The questionnaire nobody wanted to fill in

When people imagine AI in a large engineering organisation, they picture something dramatic. In reality the most valuable thing I have built started with a questionnaire.

Project teams were already answering questions about every initiative: how much it would improve collaboration, how complex the integration would be, what it would cost. The answers lived in spreadsheets, were reformatted by hand, and arrived in leadership meetings weeks late.

Automate the boring middle

The first win was not intelligent at all. Responses flowed straight from the form into an automated pipeline that cleaned and shaped them, ready to be scored. Nobody had to copy and paste anything again. That alone earned more goodwill than any model could have.

Make the picture obvious

Next came a simple quadrant: benefit on one axis, complexity on the other. Quick wins top left, strategic bets top right. Executives understood it in seconds, and that matters more than any clever visual.

People adopt the tool that makes their next meeting easier. Design for that meeting.

Then, and only then, add AI

With clean, trusted data in place, the AI layer had something to stand on. It predicted alignment scores, surfaced the drivers behind them, and read free-text answers to flag risks a human might skim past. Because the inputs were familiar, people trusted the outputs.

If your AI pilot is stuck, look upstream. The fix is usually a form, a pipeline, or a picture, long before it is a bigger model.