Synthetic Data in Market Research: Use It to Help, Not to Replace
Using generative AI to create synthetic respondents is tempting, but LLMs aren’t people and can introduce bias or false confidence when used beyond defined use cases.
SOURCE · Harvard Business Review Read the original ↗The problem: using generative AI to create synthetic respondents is tempting, but LLMs aren’t people, and they can introduce bias or false confidence when used beyond defined use cases.
What the latest work shows
HBR’s 2025 research reveals both the promise and the limits of GenAI in market research. In “Using Gen AI for Early-Stage Market Research,” Brand, Israeli & Ngwe demonstrate how LLMs can act as synthetic customers to speed early discovery: idea triage, message and concept exploration, rough willingness-to-pay hypotheses, when used with proper guardrails: clear personas, realistic price bounds, and human anchoring. Their earlier Harvard Business School working paper reaches a consistent conclusion: LLMs can mimic broad preference patterns under constraints, but outputs are prompt-sensitive and must be validated with real respondents.
The takeaway: use GenAI to accelerate the front end, not to replace human data for decisions.
Our human-first, AI-augmented approach
- Start with humans. BrandSight™ and LaunchFit™ begin with real open-ended responses, typed or spoken, to capture how customers actually talk and choose.
- Augment responsibly. We use LLMs to expand language, concepts, and scenarios, clearly labeled and calibrated to the human baseline.
- Confirm before you bet. We follow with quick choice pulses (choice-based conjoint and MaxDiff) so you get dollar-valued drivers, price bands, and adoption forecasts you can defend.
Where GenAI helps most, and where humans stay non-negotiable
It helps most in early idea sorting, drafting realistic choice alternatives, and exploring edge cases before fielding. Humans remain non-negotiable for final segmentation, willingness-to-pay and packaging decisions, board-level tracking readouts, and nuanced cultural and emotional context.
The balanced path forward
Synthetic data can accelerate discovery when used thoughtfully, but it cannot replace the complexity and authenticity of human responses for final decisions. Our human-first-augment-confirm methodology delivers both speed and reliability: AI enhancing, never replacing, the voice of your customers.