The Situation
A market research agency runs a yearly industry survey for its client, a logistics software company. The results usually ended up in a PDF report that people read once. This year the client wanted something people could actually use: explore the results, compare their own business against their peers, and ask questions in plain English.
That sounds like "add a chatbot". It was not:
- The data arrived as around eight separate files, some still being cleaned when we started
- How to read the data properly lived in the researchers' heads, not in any document
- If the AI made up even one number, the credibility of the whole study was gone
- In a small industry, a few filters could reveal exactly which business gave which answer
- It was also a sales tool, so it had to sound like the brand and never mention competitors
Ordinary AI tools work things out on the fly and sound confident even when they are wrong. For published research, that is the one thing you cannot allow.
What I Built
Together with my build partner, I built the complete system behind the research insights and the simulator, working from the agency's finished research. Not a dashboard with a chatbot on top. The dashboard people see is the last and simplest piece. Everything underneath it is there to make sure what people see is right.
The Brain
We sat down with the agency's researchers and captured how they actually read data: what they check first, which comparisons mean something, and when a result is too thin to trust. The system now follows their rules every time. Every number is calculated from the real survey answers, and every one can be traced back to where it came from.
The AI Assistant
People can ask questions in plain English and get clear answers in the brand's voice. But the AI is only allowed to talk about numbers the system has already calculated. Before any answer is shown, every number in it is checked. If one does not match, the person gets an honest "I don't have that" instead of a guess.
The Simulator
Business owners enter their own answers and see exactly where they sit against their peers. Then they can play "what if": change an answer and see where that would put them, or ask what it would take to reach the level of the businesses doing better. It shows real positions against real peers, never promises, and it runs on their own device, so nothing they type is sent anywhere.
The Privacy Protection
No one who took part in the survey can be identified, even by someone combining filters on purpose. Before anything is published, the system tries to work out private answers from the public results, the way a determined snoop would. If it succeeds, nothing is published.
People in Control
Nothing the AI writes for the report is published until a researcher has approved it. Every important decision needs a human sign-off. And the system comes with an operating manual and can diagnose its own faults, so the client can hand it to their own team to run.
What It Delivers
One trusted source now feeds everything: a public benchmark tool, an AI assistant that answers questions about the results, the simulator, a branded report, and a personal summary business owners can download as a PDF showing how they compare with their peers.
The Results
What Comes Next
Next year's survey will not need a rebuild. The new results go in, and the same brain, checks and AI assistant work on them straight away. Year-on-year comparisons open up as more years are added. And this goes well beyond surveys. Any business sitting on valuable data, whether that is customer records, sales history or years of reports, can have an AI that answers questions from it, shows where every number came from, and never guesses.