The decision
The initiative could have continued as a technical exercise. That was not enough. Moving toward scale would have required more review, exception handling and human validation than the expected outcome could justify.
I stopped the initiative before adding more technology to a weak value case. The relevant outcome was not a deployed agent. It was avoiding an operating workflow whose controls would cost more than the problem it solved.
The test I used
- Value: is the problem important enough to change how work is done?
- Evidence: does the pilot support the claims needed for a scale decision?
- Validation: can people review outputs at a cost proportionate to the value?
- Ownership: is someone accountable for exceptions, monitoring and the final decision?
Why this matters
Deployment is not adoption, and adoption is not transformation. A responsible AI programme needs explicit stop criteria before a pilot begins. If evidence, ownership or control cost does not support the next step, stopping is part of implementation discipline.
This note contains no client, sector, project duration or confidential operating detail.