Parts one and two of this series established two things: most of AI implementations fail because organizations treat them as technical deployments rather than organizational transformations, and the pre-implementation window—the period before a system goes live—is where the employee support that determines whether AI succeeds or stalls must be deliberately cultivated. Both are true. Both are also insufficient.
Pre-implementation work, however rigorous, covers only the front end of a much longer process. AI implementation is not an event—it’s a lifecycle. Each phase of that lifecycle carries distinct human and organizational risks, and demands deliberate, targeted interventions. The question isn’t whether OD practitioners and change leaders should be involved in AI implementation. It’s when and how.
Part 1 introduced four theories that explained why AI implementation fails: Organizational Support Theory (OST), sociotechnical Systems Theory (STS), Adaptive Structuration Theory (AST), and Actor-Network Theory (ANT). These frameworks aren’t solely diagnostic tools; they function as maps within each implementation stage that points toward a specific class of organizational action.