Point of View · Article 3
UK Research Industry Transition
Synthetic data: govern it, don’t debate it
Synthetic data will not replace human research, but it will reward the researchers who can prove where it works. Its value depends on behavioural grounding, transparent validation and expert judgement, which makes expertise more valuable, not less.
Few topics divide a room of insight leaders faster than synthetic data. One camp sees the end of slow, expensive fieldwork; the other sees a hall of mirrors. Both are partly right, which usually means the question has been framed badly. In Qualtrics’ 2025 Market Research Trends report, 71% of researchers expected most market research to use synthetic responses within three years. Adoption is not really in doubt. Trustworthiness is.
Fluency is not accuracy
Language models are trained to produce the most plausible answer, which in consumer research tends to be the most reasonable one. Real people are rarely that reasonable. They follow defaults, overweight losses and say one thing before doing another. An uncalibrated synthetic panel produces a confident, articulate average of good intentions: elegant, and of limited use for a pricing decision.
The commercial risk is real. In our reviews of UK agencies we found one synthetic product withdrawn within months of launch, without a published validation story to support it. Clients and codes are now asking for exactly that evidence: ICC/ESOMAR Article 7(e), EU AI Act transparency duties from August 2026 and the AI additions to ISO 20252:2026 all push towards disclosure.
Three questions worth asking
- ·
What is it grounded in? Validated behavioural measures and real-world data, not just prior survey responses.
- ·
What has it been tested against? A human benchmark, with pass marks fixed before the results are known.
- ·
What decision is it informing? Idea screening can tolerate approximation; a national launch cannot.
Why this is an argument for expertise
Here is the irony. The technology most often described as replacing researchers makes experienced researchers more valuable. Someone has to decide what a model can be trusted with, design the human benchmark, recognise when a plausible answer is wrong and take responsibility for the call. That is senior work. It does not need a large team beneath it; it needs judgement, method and the discipline to validate.
Used this way, synthetic data becomes a remarkable accelerator within a hybrid design: widening the funnel early, with real people confirming what matters. If you are weighing up where it belongs in your own programme, it is a conversation we are always glad to have.
Sources: Qualtrics, 2025 Market Research Trends Report (October 2024). ICC/ESOMAR International Code, Article 7(e); EU AI Act, Article 50; ISO 20252:2026 (Research Live). What Humans Do, UK Research Growth Matrix (base data September 2026; quadrants at £50m turnover and 4% growth) and UK agency reviews (October 2026).