The first three parts of this series built a single argument in stages. Part 1 established that AI implementation fails as an organizational problem, not a technical one, and introduced four theories to explain why. Part 2 located the decisive moment before launch, in the pre-implementation window where organizational support is cultivated or squandered. Part 3 mapped targeted interventions across the implementation lifecycle. Each part moved closer to the same uncomfortable truth, which this article confronts directly: once AI reaches the people who use it, the organization no longer controls what happens next.
Most implementation scorecards measure adoption: license activation, login frequency, usage volume. These numbers describe whether a tool is being touched. They say nothing about how—and the “how” is where success is actually decided.
Adaptive Structuration Theory, introduced in Part 1, names this gap precisely. AST distinguishes between the structures a technology is designed to carry and the appropriation of those structures by the people who use it—the concrete moves through which employees take a system and make it part of real work [1]. Crucially, DeSanctis and Poole argued that appropriation can be faithful—consistent with the designers’ intent and the underlying “spirit” of the tool—or unfaithful, where the technology is bent toward purposes its designers never envisioned [1]. The same AI assistant can be appropriated faithfully by one team as a drafting partner subject to human review, and unfaithfully by another as an unverified authority whose outputs are pasted directly into regulated work.
The decisive insight is that neither outcome is visible in its adoption of metrics. Both teams “use” the tool. Only one is using it in a way that the organization can stand behind. Appropriation, not adoption, is the variable that matters—and it is largely invisible to the dashboards most organizations rely on.