Pilot-to-Production Checklist

Did the pilot test what was uncertain, or demonstrate what was already known?

• The pilot's question was written down before it startedAnd it was a question somebody could have answered no to.
• It was run against the hardest case, not the easiestThe oldest machine, the messiest data, the shift that does it differently.
• The answer is recorded, including what did not workA pilot with no negative findings did not test anything.

What this gate catches: The demonstration pilot. It runs on the newest asset with clean data, everybody is pleased, and none of the risk has been touched. Scaling one of these is how a project discovers its real problems at ten times the cost.

What happens to the cases the system cannot handle, and who does it?

• The exception rate has been measured, not estimatedFrom the pilot's own records, over a representative period.
• There is a defined route for an exception, with an ownerNot "it falls back to the old process" — a named person and a named queue.
• The exception workload is inside somebody's capacityConfirmed with that person and their manager, in hours per week.

What this gate catches: The saving that never arrives. Exceptions turn out to be a quarter of volume, they land on the same people as before, and the business case quietly stops being true while everyone congratulates the build.

Has it been tested against failure, not only against success?

• Tested with a source unavailableNetwork down, machine off, upstream system in maintenance.
• Tested with bad input, not only missing inputA sensor reading nonsense is more dangerous than one reading nothing.
• Recovery is automatic, and has been observed recoveringNot documented as automatic. Observed.
• Someone is alerted when it stops, and that alert has been testedIncluding out of hours, to a person who is actually awake and responsible.

What this gate catches: Silent failure. The system stops updating on a Friday, the dashboard shows the last known values, and nobody notices until Tuesday because a stale number looks exactly like a current one.

Who runs this on the Monday after handover?

• A named individual owns it, and their manager agreesIn writing, with time allocated, before go-live rather than after.
• There is a second person who could keep it runningA single point of knowledge is a countdown, not an arrangement.
• Operating budget exists, separate from the buildLicences, support, and the owner's hours, in an operating line that survives the project closing.
• The owner has run it, unaided, before acceptanceIncluding the routine intervention it will need.

What this gate catches: The most common failure in this list, and the least technical. The system works, the project closes, and eighteen months later it has been worked around because nobody was funded to keep it.

When the process changes next year, what does it take?

• You hold the source, the configuration and the credentialsHeld by the company, not by an individual and not only by the vendor.
• There is a test environment, or a defined way to test safelyChanging a live integration with no way to test it is how a good system becomes a frozen one.
• The change is documented as a procedure somebody has followedAt least once, by the owner, before acceptance.
• Another integrator could pick this upAsk honestly. If the answer is no, that is a commercial position you have taken without deciding to.

What this gate catches: The frozen system. It works and nobody dares touch it, so the process grows around it instead, and within two years the workarounds cost more than the system saves.

How will you know in six months whether this was worth doing?

• A baseline was captured before go-liveWithout it, the improvement is unprovable and the next business case is harder to make.
• The measure is defined, and somebody produces it on a scheduleNamed metric, named owner, named interval.
• There is a review point with a decision attachedContinue, extend, or stop. A review with no possible outcome is a status update.

What this gate catches: The unprovable win. Everyone believes it helped, nobody can demonstrate it, and the next project has to be justified from scratch against a finance function that has learnt to discount the claims.