How to start using AI in your business (without the hype)

Most businesses know they "should be using AI." Far fewer have actually shipped anything. The gap isn't ambition or budget. It's that the advice they hear is either vague ("embrace AI!") or wildly oversized (a six-figure transformation program). Here's the version that actually works: small, specific, and measured.

I've built and shipped dozens of AI products, and the single biggest predictor of whether a business gets value from AI isn't its size or its budget. It's whether it starts narrow. The companies that win don't begin with a strategy deck. They begin with one annoying task and a two-week experiment.

01Don't start with strategy. Start with one painful task.

A sweeping "AI roadmap" feels responsible, but it's usually how good intentions die in committee. Instead, look at your week and find the single most repetitive, rules-based task you or your team grind through, drafting the same kind of email, summarizing reports, categorizing tickets, writing first drafts of proposals, reconciling data.

Pick one. Not five. The narrower the better, because a narrow task is something you can actually finish, measure, and learn from this week.

The businesses that win with AI don't boil the ocean. They ship one useful thing, measure it, and compound from there.

02Use tools you already have before buying anything.

You do not need to purchase an enterprise AI platform to find out whether AI helps. General-purpose tools, ChatGPT, Claude, Microsoft Copilot, are more than enough to validate a workflow. Run your one task through one of them. See what good looks like and where it falls short.

Buying specialized software or commissioning a custom build makes sense after you've proven the workflow is worth automating, not before. Validate first, invest second. This one sequence saves businesses an enormous amount of wasted money.

03Measure the before and after.

"It feels faster" is not a result. Before you start, write down how long the task takes today and how often you do it. After two weeks of doing it with AI, measure again. Now you have a number: hours saved per week, multiplied by how often the task happens.

That number is the foundation of every decision that follows, whether to expand, whether to buy a tool, and how to make the case to anyone who controls budget. AI adoption that can't point to a number tends to quietly fade. AI adoption with a number behind it gets funded.

04Keep a human in the loop.

The fastest way to lose trust in AI, and to create real risk, is to let it act without review. The fix is simple: AI drafts, a human approves. Set an explicit checkpoint where a person reviews the output before it goes anywhere that matters.

This does two things. It keeps accuracy and risk under control, and it builds the trust your team needs to adopt AI willingly. People are far more comfortable with a tool that assists them than one that replaces their judgment.

05Turn your first win into the next one.

Once you have a measurable win, do two things. First, write it up, a short, plain note: here's the task, here's the time we saved, here's how. That writeup is what convinces leadership and what makes the case for doing more.

Second, find the adjacent task. The workflow next to the one you just improved is almost always the easiest next target. This is how AI adoption actually compounds inside a business: not in one big leap, but in a series of small, proven steps that each build on the last.

06When to bring in help.

You can get surprisingly far on your own with this playbook. The point where outside help pays for itself is usually one of three moments: when the workflow needs real software built (an actual tool, dashboard, or automation rather than copy-paste), when you want a prioritized plan across many possible use cases instead of one, or when you need to get a whole team adopting AI rather than one person experimenting.

That's the work I do, and I do it by building, not just advising. But whether you bring in help or not, the starting move is the same: pick one task, run the experiment, measure it. Everything good follows from there.