The week in brief — AI's value lies around the model, not inside it

Four lessons this week, one thread: AI's value lies around the model, not inside it. MD Anderson spent $62M on a system that was never used, for lack of scope control and integration with existing tools; 60% of organizations are betting on digital twins but only 13% get real value from them, and the gap comes down to data discipline; 42% of regular frontline AI users save a full working day a week, which most companies let evaporate for lack of assignment; and the companies that extract measurable value from AI apply the 10-20-70 rule, with 70% of the effort going to people and processes. The model is almost never the weak link: the scope, the data, the freed-up time and the teams are. For every AI project, write these four lines before signing: the scope and its quantified exit criterion, the data feeding it, the assignment of the hours freed, and the share of budget devoted to the teams.

Sources: The Cancer Letter, University of Texas System audit, February 2017; Capgemini Research Institute, "Digital Twins", 2022 (1,000+ organizations); BCG, "AI at Work", June 2026 (11,749 employees, 14 countries); BCG, "Where's the Value in AI?", October 2024 (1,000 executives, 59 countries).