Most AI strategies released in the past year read like procurement calendars. They list platforms, timelines and spend categories but never name who can override a model output when it conflicts with commercial reality.
This creates a predictable pattern. The strategy gets signed off, the first licences arrive, and operational teams discover they still need three layers of approval to act on anything the system flags. Speed promised in the deck evaporates at the first exception.
The real constraint is not model performance. It is the refusal to redraw authority lines. Procurement owns the contract, risk owns the policy, and the business unit still owns the P&L yet cannot change the process without another steering committee.
You see the same pattern in APAC enterprises that declared AI transformation programmes last financial year. Twelve months later the only measurable output is a larger slide library and another round of vendor briefings.
Fixing it requires one structural change: give named roles explicit rights to accept or reject model-driven recommendations inside defined risk thresholds. Everything else is just more theatre.
Without that shift, every new roadmap simply accelerates the existing approval bottlenecks rather than removing them.