Hershey’s bets on AI and Media Mix Modeling to reinvent its media decisions
Hershey’s — yes, the chocolate giant — is making some really interesting moves in the digital advertising world.
And we’re not talking about just another creative campaign.
The company spent an entire year developing an Artificial Intelligence system capable of transforming the way it measures, plans, and decides where to invest in media. The idea was never to simply adopt an off-the-shelf tool, but to build something custom — something that connects sales data, campaign performance, and media modeling into a single ecosystem powered by AI agents.
At the heart of this strategy is an idea that sounds simple but changes a lot in practice: relevance over reach.
The person explaining all of this is Vinny Rinaldi, VP of media and marketing technology at Hershey’s, who shared details on how the company is using Media Mix Modeling — the well-known MMM — along with AI agents to make smarter, faster decisions that are more closely tied to actual business outcomes. 🍫
The conversation goes from the brand’s entry into Reddit all the way to who should actually own the agent that buys media. The answer goes against what a lot of people in the industry still believe.
A year building something different
It’s not every day that a company decides to hit pause and dedicate twelve full months to rethinking how it makes media decisions. But that’s exactly what Hershey’s did. The project started with a fundamental question: how do you make sure every dollar invested in advertising is actually making a difference?
The answer didn’t come from a ready-made platform or an outside vendor with a packaged solution. It came from within, through a lot of hard work, iteration, and most importantly, data. The company worked with partners like Mutinex and Tracer.tech to build a proprietary framework made up of multiple Artificial Intelligence agents that integrate various data sources. The goal was to understand, in near real time, what’s working and what’s not in active campaigns — something that used to take months to analyze and now happens in a matter of weeks.
At the core of this framework is Media Mix Modeling, a methodology that analyzes how different media channels — TV, digital, social media, paid search, among others — individually contribute to business results. MMM isn’t new to the industry, but what Hershey’s did was supercharge this approach with layers of Artificial Intelligence that make the process much faster and more precise. Instead of waiting months for a backward-looking analysis, the company can now adjust its strategies dynamically, reacting to market signals before the budget gets wasted on channels that aren’t delivering.
Another thing that stands out is the scale of the project. Hershey’s doesn’t have just one product — it manages a massive portfolio of brands, each with different audiences, distinct seasonalities, and specific goals. Building an AI system that can handle all that complexity without losing the granularity needed to make good decisions on each front is a considerable technical and strategic challenge.
Receive the best innovation content in your email.
All the news, tips, trends, and resources you're looking for, delivered to your inbox.
By subscribing to the newsletter, you agree to receive communications from Método Viral. We are committed to always protecting and respecting your privacy.
Rinaldi compared the process to building a house: on the outside it might be the most beautiful thing in the world, but if you forgot to pour the concrete foundation, the first storm is going to knock it all down. For him, that foundation — the data and technology infrastructure — is the least glamorous part of the work, but it’s also the most important. And it’s exactly what a lot of people skip over in the rush to adopt AI agents.
Relevance as a principle, not a metric
When Vinny Rinaldi talks about relevance over reach, he’s touching on something that goes way beyond a catchy phrase for a slide deck. This shift in mindset represents a break from decades of advertising logic that prioritized volume — how many people saw it, how many times, across how many channels. Hershey’s is betting that reaching the right person, at the right time, with the right message is worth far more than pushing content to the largest possible number of people.
And Media Mix Modeling is the tool that lets them test that hypothesis with real data, not gut feeling.
Rinaldi made it clear that this isn’t about choosing one over the other. Reach is still important, but the question guiding decisions has changed. Instead of asking how many people saw this?, Hershey’s started asking who was this actually relevant to right now? And did it change that person’s behavior?
In practice, this changes how the brand evaluates the success of its campaigns. A platform that delivers an impressive volume of impressions but can’t prove real impact on sales or purchase consideration starts getting questioned. Hershey’s Artificial Intelligence system cross-references media data with business data in a way that makes it much harder to hide inefficiency behind vanity metrics. If a channel isn’t contributing in a measurable way to results, it needs to justify itself or lose its share of the budget.
It’s simple in theory — but operationally very complex to execute without the right technology infrastructure.
The Reddit case and investment decisions
The relevance-over-reach philosophy led Hershey’s to explore territories that were never part of its media mix before. One of the most striking examples is the brand’s entry into Reddit, especially for the Reese’s brand.
Reddit isn’t the channel with the biggest reach in the world, but it has extremely specific and engaged communities. For certain products in Hershey’s portfolio, being present in a genuine conversation within a baking community or a candy fan group can generate far more value than a banner shown to millions of people who are thinking about something else entirely.
And the data backed up that bet. According to Rinaldi, what Hershey’s quickly discovered is that Reddit was moving more units per dollar invested than any other channel. That’s the kind of insight that MMM powered by AI can capture — and that would be very hard to identify using traditional measurement methods, which tend to take longer to deliver results and don’t always manage to isolate the impact of smaller channels.
After that discovery, the question shifted from should we invest in Reddit? to how do we do more and in a more relevant way on the platform?
At the same time, the company keeps investing in traditional channels. Hershey’s current mix sits at around 72% digital and 28% traditional channels. The diversification had been happening for a while, but the AI system brought more clarity to decisions like: if I pull money from traditional cable TV, where exactly should I put it? And more importantly — what’s the estimated impact on revenue? The system can now run those projections, which completely changes the dynamics of conversations during the American TV upfront season.
Media agents and the ownership question
One of the most thought-provoking points in the conversation with Rinaldi is about so-called media agents — autonomous Artificial Intelligence systems that can make media buying and planning decisions without direct human intervention on every transaction. This is a space that’s evolving fast in the industry, and a lot of people are still trying to figure out what role agencies, platforms, and advertisers themselves play in this new landscape.
Hershey’s has a very clear position on this, and it goes against what most of the industry still practices.
First, on planning versus buying: Rinaldi believes the real potential of AI agents lies in planning, not just buying. The market tends to focus on the transactional side — how to buy faster, cheaper, with more operational efficiency. But if the agent only focuses on buying, it gets stuck on reach and cost metrics. When the focus shifts to planning, the questions become more strategic: what’s the right message? For whom? In what context?
With a modernized MMM, it’s possible to build a planning agent powered by hyper-focused data and specific outputs. Hershey’s has already started testing this kind of approach.
On the buying side, the company isn’t standing still either. Hershey’s has been working for the past three years with Chalice, customizing bidding algorithms that integrate with major DSPs, primarily The Trade Desk, but also YouTube and Meta. This isn’t a recent adoption — it’s an incremental build that has been maturing over time.
But who owns the agent?
For Rinaldi, the agent that buys and plans media should belong to the advertiser, not the agency and not the platform. This isn’t a knock on agencies — it’s a matter of incentive alignment and control over proprietary data.
When the AI agent is owned by the brand, it operates with the brand’s objectives as its absolute north star. When it belongs to a third party, no matter how well-intentioned that third party may be, there are layers of interest that can influence decisions in ways that aren’t always transparent. 🤖
Hershey’s owns all of its technology and data contracts. The agency — in this case, Publicis Groupe, which has handled media operations in the United States since 2024 — works within the brand’s ecosystem but doesn’t own the infrastructure. It’s an important distinction. According to Rinaldi, if Hershey’s didn’t have that ownership structure, it never would have been able to build what it has today.
A concrete example: since 2018, the company has held a direct contract with The Trade Desk. It was never the one directly operating buys on the platform, but it always maintained access and control. The relationship works like a triangle — a joint business plan between Hershey’s, Publicis, and TTD.
The role of human critical thinking
In the middle of all this talk about automation and autonomous agents, Rinaldi makes a counterpoint that’s worth highlighting: human critical thinking is still indispensable.
In his view, the real value of automating planning and buying tasks through AI agents isn’t to eliminate the need for people — it’s to free up time so people can do what they do best: think strategically, question assumptions, and connect dots that a machine alone can’t connect.
If an AI agent takes over the more repetitive operational tasks, the agency team — which, in Rinaldi’s words, knows the market better than the internal team would have time to keep up with — can dedicate more energy to high-level strategic thinking. This raises the quality of decisions across the board and creates a more productive collaboration dynamic between brand and agency.
As for the possibility of agents buying media directly from publishers without intermediation, Hershey’s hasn’t gone that route yet. Rinaldi believes the separation of roles between agency and brand remains important, and that there’s a layer of human intuition that needs to be preserved in this process. Technology is a means, not an end.
What all of this means for the industry
What Hershey’s is building goes beyond a technology project. It’s a new layer of data sovereignty and media intelligence. In a market where platforms have more and more power over how budgets get allocated and how results are reported, having a proprietary Artificial Intelligence system that mediates all of this represents a real competitive advantage.
Rinaldi made an interesting analogy: the acceleration of tools like ChatGPT, Claude, and Gemini in people’s lives is reminiscent of the impact Google Search had when it first appeared. But people forget it took years to build the infrastructure that made that launch moment possible. The real value is in the base — in the foundation — not in what shows up on the surface.
Here are some key takeaways from Hershey’s experience:
- Infrastructure first: before implementing any AI agent, you need to build the data foundation and systems that will feed it. Without that, any result is fragile.
- Data ownership: whoever controls the data and technology contracts has more autonomy to innovate and ensure decisions are aligned with business objectives.
- Planning before buying: the market tends to focus on automating media buying, but the biggest strategic impact may come from automation and intelligence applied to planning.
- Measurable relevance: abandoning vanity metrics in favor of models that connect media investment to real business outcomes isn’t just a preference — it’s a competitive necessity.
- Humans at the center: AI should free up time for critical thinking, not replace it.
The Hershey’s case shows that adopting Artificial Intelligence in advertising doesn’t have to be a race for buzzwords or superficial implementations. When technology is treated as a tool in service of a clear strategy — and not as the strategy itself — the results tend to be much more solid and sustainable.
For brands considering a similar path, the main message might be this: invest time in the foundation. It’s the least glamorous part of the work, but it’s the part that will determine whether what you’ve built survives the first storm. 🏗️
