The best automation idea in your business is already in the building

A technology supplier arrives with a deck about what their platform can do. Somewhere around slide nine there is a diagram showing your business as three tidy boxes, and the assumption underneath the whole presentation is that you already know where your opportunities are, and simply need the tooling to address them.

You do not know. That is not a criticism, it is nearly universal, and there are good reasons for it.

Broken processes do not stay visible. Someone invents a workaround, the workaround becomes routine, the routine becomes muscle memory, and within about eighteen months nobody perceives it as work at all. It has become part of what Thursday is. Ask that person what takes up their time and they will not mention it, because you asked about work and this is just Thursday.

So the supplier is offering to redesign a kitchen without ever asking what you cook.

The question that actually gets answers#

The move that works is to stop asking about technology entirely. Do not ask where AI could help, because that question demands the answerer already understand what AI does, and it produces either silence or somebody repeating something they read.

Ask instead: what part of your week makes you want to lie down?

That one gets answers immediately, in detail, with feeling. People know exactly what their week costs them. They have simply never been asked in a language that let them say so, and they have usually assumed that whatever it is must be necessary, since it has been going on for years and presumably someone decided it once.

I have come to think of this as water divining, which is what I have called it since long before any of this was fashionable. The sponsor asks for a report. What they need is to stop being surprised in Monday's meeting. Those are different problems with different solutions, and the gap between what people request and what they actually want is where most projects quietly die. Extracting the real requirement, testing it against everything you have watched break before, and playing it back until everyone in the room agrees they are describing the same reality, that is the actual job. The technology afterwards is comparatively simple.

The moment worth staying for#

Get the right mix of people in a room and something specific happens, usually about forty minutes in, once the polite phase has worn off.

Someone mentions their two hours a week checking data consistency, slightly apologetically, expecting it to be too small to matter. Then someone from another department says they do that too. Then a third person says they do a version of it, downstream, on the same data, to catch the errors the first two did not.

The room goes quiet. Not because a technology has been identified, but because a shape has appeared that nobody could see from inside any single department. Three people were solving the same problem independently, none of them knew, and the cost was invisible precisely because it was distributed. That recognition is worth more than any tool you buy afterwards, and it cannot be reached by a supplier working from the outside, because it lives in the gaps between the boxes on their diagram.

Cross-department is where the value hides. Within a department, the inefficiencies have long since been optimised or accepted. Between them, nobody owns the problem, so nobody has looked at it.

Sometimes the answer has no AI in it#

This is the part that makes the approach commercially awkward, and it is also the part that makes it honest.

A decent proportion of what surfaces in these conversations does not need AI. It needs two systems configured to talk to each other. It needs a form with three fields removed. It needs somebody with authority to say that the Thursday reconciliation can stop, because the condition it was guarding against was fixed in 2019 and nobody told the person doing it.

Those findings usually justify the whole exercise on their own. They are also, obviously, unsellable by anyone whose business is selling AI, which is precisely why you rarely hear about them from suppliers and why you should be suspicious of any discovery process that never once reaches this conclusion. A discovery exercise that always discovers a need for the thing the discoverer sells is not a discovery exercise.

What you should actually be left with#

The output of this ought not to be a shopping list. Shopping lists go stale, and the world will have moved by the time you have worked through it.

What you want is the capability to keep doing it. The people in the room should leave understanding not just what came out of the session, but how to recognise the next one when their business changes, which it will. Otherwise you have simply relocated the dependency: you used to not know where your problems were, and now you know where they were in March, and you need somebody external to tell you again in November.

The test is straightforward. Six months later, has anyone in the organisation identified an opportunity on their own and acted on it? If yes, the exercise worked. If they are waiting for the next visit, it did not, whatever the report said.

Meanwhile, the person rebuilding that report every Monday is still doing it, and has been for years, and has probably stopped mentioning it.

Go and ask them. It costs nothing, and I have never once seen it fail to turn something up.