The report, Who owns understanding?, collates 10 co-authored articles featuring agency authors paired with client-side practitioners across a range of themes about AI and its impact on the market research industry. Participating clients include Shell, Co-op, Sage, Unilever and Panasonic.

In his introduction to the report, Colin Strong, chair of MRS Delphi Group and head of behavioural science at Ipsos, said that AI was performing many tasks quickly and competitively, which was leading to a shift in how research is valued.

“AI is changing who owns understanding inside organisations,” Strong wrote.

“When insight tools can be built, deployed and used by many different teams, authority no longer comes only from controlling access to data or owning the production process. It comes from knowing which outputs can be trusted, what they mean, and who is accountable when decisions are made from them.”

Adding that the future of the sector “will depend less on defending process and more on making explicit the value of expertise, judgement, interpretation and decision impact”, Strong also warned against losing human skills in an AI-led world.

“Research has never been only a process of extracting patterns from data. It is a discipline of context, interpretation and judgement, rooted in the lives and experiences of real people with often messy lives,” he argued.

“As AI becomes better at summarising , clustering and producing fluent outputs, the risk is not simply that people are replaced. It is that our understanding of human experience is thinned out: reduced to signals, averages and plausible narratives that lose contact with the lived experiences that research is meant to represent.”

He added: “AI does not make research less important but is does make the standards of research more important.

“When outputs become abundant, fluent and cheap, the scarce value is understanding: knowing what to ask, what to trust, what is missing, what matters and what should follow.”

Strong also questioned the use of the phrase “human in the loop”, and what people actually mean by the term.

“Human involvement cannot simply mean that someone reviews, approves or signs off an AI-generated output,” he wrote.

“If the human does not understand the assumptions, limitations and consequences of the system, they are not meaningfully in the loop.”

Strong wrote that the introduction of synthetic data models could create a “verification challenge” for the industry.

“The more fluent and coherent an output appears, the easier it becomes to mistake plausibility for proof,” he said.

“That places a renewed burden on the profession to define when synthetic evidence is acceptable, when it is not, and what standards of validation are required before it becomes decision-grade.”

The full report can be read here.