Why your pilot never shipped
The pilot worked, everyone applauded the demo and nothing went into production. That is rarely a model problem. It is the absence of five decisions nobody made.
Apply an operational readiness checklist to any pilot of yours and say, with evidence, what is missing for it to ship or why it should be closed.
Sign up once to unlock this course
Course 01 is open. For courses 02 to 05 we ask for six fields. It is a single signup: done once, valid for the whole track. It is not a free diagnosis and it does not trigger automatic sales contact.
A demo and an operation are different things
A demo has to work once, with chosen data, in front of people who already want to see it work. An operation has to work every day, with the data that shows up, and fail predictably when it fails.
Most pilots die because they were built for the first scenario and later judged by the second. The team presents a case that passed, someone asks what happens when the document arrives illegible, and the conversation ends in a promise to adjust.
There is no owner, there is an enthusiast
A pilot usually has a showcase sponsor and an enthusiast who makes it happen. Neither is an owner. An owner is whoever answers for the number when it fails to appear, holds budget to sustain the operation and has authority to change the surrounding process.
Simple test: if the operation stops running on a Tuesday, who gets the call? If the answer is "the technology people" or "it depends", there is no owner.
The completion criterion was never written
An operation needs to know when a unit of work is finished. The document was checked, the lead was qualified, the proposal is ready to send. Without that criterion there is no way to measure volume, no way to measure error and no way to pay for outcome.
A pilot survives without it because assessment is subjective: someone looks at the output and says it looks good. An operation does not survive that way.
The exception has no route
Every real flow produces cases outside the standard. In the pilot, the exception goes back to the enthusiast, who handles it manually and records nothing. In an operation, the exception needs a destination: who receives it, within what time, with what information, and what happens if nobody answers.
Nobody can say what percentage of cases required human intervention last month. If it is not measured, it was not designed.
Cost per unit was never calculated
A pilot runs at low volume, and the cost looks negligible. In production, that same per-case cost multiplied by real volume can invert the economics. When the invoice arrives before the arithmetic was done, the operation is suspended by finance, not by technology.
You need to know the cost of a simple case, of a case that required reprocessing and of a case that went to human exception.
Integration was left for later
A pilot that requires someone exporting a spreadsheet and pasting the result into another system is not an operation, it is manual work with an extra step. Until the output enters the system where the decision happens, the gain stays trapped in the demo.
Closing a pilot can be the right decision
Not every pilot deserves production. If the volume is small, if the required data does not exist, if the surrounding process is about to change, closing preserves budget and credibility. The mistake is not closing, it is keeping something in a permanent state of almost.
Operational readiness checklist
Assess one of your pilots across the five points. Mark each line with what already exists in writing, not with what is intended.
| Decision point | Exists today? (yes, partial, no) | Who answers |
|---|---|---|
| Owner with budget | ||
| Completion criterion | ||
| Exception route | ||
| Cost per unit | ||
| Integration into the final system |
A pilot with two or more points marked "no" is not ready for production, and pushing it consumes the budget the next initiative will need.
Basis
The five points are not management theory. They are the questions that arrive when an AI-assisted operation enters production and someone has to answer for it in a results meeting.
When the checklist becomes an investment decision
If the pilot has an owner, a criterion and volume, but lacks architecture and integration, the problem is execution. If it lacks an owner and a criterion, the problem precedes the technology. Prumo Discovery separates the two before any building starts.