Gartner’s May 2026 forecast put worldwide AI spending at $2.59 trillion this year, up 47%, with infrastructure more than 45% of the pile. McKinsey’s 2026 survey, as reported in September, still had only 37% of respondents attributing any EBIT to AI, and about 6% as high performers. Individual productivity is not the same as a cheaper close. This guide is how to stop mixing those two.
Pick a workflow with a finish line
Good: monthly close commentary, support macros, bid/no-bid memos, invoice exceptions. Bad: “make the company more AI-native.” If you cannot name the artifact that is done, you cannot measure time or cost.
Write the current path: who starts it, how long it takes, where it fails, what “done” looks like. That is your baseline. No baseline, no after.
Instrument three costs
- License or token spend for that workflow only — not the company-wide copilot bill.
- Human review time after the model drafts. Review is the cost that hides in “we saved time.”
- Rework when the draft invents a date, a customer, or a covenant.
If the only metric is seats, the organization will buy seats.
90 days, one owner
| Cost | Count this | Do not count this |
|---|---|---|
| Tokens or license | Spend on this workflow | The company-wide copilot bill |
| Review | Minutes after the draft | “We feel faster” |
| Rework | Invented dates, names, covenants | A demo that went well |
Name a manager whose bonus includes cycle time or error rate for that workflow. Put the date on the calendar now. Gartner said in September that only 22% of organizations had scaled AI across multiple business units. You do not need to be in the 22%. You need one number that moved.
If nothing moved, cut the overlapping copilots and keep the model on the one path that has an owner. That is not anti-AI. It is how you keep the next Gartner-sized invoice from becoming the strategy.
Sources
- Gartner AI spending forecast, May 19, 2026
- Gartner scaling survey, Sept. 1, 2026
- McKinsey State of AI 2026
FAQ
Why one workflow, not the whole company?
Because a company-wide copilot bill has no definition of done. One artifact you can date is a number. A stack of seats is a vibe.
What counts as review time?
The minutes after the model drafts, before the work is shipped. If a person still has to catch invented dates, that time is cost.
What if nothing moved in 90 days?
Cut the overlapping copilots. Keep the model on the one path that has an owner. Then measure again.
Do we need to be in Gartner’s 22% who scaled AI?
No. You need one number that moved on one workflow.