Observed Signal · Jul 22, 2026 · Use Case / Case Study · Source: Digiday · Impact: 3/5 · Sentiment: Positive
Agencies adopt hybrid synthetic audience research
Agencies are increasingly supplementing traditional consumer research with AI-generated synthetic audiences, using a hybrid approach that layers synthetic personas on top of human feedback, syndicated data and social listening. Crowley Webb’s senior VP of data analytics, Andrea Berki-Nnuji, says the agency has used synthetic audiences over the past 18 months but typically vets the output with real humans, estimating fully synthetic datasets reach about 80% of required validity. Vendors and platform selection focus on data provenance, pricing, multi-user capabilities and compliance review. Use cases include concept testing, naming, media messaging and pricing; however, certain contexts such as rare-disease pharma work still require strictly human research.
Describes a growing agency practice combining synthetic and human research, with vendor selection and validation implications for agencies, data providers and MarTech vendors.
Track IBM Signals & Market Shifts in Real-Time
Polaris7 autonomous intelligence agents track regulatory filings, primary sources, executive changes, and deal flow 24/7. Create your free Explorer workspace to monitor these entities.
Key Takeaways & Evidence Grounding
- Agencies are testing AI-generated personas to supplement media planning and concept testing.
- Crowley Webb has used synthetic audiences for some clients over the last 18 months, layering synthetic data with human insights and social listening.
- Andrea Berki-Nnuji (Crowley Webb) said fully synthetic datasets typically get a project about 80% of the way there and require human vetting for the remaining 20%.
- Crowley Webb uses syndicated data from MRI-Simmons and statistical modeling with SPSS (IBM) when building and validating synthetic personas.
- Some clients (e.g., travel) found the synthetic tool produced an additional, COVID-driven persona not previously identified; pharmaceutical work sometimes prohibits AI use.
Connected Companies & Entities
2 Entities mapped“We were able to paint those personas that we created using SPSS [Statistical Package for the Social Sciences, an IBM software program used f...”
“Digiday recently caught up with Berki-Nnuji about how the agency is using synthetic audiences, how it tests and vets synthetic data tools, a...”
Ontology Mapping & Concepts
Related Market Signals & Shifts
Recent verified developments and strategic activity across this market segment.
Synthetic Research Depends on Audience Model Trust
Eric Ayzenberg (Soulmates.ai) argues that the real debate is not AI versus human research but whether the models behind synthetic audiences are trustworthy enough to inform decisions. Citing an evaluation by The Good, he says AI research tools can surface friction, accelerate workflows and provide directional input, but they often fail to reproduce observed behaviour, emotional nuance and the subtle signals needed for high‑stakes decisions. Ayzenberg proposes a three‑step approach: use synthetic audience models for quick directional clarity, use high‑fidelity synthetic models (validated against human data) when behavioural understanding is needed fast, and use traditional recruited field research when observed in‑the‑wild behaviour or regulatory/ investment rigor is required. The piece positions synthetic tools as complementary to human research, not replacements.
Synthetic Data's Role in Customer Research
The article explains how synthetic data — AI-generated, statistically representative datasets — can accelerate and broaden customer research by simulating audience responses, enabling synthetic focus groups, virtual personas, and digital twins. It argues synthetic data is most useful where real data is scarce or expensive to collect, and recommends starting with focused pilots tied to manageable decisions (for example, message testing or early product concept validation). The piece warns synthetic outputs reflect the quality and biases of their inputs and should inform but not replace real-world validation or human judgment. Governance, transparency about source data and model assumptions, vendor evaluation, and validation steps against observed behavior are presented as prerequisites for trust and long-term adoption. Over time, organizations that treat synthetic data as a disciplined capability can speed experimentation and make real customer input more targeted and valuable.
Synthetic Research: Promise With a Catch
This MarTech analysis examines the rapid rise of synthetic research—using generative models and synthetic personas to produce market and product insights—and the tradeoffs between speed/cost and scientific rigor. The synthetic data market is projected to grow dramatically, and many insight leaders plan to adopt synthetic approaches for scale and niche sampling. But off-the-shelf LLMs (e.g., ChatGPT, Claude, Gemini) can introduce bias, homogeneity and overly positive responses (the “Pollyanna Principle”), producing outputs that are difficult to validate. The article highlights techniques to improve reliability: fine-tuning models on proprietary survey data, applying a train-synthetic, test-real (TSTR) validation approach, and enforcing governance, transparency and persona provenance checklists. Case studies and experiments (including Stanford/Google DeepMind results and a Dollar Shave Club example) show promise when synthetic methods are benchmarked and verified against real-world data.
Track Real-Time Market Signals & Shifts
Set up custom watchlists to receive automated, evidence-grounded executive digests whenever material signals or shifts occur across your tracked landscape.