We've deployed AI. Now what? Designing how People and AI actually work together

Most organisations I work with have already rolled out AI. Licences are bought, pilots are done, an AI policy sits on the intranet – and yet people either ignore the tools or trust them far more than they should. The problem is rarely the model. More often it is the way collaboration between people and AI was, or wasn't, designed.

In this talk I will show what behavioural science tells us about how people actually make decisions when AI is in the loop. We will look at why transparency notices required by the AI Act lose their effect after a few exposures, why information on its own rarely changes behaviour, and how over-reliance and quiet rejection grow out of the same design gaps. Drawing on enterprise projects, I will walk through FASCAI, a practical way to decide which tasks can be fully automated, where AI should only suggest, and where a human needs to slow down and verify. I will also show how to calibrate trust instead of simply trying to increase it.

You will leave with a behavioural lens for looking at your own AI product or internal tool, and a handful of questions worth asking before the next rollout.