Challenge
A global market research firm designs surveys for companies that live on research: retailers, branding agencies and other businesses that need a steady flow of insight about their markets and customers. Every survey is built to a careful set of rules: how questions are worded, ordered and routed, how bias is avoided, how answers will be analysed. That expertise is what clients pay for, and it lived largely in the heads of experienced researchers.
So every new survey was a craft project. Designs were rebuilt from scratch or adapted from old ones by hand, the firm's rules were applied as well as each person remembered them, and the lessons from past projects, what clients found useful and what they did not, were rarely fed back into the next design.
Solution
We built an AI layer that designs surveys the way the firm's best researchers would. It works from the firm's own rules, written down as rules rather than habits, and connects directly to the survey systems through their APIs, so a design goes from brief to a working survey without being rebuilt by hand.
The layer also learns. Client feedback on every project, what worked, what was confusing, what produced the insight the client needed, is condensed and fed back into how the next survey is designed. Each project makes the next one better, which is how the firm's expertise starts to scale beyond the people who hold it.
Execution
Turning expertise into rules
The first step was to write down what experienced researchers do instinctively: the conventions for wording, the order questions should come in, how routing should work, what to avoid. Those rules became the foundation the AI designs against, and the standard every generated survey is checked against before a researcher sees it.
Connecting to the survey systems
Rather than producing documents that someone then rebuilds, the layer works through the APIs of the firm's survey systems, so a design arrives as a working survey ready for review. That removed the slowest and most error-prone step in the process.
Closing the loop with client feedback
Every project ends with a view of what worked. We built a process that captures that feedback, condenses it into lessons and attaches them to the kinds of survey they apply to, so the next design for a similar brief starts from what was learned, not from a blank page.
Researchers stay in charge
The AI drafts; researchers decide. Every survey is reviewed before it goes to a client, and the researchers' changes are themselves a signal the system learns from.
