Plenty of businesses we meet have already tried something. A pilot was run, it worked, everyone was pleased, and it is not in use today. Nobody can quite say when it stopped.

Having seen enough of these, the causes are consistent. Only one of them is technical.

01 It was scoped to impress rather than to run

A demonstration has to be interesting. A production system has to be dull. It has to handle the ugly record, the supplier who sends a PDF inside a PDF, the month where volume triples, and the Tuesday nobody is watching.

When a pilot is scoped to be shown to a board, it optimises for the impressive case. Every hour spent on that is an hour not spent on the boring cases, which are the ones that decide whether anyone can rely on it.

02 Nobody owned it after the demo

The person who championed the pilot is usually not the person who has to live with it. When it moves into operations, it arrives at someone with a full workload who was not consulted and has no stake in it succeeding.

They will not sabotage it. They will simply revert to the process they trust the first time it does something strange, and there is no mechanism to bring them back. Six weeks later it is unused, and nobody has made a decision to stop using it.

03 The exception rate was never measured

A system that handles eighty percent of cases sounds like success. Whether it is depends entirely on what happens to the other twenty.

If exceptions surface clearly, land with a named person and get resolved in a defined way, eighty percent is a strong result. If exceptions fail silently, or land in a queue nobody owns, then the eighty percent is worthless, because the team has to check everything anyway. They now have their old workload plus a system to supervise.

The exception path is not a detail to sort out later. It is the product.

04 It was measured on hours saved, and nobody could find the hours

This one is close to universal and it is largely our industry’s fault. Automation gets sold on cost reduction, so the pilot is measured on hours removed, and then someone has to point at those hours in a real business.

They are rarely findable. The work was distributed across six people in fifteen minute fragments. Nobody was made redundant, because nobody wanted to make anyone redundant, and the business is not obviously cheaper to run. Judged on the stated measure, the pilot failed, and it gets quietly shelved even where it was genuinely working.

Capacity is the honest measure, and it survives contact with reality. Orders processed per person. Quotes out the door in a week. Time to first response on an enquiry. Jobs completed without adding headcount. How much bigger this operation can get before the next hire becomes unavoidable.

Those are numbers a business can act on, and they are the reason to do this work. Automation that only makes an existing business slightly cheaper is a poor use of anyone’s money. Automation that removes the ceiling on what the business can take on is a different proposition entirely.

05 The data was not ready and nobody said so

The technical one. Product codes that do not match between systems. Customer records duplicated four ways. A field everyone agreed to use consistently in 2019.

Pilots run on a clean extract, which is how this stays hidden. Production runs on the live mess. Any honest scope prices the cleanup as part of the work rather than treating it as a surprise discovered in week five.

What a production ready engagement looks like

Five things, agreed before anything is built.

A written definition of done that describes the system running in daily operations, not a demonstration. A named owner in operations who was involved in scoping and has some say in the outcome. An exception path with an owner and a resolution time. A measure agreed in advance, expressed in capacity rather than hours. A handover that includes documentation and a named person who can maintain it after we leave.

None of this is exotic. It is the difference between a project that gets used and one that gets talked about.

One more thing

If a pilot is not worth putting into production, it is usually not worth running. Two processes taken properly into daily use will teach a business more, and earn more, than eight pilots that impressed a meeting.

Tried something that did not stick? Tell us what happened. The diagnosis is usually quick and it is often not the technology. Start a conversation.

Related reading: A practical readiness checklist and what this work actually costs.