Everyone wants AI. Almost nobody knows what they want.

“I want AI to do my marketing.”

I’ve heard that sentence, or a version of it, from a brewer, a broker, and a parts distributor in the last six months. Sometimes it’s marketing. Sometimes it’s quoting, scheduling, follow-ups, the books. The shape is always the same: I want AI to do X.

My reply is always the same too: what does X look like for you?

That’s usually where the conversation stops.

Not because these people are slow. They can read a P&L, hire, negotiate a lease, and keep payroll moving through a bad month. They run things. But when I ask what X actually is in their business, the tasks, the cadence, who does it now and what it’s supposed to produce, there’s no answer. Because there’s no system. Marketing is a Mailchimp account someone set up in 2019 and an Instagram that gets posted to in bursts of guilt. Follow-ups are whatever survives the week. The books are a shoebox with a login.

X, in most small businesses, isn’t a process. It’s a feeling. Usually a guilty one.

And you can’t hand a feeling to AI. You can’t hand it to a new employee either, which is the tell. If you hired someone tomorrow and said “do my marketing” or “handle my follow-ups,” they’d ask the same questions I do, and they’d get the same silence. The problem was never that the work is hard to automate. The problem is the work was never defined in the first place.

This is the gap I keep seeing with AI adoption. Operators have been told, loudly and constantly, that they should be doing something with AI, that they’re falling behind, that their competitors are already ahead. So they arrive with an appetite and no order. The pitch they’ve absorbed is AI as an employee: hire it, point it at a department, walk away.

It’s not an employee. It’s a tool. A very good one, but a tool. And tools need a job. Nobody buys an impact driver and asks it to do their construction.

So here’s what I tell people instead, and it costs nothing. Skip the strategy. Make a list. For one week, write down every task that’s repetitive, annoying, or typed more than once. The email you send every Monday that’s the same as last Monday’s. The quote that matches the previous quote except for three numbers. The customer question you’ve answered forty times. Retyping order details from a PDF into a spreadsheet.

That list is your AI roadmap. The whole thing. Pick the three most annoying items and start there.

AI makes a defined process cheap to run. It cannot make an undefined one exist.

And if X is genuinely the thing you want help with, the first job isn’t an AI job at all. It’s deciding what X is. That part is still yours. Take the smallest honest version you’d be happy with. For marketing, maybe that’s one email a month, two posts a week, and asking every happy customer for a review. Write it down. Now it’s a defined, repeating job, and AI becomes properly useful: drafting, repurposing, keeping the cadence going when you’d rather not.

The operators who get real value out of AI this year won’t be the ones with a strategy deck. They’ll be the ones who picked three boring jobs, handed them over, and noticed they got an hour of their week back. Start simple. It’s a tool. Use it like one.