Most organisations have run an AI pilot. Far fewer have an AI capability in production, owned by a team, measured, and trusted by the business.
Where pilots stall
A pilot proves a model can do something interesting on a slice of data. Production asks a different question: can this run every day, on real inputs, with someone accountable for the result. That gap rarely comes down to the model. It comes down to the operating model around it.
What to fix first
- Ownership. Name the team that runs the capability once the pilot ends. Without an owner, nothing reaches production.
- Data path. Wire the model to the live data source rather than an exported sample. The export is why the pilot looked easy.
- Guardrails. Decide what the system may do on its own and where a human signs off. Put that checkpoint in the product, and keep it out of a policy note.
- Measurement. Agree one metric the business already cares about, and track it from day one.
The shift
Production is first a question of accountability. Treat the operating model as the deliverable, and the model becomes the straightforward part.


