Is AI worth it for a small business?

It's 9pm. You've just finished the quotes you didn't get to during the day, and your phone shows you an ad promising that AI could have done them in seconds. Then another one. Then a LinkedIn post from someone who "10x'd" their business with a chatbot.

Is any of it true for a small business like yours?

Sometimes. Usually for less than you've been told, and rarely in the place you'd expect. AI earns its keep when it takes over one dull job that happens a lot. It almost never pays when you buy a tool first and go looking for something for it to do.

First, a quick test

Think of the one task in your business you'd most like never to do again. Now ask three questions about it.

Does it happen every day? Not every month. Every day, ideally many times.

Could you explain it to a new hire in five minutes? If the honest answer is "it depends, you just have to know," AI will struggle too.

Would you spot a mistake in ten seconds? AI gets things wrong. If checking its work takes as long as doing it yourself, you've gained nothing.

Three yeses and you've probably found a job AI can take off your hands. Anything less, and AI probably isn't the first fix. The task may need tidying up, writing down, or simply doing less often before any tool will help. None at all, and you can close those ads with a clear conscience.

Hold on to that task. We'll come back to it.

Everyone says they use AI. Far fewer actually do

Last year the U.S. Chamber of Commerce surveyed 3,870 small businesses, and 58% said they use generative AI. In 2023 it was 23% (U.S. Chamber of Commerce). If you feel like you're the last one not using it, that's why.

The Census Bureau sets a stricter bar. Every two weeks it asks businesses whether they actually used AI in one of their business functions in the past fortnight, not whether they've ever opened a chatbot. Between December 2025 and May 2026, only 17% to 20% said yes. Firms with 20 or more staff were pulling ahead; among firms with fewer than 20 employees, the number didn't meaningfully change over those six months (U.S. Census Bureau).

So you're not behind. Most small businesses have someone using ChatGPT to tidy up emails. Far fewer have AI doing a piece of the real work every week. The gap between "we use it" and "it does something for us" is where most of the wasted money sits.

The research is less exciting than the ads

A study of 25,000 workers across 7,000 workplaces in Denmark found AI chatbots saved people about 2.8% of their working hours (Humlum & Vestergaard, NBER). On a 40-hour week, that's about 70 minutes. Useful, but it won't change how you run the place.

An MIT report on larger companies, widely debated since it came out in 2025, claimed 95% of their generative AI pilots showed no measurable effect on profit. Its explanation was that the tools never got wired into how the work actually happened (MIT Project NANDA, via Virtualization Review).

One study is worth slowing down for. A company gave its 5,179 customer support agents an AI assistant built directly into the tool they already worked in. Productivity rose 14% on average. For the newest, least experienced staff it rose 34% (Brynjolfsson, Li & Raymond, Quarterly Journal of Economics).

That second number is the one to remember. The assistant helped new people work more like the team's best. If you've ever watched a business slow down because one experienced person was on holiday, you know what that's worth.

Why such different results? The studies measured different things, so don't read too much into the numbers side by side. But the pattern fits what you'd expect: AI in a separate tab helps a little, and AI built into one specific job tends to help more.

Where AI tends to pay off for small businesses

If you run an insurance agency, a haulage firm, a clinic or anything like them, this is where I'd look first. Notice how many of these pass the three-question test.

That fourth row catches a lot of people out. Plenty of what gets sold as an "AI project" is really a missing automation: someone copies data from an inbox into a spreadsheet every morning because a form was never connected to anything. You don't need AI for that. The fix is cheaper and more reliable without it.

When I'd tell you to wait

Sometimes the smartest AI decision is "not yet."

If the task only happens a few times a week. Ten minutes saved twice a week is about 17 hours a year. That won't cover the setup, never mind the upkeep.

If your customer information lives in three spreadsheets and someone's phone. AI will give you confident, wrong answers. Getting that data into one place is the real first project, and it's worth doing whether AI comes later or not.

If the process is a mess. Automating a messy process just produces the mistakes quicker. Sort out the steps first.

If the software you already pay for has an AI feature. Switch it on before you build anything. A lot of CRMs and practice management systems have quietly added them this past year.

If nobody will own it. Someone has to check what it produces and notice when it drifts. Without that person it slowly gets worse until someone switches it off.

And if you're a one- or two-person business, the honest answer might be a business plan on ChatGPT, Claude or Gemini, around $25 to $30 a month per person, and nothing else. Use the business tier rather than the personal one if customer details go anywhere near it. That's a perfectly good outcome.

Try it small before you spend big

Go back to the task you picked at the start.

Time it for two weeks before you change anything. How often it happens, how long each one takes, who does it. Write the number down. Without it you'll never know whether AI helped or just felt like it did.

Then try the simplest thing that could work, with a person checking every result. That could be a chatbot and a well-written prompt, a feature in software you already own, or a small custom automation. Count the checking time too. If the saving holds up, build it properly. If it doesn't, you've spent very little to find out.

To see what's at stake, multiply hours by cost. Say two people spend 10 hours a week between them re-typing documents, at $30 an hour once you include everything. That one job costs you about $15,600 a year. Take 70% of it off their plate, with someone still checking, and it's worth roughly $10,900 a year, plus about seven hours a week of your team's attention back. That's an illustration, not a real business's numbers, but it's the sum worth doing. You can run it with your own figures in an ROI calculator.

One thing to start this week

Don't buy anything yet. Take the task you picked at the top, and for the next two weeks, jot down every time it happens and roughly how long it takes. A note on your phone is fine.

In two weeks you'll know how often it really happens and what it costs you. That's the number to have in front of you before anyone tries to sell you anything.

Sources

• U.S. Census Bureau, "Large Firms With at Least 20 Employees Biggest AI Users", Business Trends and Outlook Survey, May 2026: census.gov
• U.S. Chamber of Commerce, Empowering Small Business: The Impact of Technology on U.S. Small Business, 2025: uschamber.com
• Humlum & Vestergaard, "Large Language Models, Small Labor Market Effects", NBER Working Paper 33777, 2025: nber.org
• Brynjolfsson, Li & Raymond, "Generative AI at Work", Quarterly Journal of Economics, 2025 (NBER Working Paper 31161): nber.org
• MIT Project NANDA, The GenAI Divide: State of AI in Business 2025, via Virtualization Review: virtualizationreview.com