The enterprise AI economics problem no one owns.

Ask a large organization what it spends on AI and you will get five answers. Licensing knows the Copilot seat count. Cloud finance knows the metered Azure consumption. The program office knows the delivery labor. The platform team knows something about quality. HR or a champions network knows something about adoption. Each is right about its own column and nobody reads the row.

The row is where the money is. A seat that is paid for and idle is a licensing fact and an adoption fact at the same time; it only becomes a decision when the two are on one page. Metered consumption that grows in a department with no measured lift is a cloud-finance fact and a quality fact; separately they look like growth and like noise. Together they are a question.

This is why the first thing we count is the estate as one row: seats, agents, credits, spend, and the never-nudged control group beside them. The score matters less than the fact that every input on the page is traceable to a Microsoft report and sits next to the input it should be read against. Once the row exists, ownership follows — someone has to sign the one costed move each month, and that person becomes the owner the economics never had.

The practical test for any AI program: can one person, in one sitting, see license cost, metered cost, adoption and measured effect for a single department, from the same run? If not, the budget is a forecast and the outcome is a story.

Published 14 September 2026 · All insights