Where AI actually helps — and where it still needs a researcher

Useful, not magic

AI has made parts of research genuinely faster: drafting questionnaires, coding open-ends, summarizing interviews. It hasn't replaced the hard parts — deciding what to ask, who counts as the right respondent, and what a finding means for a client with real money on the line.

Where we use it

• Questionnaire drafts: Good first drafts from a brief, always rewritten by a researcher before field.
• Open-end coding: Theme suggestions at scale, with humans reviewing anything ambiguous.
• Forecasting: Trend models that are only as good as the input data — garbage in still applies.
• Quality control: Flagging speeders, straight-liners, and bot-like patterns during fieldwork.

Keep humans on the hook

The failures we've seen all look the same: nobody owned the output. AI drafts the survey, AI codes the answers, AI writes the summary — and nobody who understands the market reads it end to end. Assign an owner to every AI-assisted step or don't use it for that step.