AI strategy for hospitals: the 5 mistakes boards make

DR GPT™ perspective

Most failed deployments I see were decided in a procurement meeting, not an engineering one. These are the five ways that happens.

The first mistake is the expensive one: treating AI as an IT project. It gets delegated to a function that cannot change clinical workflow, cannot change staffing, and cannot change how a decision gets made. Then everyone is surprised when the tool goes in and nothing changes. AI is an operating model question that happens to involve software.

The fifth mistake is the one that hides for years: assuming adoption equals success. Login counts are the easiest number to produce and the least informative. A tool used daily by clinicians who have quietly stopped reading its output is worse than a tool nobody opened, because it generates a false sense of coverage and somebody downstream is trusting it.

If a board fixes only one of the five, fix the fourth. Excluding clinicians is the mistake that causes the other four. A board with a practising clinician in the room when the decision is made rarely treats AI as an IT project, rarely buys before defining the problem, and rarely mistakes usage for value. Bring the person who will have to live with it into the meeting where it is chosen.

One question I would add to any board packet: what would make us turn this off? If nobody can answer that, the organization has bought something it cannot withdraw, and reversibility is the only real protection at this stage of the technology.

Originally published on KevinMD. Read the full essay at the link below, or see how I build these arguments for a room as a speaker.