# How AI Is Reshaping Consumer Insights and Marketing Research By Aby Varma · Published 2026-07-06 · Updated 2026-11-02 · Professional AV on MarketScale Canonical: https://www.marketscale.com/industries/pro-av/how-ai-is-reshaping-consumer-insights-and-marketing-research Creator hub: The Marketing AI SparkCast > David Jacobson of The Coca-Cola Company joins Aby Varma on The Marketing AI SparkCast to explain where synthetic personas work, how to separate a real trend from AI noise, and how AI agents are taking on repetitive insights work. ## Key points - Synthetic personas are useful in specific situations, and their outputs still need a human in the loop. - Data quality determines whether AI-generated research holds up, and one data point is not a trend. - AI is making personalized creative cost-efficient at scale and speeding up creative testing. On this episode of The Marketing AI SparkCast, host Aby Varma, founder and CEO of Spark Novus and the Marketing AI Pulse community, talks with David Jacobson, Global Senior Director of Applied Consumer Science at The Coca-Cola Company. David brings 25 years in marketing strategy, starting as a brand strategist and researcher in traditional ad agencies, moving through digital and experiential work and consulting roles at Publicis Sapient and PwC, and landing on the brand side at Coca-Cola about three years ago. He now sits on Coca-Cola's global human insights team, piloting new research methods and applying AI to audience segmentation and consumer understanding. ## Where synthetic personas work best David says Coca-Cola was already embracing AI when he arrived. Employees use enterprise versions of Copilot and ChatGPT integrated into company systems, with training that runs from AI basics to prompting to image creation. Every AI project goes through an AI governance portal, which he describes as a way to guide people, avoid duplicate projects and share resources across the organization. The open question is confidence. AI is fast. The harder part is knowing whether its answers reflect what real people actually say, experience and respond to. So the team still runs focus groups, diaries, surveys and ethnography alongside its AI work. > We do definitely go by the mantra that there needs to be a human in the loop, even if we're using AI. — David Jacobson, The Coca-Cola Company, on The Marketing AI SparkCast David defines a synthetic persona as a synthetic version of a human consumer. It can be a digital twin or a set of synthetic respondents in a quantitative survey, built by programming an LLM with the same attributes a traditional persona uses: demographics, psychographics, income, lifestyle interests, passion points, motivations and values. The appeal is reach. A team can generate 60,000 survey respondents in a short amount of time, while recruiting that many real people is slow, hard to target and expensive, especially for a global brand researching many markets at once. His rule of thumb is that synthetic personas earn the most confidence early. If a team is exploring 10 ideas for a campaign or brand platform, synthetic research can narrow the field to the top three or four faster and more cost efficiently. As ideas move toward final development, especially for larger campaigns, Coca-Cola still validates with human-based pretesting and copy testing. He says getting AI models to 80 to 90% accuracy is a pretty good number, and his team checks outputs against years of historical research on switching behaviors, drinking occasions and motivations. He also flags a pattern from pilots with synthetic persona vendors: output that looks thorough and still says little. > It's easy to get a large volume output, a 100 page slide deck that was mostly put together through an AI model after the synthetic persona research. But then the question for us is, are there insights there? — David Jacobson, The Coca-Cola Company, on The Marketing AI SparkCast The test, he says, is whether an insight comes down to a core human tension or human need. ## Data quality and the trend-versus-noise problem Aby presses on data quality, especially for research spanning countries where data quality varies. David agrees. > The old adage is definitely true with AI, garbage in, garbage out. — David Jacobson, The Coca-Cola Company, on The Marketing AI SparkCast Coca-Cola trains its personas on internal datasets and trusted third-party data partners such as Nielsen, rather than having models scrape the public internet. He still wants local context, including what's happening politically, economically and culturally in a market and with content creators, so the team combines trusted internal data with outside signals it doesn't have in-house. Separating a genuine trend from a skewed signal follows the same discipline as traditional research. > One data point is not a trend. ... Let's not use AI to validate AI generated insight. Let's see if we see that same point elsewhere. — David Jacobson, The Coca-Cola Company, on The Marketing AI SparkCast That means talking to local experts and people in the market, and checking trend reports from firms such as Kantar Global Monitor and Morning Consult. His reasoning: AI models are good at averaging huge amounts of data and extrapolating what comes next. A few small signals can look like a growing trend to the model, when in real life the behavior may stay with a small group or a single local area. ## Personalization at scale and faster creative testing Audience segmentation is where David sees AI doing what it was built for: finding commonalities across large amounts of data. Fusing datasets used to take highly trained data scientists and a lot of spreadsheets. Now an AI model can combine segmentation with retail data, first-party data and third-party sources, then match a person in a segment with a piece of creative and a media channel such as Meta or TikTok. That changes the math on small audiences. A sub-segment may lack volume but still be very valuable, and it no longer has to go without tailored creative. > Now you can create that creative or that offer or that promotion specifically for that group in a cost efficient way. — David Jacobson, The Coca-Cola Company, on The Marketing AI SparkCast Creative testing is shifting, too. Older testing relied on rough animatics and storyboards that didn't look like a final ad. AI can now generate polished stimuli. Traditional research companies have trained AI models on thousands of ads they've tested over decades, and David says he has seen 90% confidence levels and higher with some of Coca-Cola's partners and vendors. It doesn't replace all human testing, but it lets teams test many more versions and variations more cost efficiently, and it helps in markets where recruiting respondents is hard. The payoff isn't necessarily more campaigns. It's speed. David notes it could sometimes take a year or more from initial research to a campaign seeing the light of day. The path from insight to brief to creative development to market is now shorter, which also makes it easier to localize campaigns to local trends and audience segments. ## AI agents take on the repetitive work David calls agents a hot topic for his team. The fit is any process repeated over and over: the same campaign report, the same quarterly report, the same metrics. An agent can pull the data inputs on a regular basis so people spend their time on what the report actually says. Agents are also starting to talk to each other, with one focused on first-party data and another on a partner's retail media data, looking for common attributes and insights. That's where he urges caution. > One piece of bad data from one agent ... that hallucinates or gives a false positive or whatever, then that gets used by a second agent, and then it starts to expand and grow. — David Jacobson, The Coca-Cola Company, on The Marketing AI SparkCast Data privacy and sharing requirements, he adds, have to be worked out ahead of time, whether through a data clean room or internal systems. ## Where marketing leaders should start David's advice applies to businesses of any size. A small business doesn't need a bespoke LLM. It can build a rudimentary synthetic persona by prompting ChatGPT, Gemini or Copilot with the attributes of its ideal customer profile, then experiment with questions or creative stimuli. Working with a specialized partner raises confidence. Either way, he recommends having your own governance approach for validating outputs, whether that's past business performance, checks with real humans or gut checks with colleagues. > One, don't be afraid. And two, just start experimenting ... and seeing what works for your situation and your use cases. — David Jacobson, The Coca-Cola Company, on The Marketing AI SparkCast His third point is to learn, validate and share notes with others, including at Marketing AI Pulse events. As he puts it, "we're all learning at the same time." Watch or listen to the full episode of The Marketing AI SparkCast to hear David Jacobson and Aby Varma on synthetic personas, trend validation, creative testing and AI agents in consumer insights. ## Sources - [https://open.spotify.com/episode/0Rr4ifGUF9tNi29XtaKtaq](https://open.spotify.com/episode/0Rr4ifGUF9tNi29XtaKtaq) - [https://marketscale-4.wistia.com/medias/iu56o4552s](https://marketscale-4.wistia.com/medias/iu56o4552s) Tags: Aby Varma, David Jacobson, The Coca-Cola Company, The Marketing AI SparkCast, Spark Novus, Synthetic Personas --- Source: MarketScale, https://www.marketscale.com/industries/pro-av/how-ai-is-reshaping-consumer-insights-and-marketing-research. Published for AI indexing and citation; cite the canonical URL. Site guide for agents: https://www.marketscale.com/llms.txt