Overwhelm is a common feeling in this market and an underexposed problem. There's something new every week, everyone says you're behind, and the advice you get everywhere is the same: start small.

That advice isn't wrong. It's just empty. It says how big your first step should be, not where you place it. And the place determines whether, after two months, you have something that keeps running or a trial subscription you forget to cancel.

01 / the definitionWhere do you start with AI in your business?

At the most expensive leak. That's the work that comes up often, requires little judgement and is still done by you. Usually that's email, follow-up or recording what was discussed. You choose that place first and only then the tool. Do it the other way round and you pick a tool and go looking for a problem to fit it.

That's the whole order. Everything below is explanation of that one reversal.

02 / the staircaseWhich step are you on now?

It helps to know where you're coming from, because you may be further along than you think, while at the same time expecting something from the next step that it doesn't give. There are three.

The difference between step 2 and 3 isn't the intelligence of the model. It's who fires the starting shot. At step 2 that's you, every time. At step 3 that's an event in your business: a conversation that ends, an email that comes in, a lead that's been quiet too long.

The jump that counts isn't from dumb to smart. It's from "when I think of it" to "even when I don't think of it".

You can do steps 1 and 2 yourself, today, free of charge. Do that too. Whoever buys step 3 without ever having stood on step 2 is buying something they can't judge is any good.

03 / the adviceWhy is "start small" half an answer?

Because it's about size, not place. Start small on the wrong process and you have a small solution for a problem you didn't have. The advice comes from everywhere because it can never be wrong, and precisely for that reason it doesn't guide you.

There's also a hidden assumption in it: that every process is equally worth tackling. That's not true. Automating five minutes a week is a hobby. The work that comes back every day and that you're stuck doing in the evening is a business problem. Those aren't two sizes of the same thing, they're two different things.

What does guide you is the question which bottleneck is keeping your business small, and whether that bottleneck is even the kind a system solves. That test is set out in how do you know you're ready for an AI system, including the cases where the answer is no.

04 / the choiceHow do you find the most expensive leak?

With three questions, and you don't need anyone for it. Together they take half an hour and they work better than any tool comparison you'll find online.

• What have I done more than twice this week? Go through your sent mail and last week's calendar. Repetition is the raw material; without repetition there's nothing to hand over.
• What consistently gets left undone? Not what's difficult, but what always gets pushed back. That's almost never unimportant work, it's work without a deadline. Follow-up is usually top of this list.
• What can only I do? That's the part you don't give away. Everything outside that is a candidate.

What remains after those three questions is usually one of three places: your mailbox, your follow-up, or recording things after a conversation. That's no coincidence. Those are the three places in a small business where volume, repetition and lack of a deadline come together.

We ourselves usually start with email, because within two weeks you can see whether it's right: you read each draft to check if it hits the mark and you know exactly what you'd have written yourself. That reasoning is set out in your inbox as first AI employee. If your leak is in recording things after conversations, then client files that keep themselves up to date is the better first step.

05 / the catchWhat goes wrong if you start with the tool?

You get a collection of separate things that each do something and together do nothing. Five subscriptions, three logins, none of them knowing what the other knows. Each piece works on the day you set it up and nobody notices when one falls silent two months later, because there was never a moment it had to happen.

The cost of that isn't the subscriptions. It's that after six months you think AI doesn't work in your business, while you never set up anything that could have worked. This is how it goes wrong, and from the inside it looks like a fair experiment.

There's also a genuine drawback to step 3, and we'd rather point it out ourselves: everything that runs needs maintenance. Something in your offering changes, an integration shifts, a supplier adjusts something. What runs has to be kept up, by you or by someone else. Whoever sells you a system without mentioning that word is selling you the first month and not the rest.

06 / usableWhat do you do this week, without buying anything?

Three things, in this order. Together they cost a few hours and they produce something that lasts, even if you never buy anything.

1. Point out the leak. The three questions above, with your calendar and sent mail at hand. Write down one sentence: this is the work that comes back most often and costs the most.
2. Describe that one process. Do it three times deliberately and note what you did, in what order and why you chose what you chose at each turn. That why is the part that normally doesn't come out.
3. Test it by hand. Use that description for a week in an ordinary chat. If it doesn't work with your own description included, a more expensive version won't save it either.

What you end up with is a description of one process in your own words. That's the building block for everything that comes after, and it's also the only thing a provider can build on. If you then come to us or to someone else, the conversation starts two steps further along.

This text was produced with AI assistance and checked and approved by a human before publication.