The most useful result of an advertising campaign may never appear in a click report. Someone can come away trusting a company more without visiting its website. Understanding what shaped that response could help marketers decide what to say next, before they commit millions to another campaign.
Sara Fagen, CEO and co-founder of Tunnl, wants to make that process conversational. Her Arlington, Virginia-based audience intelligence company turns survey responses into predictive models that help brands reach relevant audiences and track shifts in attitudes. She now envisions marketers talking to an AI version of the people they hope to persuade, drawing on years of human research to explore how a message might land.
“Synthetic research has to remain grounded in a human-in-the-loop conversation, with people’s responses providing the foundation for that work. We conduct several hundred thousand interviews every year, combining the research that powers our platform for subscribers with the custom work our brand clients ask us to undertake,” Fagen tells me in an exclusive interview. “Every one of those questions produces a signal that we use to build a corresponding model. Our technology automates that process, allowing us to translate survey responses into models quickly, with machine learning powering much of the work.”
A four-time founder, Fagen co-founded Tunnl in 2021 after earlier ventures that included Resonate and Deep Root Analytics. Her career began in politics and polling, where a campaign deadline made precision essential. That experience informs her approach to understanding what might persuade someone, beyond what demographic characteristics reveal.
Connecting Consumer Insights With AI Modeling
Fagen says Tunnl’s platform holds more than 25,000 variables on each U.S. person. Research Studio connects survey work to audience building, while the company’s media capabilities help clients put those insights into advertising. Moreover, the platform’s television tools help advertisers plan spending around their target audiences’ viewing habits across traditional and connected TV, building on Fagen’s earlier work in TV ad targeting at Deep Root Analytics.
Tunnl’s historical research also gives its team a way to study how opinions evolve. Fagen describes more than a decade of data and an ongoing effort to determine how much fresh research its models need.
“Some people’s attitudes change regularly, while others remain relatively consistent. Our research gives us a good sense of those patterns, but we still need to capture how opinions evolve over time,” she says. “That is part of what we are exploring now, understanding how much fresh information we need to bring into the existing data set so that our models continue to reflect what people think as their views change.”
She expects synthetic research to develop alongside established methods, giving brands room to explore an idea before undertaking a larger study. Over time, she believes that relationship could change how companies commission research itself.
“We could eventually reach a point where the only research taking place is the work needed to feed synthetic models. That is a possibility. For now, though, most brands want their research rooted in methods they already know and trust,” Fagen says. “Perhaps you want to have a conversation that helps you work through how an audience might respond to a particular message. You could then explore what to say next, using that exchange to think through a follow-up message. Those conversational applications of synthetic research are very valuable.”
How Human Oversight Shapes AI Marketing
At Tunnl, faster modeling is changing how specialists spend their time. AI allows them to take on other assignments while retaining responsibility for reviewing the models they produce.
“In the data space, human review remains an important part of the modeling process. About 10 years ago, we might have had four or five people working on a model over time. Today, one person with the right skills and an understanding of the code can do much of that work quickly,” Fagen says. “The same principle applies to creative work. Someone still needs to review what AI produces and make sure the output makes sense. Adding a touch of humor can also help people become more receptive to AI-generated advertising.”
Fagen uses AI herself, including to sharpen emails, and reminds employees to review their work before sharing it. She expects consumers to become more comfortable with AI-generated advertising as they experience the technology’s practical benefits in their daily lives.
Tunnl’s September financial trust report shows why retaining access to people matters. In a survey of 3,066 adults conducted during July 31-Aug. 4, 63% said they would not accept a loan or claims decision if AI made it entirely without a human appeal. Fagen applies the same sense of responsibility to audience targeting.
She believes that working with aggregated segments can reduce employees’ need to examine individual records, while companies remain responsible for honoring removal requests and complying with privacy laws. Tunnl also reviews how clients intend to use sensitive information and declines work when it identifies harmful intent. “As a data company, you have to understand how people intend to use the information you provide and decide where to draw the line. That responsibility rests with the operator, who needs to understand what is happening with the data and the purpose behind its use,” she says.
Tracking Brand Impact Beyond Advertising Clicks
Tunnl reports more than 450 platform users and over 80 active clients, including 17 Fortune 500 companies. Moderna is among its clients, and more than 8,800 advertisers activate Tunnl audiences through marketplace destinations. The company reports 14 direct advertising integration partnerships, including Vizio and TiVo alongside The Trade Desk and FreeWheel. It added 31 staff members over the past year and cites 60% growth over three years in its Inc. 5000 listing.
The platform’s Campaign Effectiveness product examines whether people’s attitudes shifted after a campaign, comparing modeled results among audiences that did and did not receive the advertising. External media buyers handle the spending, while Tunnl supplies the audience data and studies the response. For a brand advertising to build trust, that means looking for evidence that trust increased among the people it reached.
“Our audience data goes to an external media buyer who executes the campaign. The research then focuses on how that audience responds to the advertising,” Fagen says. She wants to bring more campaign data back from advertising platforms and connect it with survey findings. Linking exposure to changes in opinion could give marketers a stronger basis for allocating their next campaign budget.
“Buying media isn’t our role. The goal is to bring campaign data back and combine it with our survey results. There is still work to do to reach that point, but I think the combination could be very powerful,” she explained. “It would give brand clients greater confidence in the independence of the measurement and a clearer understanding of their campaigns’ impact.”
As AI takes a larger role in product discovery, Fagen sees opportunities to analyze more of the timestamped data behind consumer decisions. Survey research complements that information by tracking attitudes before and after a campaign.
“Defining the audience and establishing a research baseline gives us a clear starting point. Knowing who received the advertising and who did not provides a control group against which to assess the results,” says Fagen. “A survey before the campaign and another afterward capture attitudes at two distinct points in time. Keeping those findings in the same system allows us to measure the movement. That adds an important perspective alongside the clickstream data, helping brands understand how their audience’s views have changed.”
Tunnl wants to give a brand’s research a life beyond the campaign that produced it. Fresh survey responses would refine a synthetic audience that marketers could consult before committing their next budget. Over time, a brand could return to years of accumulated research with a question it had never thought to ask when it commissioned the work. Long after the ads stopped running, the research could still have something useful to say.
