By: Matt Emma
The CEO and co-founder on why the rush to bolt AI onto static research is solving the wrong problem, and what the industry needs to look at instead.
Did the message actually move anyone?
It’s the question Sara Fagen has spent more than two decades trying to answer. And it’s the question, she argues, that market research has been quietly failing to answer for that entire time.
The market research industry took in $153 billion globally in 2025, up from $140 billion the year before, according to ESOMAR. But Fagen’s argument is that most of that spending has been buying the wrong thing for two decades, and the AI moment is finally making it impossible to ignore.
Research vendors are racing to bolt generative AI onto static workflows, replace human respondents with synthetic data, and rebrand summary tools as intelligence. The cycle is moving fast. The underlying problem, in Fagen’s view, has been there the whole time.
She’s spent those two decades at the front of data-driven persuasion, helping pioneer the modern microtargeting playbook in campaign politics and at Deep Root Analytics before co-founding Tunnl. Across politics, advocacy, and the Fortune 500, she says she’s never met a consultant or CMO who didn’t want a real answer to her one question.
“Honestly, I’ve never met a single consultant or CMO who wanted to waste a message, burn a dollar on the wrong audience, or walk away from a campaign unable to prove what it accomplished,” Fagen says. “Nobody wants that. And yet it happens constantly.”
The reason it keeps happening, in her view, is structural. “The industry has spent two decades getting better at understanding audiences without getting any better at moving them,” she says. The AI cycle, she argues, threatens to repeat the same mistake at much greater speed.
That argument is the foundation of Tunnl, the platform Fagen co-founded to operate as what she describes as the bridge between two systems that should have been connected all along: research tech and marketing tech.
Why Now
For most of the last two decades, the gap between research and outcomes was a slow-burning problem. Findings produced one report. Marketing acted on a different dataset. The two systems didn’t talk to each other, and somewhere in the middle, value leaked out at the seam.
Three things have made that gap suddenly urgent. Cookie-based targeting is breaking down, forcing brands to rethink how they identify audiences without the proxies they’ve leaned on for years. CMOs are under harder pressure than ever to prove ROI in dollar terms. And the AI moment has flooded the market with tools that promise insight at speed, leaving buyers struggling to tell the difference between a real prediction and a plausible-sounding one.
The result, Fagen says, is that “close enough” no longer survives in environments where being wrong has consequences.
“When you’re a pharmaceutical company navigating a public health crisis, or a brand managing a reputational crisis, you don’t really have the luxury of being close enough. What changes is accountability. You need to know not just that you reached an audience, but that you moved them. And you need to be able to measure that in real time against your actual KPI, not against impressions or click rates.”
The Wrong Problem
Fagen’s critique of the current research moment is precise. The industry, she argues, is racing to make insight generation faster, while almost no one is focused on whether those insights actually lead to measurable change.
That’s not a critique of AI itself. Fagen’s company is built on it. The critique is about how market research is using it, layering AI summarization onto static reports, replacing real survey respondents with synthetic ones, and treating speed as a substitute for ground truth.
The Qualtrics 2026 Market Research Report states that 69% of market researchers have already incorporated synthetic data into their work, even as the industry continues to debate whether AI-generated respondents can substitute for human ones in high-stakes research.
“In a market full of AI-native competitors, our edge isn’t really the model itself,” she says. “It’s what the model is trained on. Any platform can generate a plausible-sounding audience description. Only a platform with calibrated, ground-truth data can predict how that audience will actually behave.”
Tunnl’s training data is built on hundreds of thousands of annual survey interviews, tens of thousands of individual-level signals, and more than a decade of observed campaign outcomes. The point, Fagen says, is not the size of the dataset. It’s the connection between what the data describes and what actually happens in the real world.
“LLMs don’t threaten that, they amplify it,” she adds.
Where the Money Goes
The waste, Fagen says, hides in three places:
- Research that produces a deck no one can activate
- Teams working off different sources of truth, duplicating work and making decisions that contradict each other
- Reach without precision, where impressions land on people who were never going to be persuaded in the first place
“The waste comes from the gap between understanding and action,” she says, “and from the sheer number of vendors and handoffs required to close that gap.”
The Bridge
The reason research and marketing have stayed in separate systems for so long, Fagen says, is partly historical. Research vendors produce findings. Marketing technology platforms act on data. The two have never been built to talk to each other.
“The moment they’re separate, value leaks out at the seam between them,” she says. “Tunnl’s core position is that every survey question you ask should be tied to real, identifiable individuals, not just a sample of them, so the learning immediately becomes an activatable audience.”
Inside the Tunnl platform, that thesis takes the form of Research Studio, which closes the loop from survey fielding to AI-powered modeling to audience activation in a single pipeline. The research doesn’t produce a deliverable to hand off. It produces targeting that gets acted on, and outcomes that can actually be measured.
It’s the kind of system, Fagen argues, that a well-funded competitor cannot simply replicate.
“We started in campaign politics, where the cost of bad targeting was an election lost, before extending into commercial markets. That ground truth isn’t something you replicate by hiring a team and raising a Series B.”
So What?
Fagen sums up her argument in a phrase her team uses internally: understanding is cheap, changing outcomes is hard.
“The market is full of tools that give you understanding. Dashboards, surveys, analytics platforms, AI summarization, all of it. Getting from ‘here’s what your audience thinks’ to ‘here’s how you moved them’ is much harder.”
“Understanding without activation is just a deck. Understanding without measurement tells you nothing about whether you actually changed anything in the real world. So it’s not really a tagline so much as a challenge to the standard research and analytics value proposition: so what?”
Her bet is that the only thing that matters at the end of a campaign is whether something changed. Opinions, behaviors, purchases, perceptions. They have to move. And that requires connecting the intelligence layer to the targeting layer to the measurement layer.
“Most companies in this space own one of those three,” Fagen says. “Tunnl knows that owning all of it, in one platform, and continuing to build out AI layers that keep improving how strategists work with it, is really the only way to actually change outcomes rather than just describe them.”
For two decades, that was the question market research didn’t have a system to answer. Fagen is betting it does now, and that the rest of the industry will eventually have to look at the problem differently.
