The pilot that met a normal Tuesday
A pilot can handle the happy path and still fail on ordinary production work. Four controls close the gap between a demo and a system that can be operated.
A controlled demo covers the happy path. Production also needs exception handling, monitoring, ownership, and documentation.
A pilot can handle the happy path and still fail on ordinary production work. Four controls close the gap between a demo and a system that can be operated.
Clean examples hide production exceptions: a document with a missing field, a lead whose message breaks the template, or a source export that changes shape without warning.
A pilot that has not defined what happens on the weird cases has not defined the system. The correct behavior on an exception is to stop, hold the work, and hand a person the context. Building that stopping behavior is most of the build.
The dangerous failures stay quiet: a scheduled job stops running, a data feed goes stale while reports keep generating, or a queue backs up without an alert.
I have had a job fail quietly for two weeks before I caught it. I run monitoring as its own workflow now: a system whose only purpose is to watch everything else and hand me the one thing that broke before it becomes a problem. During a pilot everyone is watching. Once the pilot ends, nobody is, so the system has to watch itself.
A production workflow needs a named person who reviews its output, feeds corrections back into the rules, and decides when it is wrong. Without that owner, corrections and exceptions accumulate.
The handoff standard includes a runbook: what the workflow does, what it needs, and what to do when it stops. Another person should be able to operate it from that record.
Every part of my own stack has died at least once. The brief generator crashed. The clip pipeline shipped garbage through its first several versions. The reason it all runs now is the layer around the tools: exception handling, monitoring, ownership, documentation. That layer does not demo well, which is why pilots skip it and why they stall.
If you have a stalled pilot, the audit maps what exists, what is missing from that layer, and whether the workflow is worth finishing.
Describe what it was supposed to do and where it stopped. The written assessment comes back with what is missing.