Compare similar work with and without AI, keeping the category, time window, quality measures, and cost definitions consistent.
A useful comparison follows the work beyond the AI-generated draft.
Oximy compares the same kind of finished work with AI and without it, in one window.Maintenance fixes merge 9 hours sooner at $8 less per change, and the change failure rate did not rise.
The comparison supports taking maintenance fixes to the next eligible team with the same definitions attached. The ninety-day window on customer bug reports is still open, so the customer result stays under measurement.
Attribution supports an inspectable comparison; it does not by itself prove that AI caused the change.
Define the comparison before interpreting the difference.
Use the event the business already accepts as completion. A draft, a suggestion or an agent run is an input to the workflow unless it is itself the agreed finished unit.
Keep category and scope consistent: maintenance changes are not compared with feature work because both reached the same repository. Complexity, open windows and shared costs are recorded beside the result rather than folded into it.
Only when the method supports it. A before-and-after comparison can also reflect changes in staffing, complexity or process, so those explanations stay visible beside the finding.
