The organization expected to find a long list of AI tools. It did not expect to find the same work repeated across teams, paid for through several channels, and hidden behind personal accounts.
The useful signal was not the number of tools. It was the relationship between people, workflows, spend, and outcomes.
What appeared once the map was complete
Approved products accounted for only part of active AI usage. Browser tools, embedded assistants, APIs, and team-level purchases made the real footprint much larger than procurement records suggested.
Several departments were also solving the same problem in parallel. The overlap created duplicate spend, but it also exposed workflows worth sharing.
What mattered more than tool count
Leaders needed to distinguish experimentation from repeatable work. Usage alone could not answer that question. The stronger signals were recurrence, team adoption, business context, and whether the result changed how work was completed.
The next step was allocation, not prohibition
The response was not to block every unapproved tool. It was to reduce obvious duplication, protect sensitive workflows, and move strong patterns into approved defaults that other teams could reuse.
