a consumer, how [you’re] speaking to them and where [you’re] speaking to them. You start to ask very different questions.

Related Insights

Life Beyond the Cookie

More brands are blending deterministic and probabilistic data for hybrid targeting approaches Read More

With a modernized MMM, I think you can build a planning agent around hyper-focused data and outputs. We have started to test out what does that look like from a planning perspective.

From a buying perspective, the same can also be true. You can take all of this data and feed it into a buying agent and/or custom algorithm — which we’ve been actually spending a lot of time and effort on for the last three years. We’ve been one of the early partners of Chalice. We’ve customized a lot of our bidding ecosystem using their data sets and using their integration into a lot of the major DSPs, mainly The Trade Desk, but across YouTube and Meta as well. 

Do you foresee agents eventually being able to buy direct from publishers?

We haven’t really gone down that path. The separation of both agency and brand is really important… I still think there’s human intuitiveness that’s needed. I think critical thinking is one of the most important levers that we have not focused on as an industry in general.

If you can replace time spent for both planning and buying through an agentic layer, you’re now elevating the critical thinking capacity of your agency [team] — who, by the way, know more about the industry than we have time to spend on.

So, should Hershey’s own that agent, or should the agency?

That’s a great question. To make it differentiated for a company like Hershey, inherently, we should own it. I say that because it’s our data fueling it. You might want to pile on top some agency data, but at the end of the day, it becomes our asset because we’re feeding it sales data and other assets that an agency won’t own.

We own every single technology and data contract. The agency just sits in our buying seats in our ecosystem, but they are still our agents-of-record, and I think it’s a really important distinction because if I didn’t own all of my contracts, we would have never been able to build that infrastructure out the way that we have. [Publicis Groupe has handled U.S. media duties for Hershey’s since 2024.]

[For example], since 2018 we’ve owned our contract directly with The Trade Desk. Again, we’ve never been the one sitting and buying in the platform, but we have always owned access. it’s a triangular relationship, always. So our JBP [joint business plan] is between us, Publicis and TTD.

What were the big lessons you took from the year working on this project?

It’s funny. You look at what’s happening in the world and everything you read about is agentic-this, or AI-that. The acceleration of ChatGPT, Claude and Gemini into the forefront of everyone’s lives feels like the replacement of what Google Search was back when it first launched. [But] it took years to build to that launch moment [and] the infrastructure that had to be built for those things to work is very undervalued.

That’s the journey we’ve gone on. A lot of the time that we’ve spent in the last year was building out our infrastructure. It’s like building a home: it can be the most beautiful thing on the outside, but if you forgot to pour the concrete foundation, the first storm is going to blow it over.

Pouring that foundation is critical to this, to the success of any of these agentic AI capabilities, in my opinion, and I think it’s the first thing that people overlook, because it’s the most unsexy part of the job.