Why we don’t run a pilot.

The conventional opening for a project like this is a discovery workshop, followed by a scoped pilot, followed by a business case. It is a sensible-looking sequence and it has one structural flaw: it asks an operator to specify a system whose value nobody can see yet — including us.

A workshop produces a list of what people can articulate in a room. That is a genuinely useful artefact, and it is not the same as what the operation is actually losing. The things that cost the most are the ones nobody thinks to mention, because they do not feel like problems. Orders keep arriving. Every run passes. The feed is still there.

A pilot has the same flaw with a deadline attached. Pick a narrow use case, prove it in ninety days, and you have proven a narrow use case. The compounding part — the bit where the system knows what a normal winter looks like here, or which engagements always run over — is precisely what a ringfenced pilot cannot demonstrate, because it has not had time to learn anything.

So we start the other way around. The platform goes in and connects to what is already there: the logs, the time records, the receipts, the schedules. It observes. In the first week it can already compare this season, this month, this run against the last few, which is usually enough to surface something nobody had noticed.

Then, every month, there is a conversation about two things: what it has learned about how the business runs, and where it looks like it will pay to build next. Those are not our opinions. They come out of the operation’s own records.

This has a consequence we are comfortable with. We do not write the business case — the operation does. If, after observing for a month, there is nothing worth building, that is a real answer and it is cheaper for everyone to have it early.

It also sets the standard for what counts as progress. Not a slide deck at the end of a discovery phase, but something an operator can check: a claim, the records it came from, and a decision they can make six weeks earlier than they otherwise would have.

The rest is patience. A system like this is worth more in month six than in month one, because by then it knows things. There is no way to compress that, and any process that pretends otherwise is selling the wrong thing.