I recently hosted a roundtable with two longtime friends and marketing science veterans: Josh Chasin and Rick Bruner. Between them, they’ve led research and measurement at Comscore, VideoAmp, DoubleClick, Viacom, and more. We explored how AI is fundamentally changing market research, from classic surveys to data science, synthetic respondents, and agentic AI interviewers.
From Survey Sampling to Data Science
Josh framed the biggest shift clearly: research used to be the domain of survey sampling statisticians. Today, the core practitioner is the data scientist.
Instead of “What do I want to know? Let me draw a sample and ask,” we increasingly ask, “What do I want to know? Let me find and aggregate existing data assets.” That shift, from creating data via surveys to extracting insight from behavioral data, is exactly where AI excels, ingesting massive data sets and surfacing patterns and predictions at scale.
Synthetic Respondents and Survey Limits
Rick tackled one of the more provocative themes: synthetic research and synthetic respondents. Traditional survey research faces real problems:
- Fraud and “professional” respondents
- Survey fatigue and straight-lining
- Underrepresentation of key groups (e.g., young men)
- Human limits in predicting our own future behavior
At the same time, billions of people leave trillions of behavioral signals online. AI can build synthetic personas that often predict behavior more reliably than rushed, low-quality survey responses.
Josh added that even for past behavior, self-reporting is shaky. When metered or passively observed media exposure is compared with what people say they watched or did, the answers frequently diverge and typically underreport true consumption. Surveys aren’t dead, but they are no longer the unquestioned backbone of insight.
Identity, Privacy, and Synthetic Consumers
As research becomes more data-driven, identity resolution and privacy sit at the center. Brands want to enrich and activate their first-party data without violating consumer trust.
Josh discussed work on synthetic ID spines: “virtual identities” that stand in for real individuals. These ID constructs:
- Reduce privacy risk
- Can be enriched via data matches (e.g., smart TV viewing) and AI inferences
- Can be further trained with surveys for attributes that don’t leave digital traces, like underlying motivations
In this model, surveys increasingly serve to train and validate AI rather than acting as the sole source of truth.
Agentic AI and Voice-Based Interviews
I’m particularly interested in agentic, voice-based AI interviewers, like my client ReadingMinds.ai, that conduct self-driving research:
- Conversational, voice-based interviews
- Dynamic follow-up questions that “go down rabbit holes”
- Rich capture of emotions, motivations, and language patterns
This approach merges the depth of qualitative research with the scale of quantitative. Respondents get to talk instead of ticking boxes, and AI plus NLP can mine the resulting conversations for themes and signals that static questionnaires often miss.
Experiments, Causality, and Impression Quality
On media ROI, Rick stressed that AI will happily optimize to correlations, but correlation is not causation. He argued for randomized controlled trials (RCTs) in media, automated experiments that compare exposed vs. unexposed groups to reveal true incremental impact.
Josh closed by shifting focus from just reach and frequency to impression quality, what actually happens between the screen and the face. In a fragmented, precision-targeted world, attention, emotion, and resonance matter as much as raw exposure counts.
AI isn’t killing research; it’s forcing it to evolve, toward richer data, better experiments, and deeper understanding of human motivation.
Discussion points include:
- The shift from survey sampling to data science as the core of market research
- How AI and synthetic respondents can address survey fraud, fatigue, and sample gaps
- Why self-reported behavior (past and future) is often unreliable
- The role of identity resolution, privacy, and synthetic ID spines
- Using surveys as training data to enrich AI-driven consumer models
- The potential of agentic, voice-based AI interviewers to scale qualitative depth
- The importance of randomized controlled trials (RCTs) for true media causality
- Moving beyond reach and frequency to attention, emotion, and impression quality in media measurement