The week in brief — with AI, everything depends on what you choose to look at

Four nuggets this week, one common thread: with AI, everything turns on what you choose to look at. Wimbledon handed its 300 line judges to AI, then found itself with no backup the day the system was switched off by mistake: automating a decision moves the safety net, it does not remove it. A predictive AI, for its part, loses accuracy as the world changes, and without warning: this is “model drift”. An AI that is installed is never an AI that is finished; it has to be monitored over time. And when AI adoption among French small firms doubles in a year yet ranges from 9% in farming to 41% in tech, it is your own sector you should measure against, not the national average.

The last lesson sheds light on the others: across 10,000 businesses, only 5% consider their data ready for AI, when it is the data, not the model, that decides whether AI scales. Before investing in a tool, it is worth writing down what you will monitor afterward: a fallback for when the AI gets it wrong, an accuracy threshold reviewed regularly, data prepared upfront, and your own sector as the benchmark.

Source: YPrtnrs Institute nuggets, week of October 19–22, 2026, drawing on AP (Wimbledon), FDA, France Num 2025 (DGE), and Dun & Bradstreet.