Hybrid Research

Why this matters for your decision

Each method answers a different question. Synthetic respondents quickly answer what is plausible. Surveys and qualitative research explain how people think and why. A behavioural test shows whether people act in a realistic offer situation. Problems arise when a method is expected to answer a question it was not built for, for example when a synthetic ranking is treated as proof of demand.

In the Journal of Marketing, Arora and colleagues describe language models as collaborators in the research process that complement human data and human judgement. That is the core of hybrid research: clear roles instead of replacement.

Roles in the process

Synthetic methods for breadth: hypotheses, segments, objections, initial pre-selection from many ideas. Surveys and qualitative research for understanding: the target group's language, motives, barriers, details of the concepts. Behaviour for the investment question: which of the remaining variants, which price, which claim triggers purchase intent?

After the behavioural test, the circle closes: qualitative research explains why a variant won, and the results sharpen the next round of exploration.

For a pricing question, this can mean: synthetic methods collect possible price anchors and objections, a survey checks which price levels are understandable, and the behavioural test measures at which price level measured purchase intent drops markedly.

In practice, a simple rule helps: the more expensive and the harder to reverse a decision is, the more it should rest on observed behaviour. Early, inexpensive steps can take more synthetic input.

Example

A retailer is planning a subscription for household goods. A language model generates 30 offer variants and pre-sorts them down to 8. A short survey of 400 people shows that flexibility matters more than a discount, and narrows them down to 4 variants. A four-week behavioural test measures with which variant and at which price real people want to sign up for the subscription. Two rounds of interviews then clarify what makes the winning variant convincing.

Distinction

Hybrid research is more than synthetic boosting, in which real and simulated responses are mixed in one dataset. It separates the roles of the methods instead of mixing their data. It differs from mixed methods in classic market research in that synthetic methods and behavioural measurement are explicitly planned in.

Limits

A hybrid process is only as good as its transitions. If AI-based pre-selection filters out unusual ideas too early, even the best behavioural test can no longer test them. So deliberately carry forward a variant that deviates.

The behavioural test itself answers whether people act, not always why. It measures purchase intent, not a sale. In this process, Horizon takes on the behavioural measurement and is complementary to surveys and research institutes.