Where should a business start with AI?

The short answer

Map how work actually gets done in your business today. Pick the one task where saved time clearly turns into money. Put AI to work on that single task. Measure the result within 30 days. Then, and only then, pick the next task.

Notice what is not in that list: choosing a tool. The tool is the last decision, not the first. Most businesses get this exactly backwards, and it is the main reason so many walk away saying AI did not work for them.

Why this question feels so hard

If you run a business right now, you are being told from every direction that you need AI. Your industry newsletter says it. Other business owners say it. Your kids probably say it. What nobody tells you is where to begin, and the honest reason is that most people selling AI are selling a tool, and the answer to "where do I start" is always their product.

So owners do the sensible-looking thing. They buy a subscription or two, forward the login to a manager, and wait. Three months later the tools sit unused, the team is mildly annoyed, and the conclusion writes itself: this AI thing is not for businesses like ours.

The tools were never the problem. The order was.

The wrong way: tools first

Buying an AI tool before you understand your own workflows is like buying gym equipment before deciding what you are training for. The purchase feels like progress. It is not. A tool without a specific, named problem attached to it is a solution looking for a problem, and in a busy company, nobody has time to go looking on its behalf.

There is a second, quieter cost. Every tool that gets bought and abandoned makes your team a little more skeptical. By the third one, "we tried AI and it didn't stick" has become part of the company's story about itself. That story is expensive to undo.

The right order, in four steps

1. See how work actually gets done

Not how the org chart says it gets done. How it actually happens. Where do enquiries land? Who types what into which spreadsheet? Where does a quote sit for two days waiting for someone to get back from site? You are looking for the places where time leaks: the same information typed twice, the follow-up that depends on someone remembering, the report assembled by hand every Monday.

You do not need special software for this. You need honesty and a few hours. Most owners who do this exercise properly are surprised twice: first by how much of the business runs on memory and habit, and second by how obvious the fix list becomes once it is written down.

2. Pick one task. Not five. One.

This is the step that separates businesses that get results from businesses that get subscriptions. From your list of leaks, pick a single task using four tests:

• You can measure it within 30 days. Quotes going out same-day instead of in three days is measurable. "Better productivity" is not.
• One fix helps several things. The best first tasks feed multiple outcomes. Faster quotes also means fewer lost deals and less chasing.
• The information it needs already exists. If the fix requires six months of collecting new data first, it is not a first task.
• If it stumbles, nothing breaks. Do not start with payroll, compliance, or anything a lawyer would want to review. Start where a mistake is an inconvenience, not a crisis.

Everything else that made your list stays on the list. It is not abandoned, it is queued. That distinction is what keeps this from becoming overwhelming.

3. Make it live, in weeks

A first AI project should go live in weeks, not quarters. If someone proposes a six-month transformation program as your first step, walk away. Speed matters here for a business reason: your team decides in the first month whether this is real or another management fad, and a small win they can see beats a grand plan they have to take on faith.

4. Measure it, out loud

Before the work starts, write down the number you expect to move and what it is today. After 30 days of running, compare. Share the result with the team either way. If it worked, the next project just got easier to sell internally. If it fell short, you learned something specific about your business for a modest price, which beats learning nothing for a large one.

Who should drive this?

You, or someone who genuinely knows how the business runs. This is the part that surprises owners: the hard work of starting with AI is business thinking, not technical work. Mapping workflows, choosing the first task, defining what success looks like: those are judgment calls only someone who knows the company can make. The technical build can be hired. The judgment cannot.

That split, judgment from people and heavy lifting from machines, is more than a way to start with AI. It is the whole model for running a business with it. We call a company that works this way a centaur business, and it is the end state all of this points toward.