Somewhere in your business right now, someone has built a routine. Each morning, before anyone else is in, they have Claude read the new invoices in a shared folder, pull out the supplier and the amount, draft the queries for the ones that do not match, and update a sheet. Nobody asked them to. They did it because it gave them back an hour, and they are slightly embarrassed about it, because it feels like cheating.

That routine is the most valuable thing in your AI programme, and you probably do not know it exists.

Individual change has a ceiling

Everything I see in businesses starts with individual productivity, and I would never talk it down. A team that uses the tools well gets hours back every week. Emails answered, documents summarised, the first draft of anything done before the coffee is cold. That is real and it compounds.

But it has a ceiling, and the ceiling is the process. If the Monday numbers still take a person to assemble, and that person is now assembling them thirty percent faster, you have a faster Sunday evening. You do not have a business that produces its Monday numbers by itself. The volume keeps rising, it keeps landing on the same lean head office, and every step change still needs a hire.

The step that matters is when a piece of recurring work stops being something a person does faster and becomes something the business does without them, in the company’s accounts, on a schedule, with a log. That is a system change rather than an individual one, and it is where revenue per head actually moves.

The promotion path

Here is the thing I have come to believe after watching this a few times: you do not get to system change by planning it top-down. You get there by promoting the routines that people build for themselves.

It works like this. Someone builds a personal routine, like the invoice one above. It runs under their login, on their machine, when they remember to run it. It is unaudited and slightly fragile and it is theirs. That is fine, because it is doing something more important than saving an hour. It is discovering the process. It is finding out which systems are involved, where the exceptions live, what the real steps are once the human hand-offs are stripped out, and whether anyone actually cares.

Then, at some point, the routine passes a test. I use three questions. Does more than one person want it? Does it touch a system of record, the accounting package, the point of sale, the customer list? Does it write to the outside world, an email to a supplier, a payment, a price change? Any yes means the routine has outgrown one desk and should be promoted: rebuilt once, properly, in accounts the business owns, on a central schedule, with every run recorded and every outward action waiting for a named person to approve it.

The person who built the routine is not sidelined by this. They become the expert on the process, the one who knows why the exceptions happen, the one whose feedback shapes the promoted version. And they get to stop running it by hand.

Simplify before you automate

The gotcha in the middle of this, and the reason so many automation projects fail, is that the routine is usually built on top of a process designed for humans handing things to humans. Ten steps, three approvals, two check-ins, a hand-off to someone who forwards it to someone else.

If you automate that, you get faster hand-offs. Salim Ismail has a line about this I keep coming back to: when television arrived, the first thing anyone did was put radio announcers in front of a camera. It took years to work out that the new medium needed a different show. So before a routine is promoted, the process gets simplified, ten steps to three, and the approvals get reduced to the one that actually matters, the one where a person says yes before money moves or a supplier hears from us. Most of the others existed because humans forget and drop things. The machine does not, and it logs everything, so the check-ins become a page you can look at instead of a meeting.

Then the roles change, on their own

This is the part people brace for and it turns out to be the gentlest part, if the order is right.

When the Monday numbers assemble themselves and the invoices match themselves and bounce the exceptions, the people who used to do that work do not vanish. They move up. The finance person becomes the person who decides what to do about the mismatch, not the person who found it. The operations lead spends Monday morning in the meeting with the numbers rather than making them. The people closest to the work start asking for new views, “show me this by channel”, “what changed since last week”, and those become two-hour additions to the machine rather than another Sunday.

Roles redefine themselves around judgement, exceptions and relationships, which is what people wanted to be doing all along. And because every stage visibly made their week better before the next one began, nobody experiences it as a restructure. They experience it as finally getting to do the interesting part of the job.

This is what I mean when I say a business has become AI-native. Not that it bought a particular tool. That its default for any recurring task is to have a machine do it and a person check it, that it designs new processes that way from the start, that people share what they work out, and that the leadership team gets its questions answered by asking rather than by waiting for a report.

The advantage of being small

The advice written for large companies is full of scaffolding for this: transformation offices, board sign-off, layers of agent governance. If you have three thousand staff you probably need it. If you have thirty, you do not. You can simply do the work, one process at a time, two or three people and ninety days each, running the new way alongside the old until the numbers match and then cutting over. The whole business can be walked through in a year or so.

That is not a consolation prize. It is the actual advantage, and the businesses that use it will look structurally different from their competitors sooner than anyone expects.

If someone in your business has built a routine

Find out. Ask, without any hint of disapproval, who has quietly got Claude doing part of their job. Then ask them to show you. What you are looking at is your discovery phase, already done, for free, by the person who understands the process best.

And if you would like help working out which of those routines is ready to be promoted and what it would take, send me a description of two or three of them. I will tell you honestly which one to start with, and whether it needs me at all.