Vendor dashboards love to report adoption as the percentage of licensed users who logged in at least once. By that standard, most enterprise AI deployments look spectacular. By every standard that matters, they don't.
Real adoption is a pattern, not an event. It's whether a developer uses Copilot on the fifth task of the week as often as the first. It's whether a sales team opens their AI-drafted emails before sending them, or just ships the template. It's the distribution of usage over time, across features, and across the population that was supposed to benefit.
When Umbral Intelligence measures adoption, we measure four things: reach (who's touched the tool), depth (how many features they've used), persistence (are they still using it in month six), and integration (is it changing how the surrounding work gets done). A tool can score beautifully on reach and catastrophically on the other three. Most do.