The five-gate AI value review: a steering checklist for COOs

McKinsey's COO100 survey covers 101 senior operating executives at organizations with at least $1 billion in revenue, fielded in June-July 2025. About two-thirds describe AI as still in exploration or targeted implementation, about one-third as starting to scale, and only 2% as fully embedded across operations.

Intent is not the constraint. Of the respondents, 93% plan to spend more on digital and AI over the next five years, and almost one-third plan to spend at least 5% of cost of goods sold. COOs also plan to scale a focused portfolio of 5 to 12 use cases by 2030. The constraint is how decisions about those use cases are made.

Nearly 60% lack clear targets for measuring AI value. That matters economically because the same survey finds that where targets exist, nearly two-thirds of companies meet or exceed them. Accountability is associated with results. The causality is not proven by a survey, but the pattern is consistent with basic incentive theory: what is not measured is not managed.

Barriers confirm the shop-floor reality. Half of respondents cite cultural change as a major barrier, almost as many cite reskilling, 46% cite data or IT/OT limitations, and a quarter struggle to build reusable, scalable applications. A July 2026 study by the Manufacturers Alliance Foundation with Roland Berger (100+ manufacturing leaders, about 40 interviews) adds that nearly two-thirds needed significant data clean-up before launching AI initiatives.

McKinsey's own steps for COOs are to own the AI value agenda with KPI-linked targets and ring-fenced funding, redesign production end to end, build scalable technology on an interoperable backbone, and drive adoption through reskilling and an operating model. The Cardan-AI method below turns these into a recurring committee routine, applied use case by use case. It is Cardan-AI's own synthesis, not a McKinsey framework.

Gate 1, target: a KPI, a size of change and a date, agreed before the build. Gate 2, owner: one operations leader, not the IT function, who signs the result. Gate 3, data: is the data usable now, and who funds the clean-up. Roland Berger's advice is to solve problems with the data available while strengthening the foundation, because perfection should not be the goal.

Gate 4, funding: a ring-fenced budget released stage by stage, so that a failing use case stops costing money. Gate 5, adoption: which roles, routines and skills change on site, and who is trained. The committee ends each review with one of three decisions on a fixed date: stop, scale or fix.

Limits: the sample is 101 large organizations, mostly in North America and Western Europe, and the survey was fielded in 2025, so mid-sized suppliers and 2026 conditions may differ. Executives self-report maturity. A five-gate review adds governance cost; apply it to use cases above a materiality threshold, not to every experiment, and keep a small explicit option budget for exploration.