AI for Small Business: Where to Start

AI for small business works best when it starts with one painful task, not one tool. Pick a task that costs you hours each week, try one AI tool on it for four weeks, measure the result against today’s numbers, and widen only if it worked.
To talk your own case through, see the AI consulting page. Below: the path, a table of first use cases, and adoption data.
How many small businesses use AI, and why do the numbers disagree?
It depends on who is counted and what “use” means, which is why you see anything from under one in five to nearly six in ten. Treat both numbers below as real and as measuring different things.
- U.S. Census Bureau, May 26, 2026. In its Business Trends and Outlook Survey, 19.8% of businesses used AI as of May 3, 2026, and overall use stayed between 17% and 20% from December 2025 to May 2026. The survey counts AI used “in any business function.” Fewer than 20% of firms with four or fewer employees reported using it, and use did not change significantly among firms with fewer than 20 employees.
- U.S. Chamber of Commerce, August 18, 2025. Its Empowering Small Business report says 58% of small businesses use generative AI. The full report shows the survey was run by Teneo Research for the Chamber’s technology group, covered 3,870 U.S. businesses with fewer than 250 employees and was fielded online from June 6 to June 26, 2025. In it, 44% used a generative AI chatbot.
The gap is large, and neither source explains it. A likely reason (inference): the Census asks businesses of every size whether AI is used in a business function, while the Chamber asks small business owners, in an online survey, which technologies they use at work, chatbots included. One person asking a chatbot to draft an email counts in the second survey and may not in the first. The Chamber is also an industry group, and its release voices concern about a “patchwork” of state AI and privacy rules, so read its numbers with that in mind.
What both agree on: use is rising, and most small businesses are still early. That is useful for you. Starting now does not mean being a pioneer, and it does not mean being late.
For the Chamber’s own breakdown, the report says the main uses are about understanding and engaging customers and managing inventory. If you want to see how a larger team approaches the same step, read AI adoption.
What counts as AI for a small business?
For a first pilot, think of two kinds and pick from the first.
- AI inside software you already pay for. Many email, accounting, customer service and sales tools now include an assistant. This is the lowest-effort start, because your data and your team are already there.
- A general AI assistant. A chat tool such as ChatGPT, Claude or Gemini that you give a task and some context. It is more flexible and needs more care about what you paste into it.
A third kind is automation built for you: a custom system that links your tools and runs parts of your operations without a person starting each step. That is where an ai consultant can help, and it comes after a pilot, not before. The wider question of whether a given tool is “AI” matters less than whether it takes a real task off someone’s plate.
“AI for business” at large company scale, with data teams and models trained on your own data, is a different job. This post is about the small version.
Where should a small business start with AI?
Follow five steps, in order. If you skip the first one, you have nothing to measure the other four against.
1. Pick one painful task
Choose a task with four traits: it happens every week, it is mostly text or routine decisions, a wrong answer is cheap to catch, and you can count something about it today, such as hours, replies per day or errors.
Good candidates: replying to the same customer questions, writing first drafts of quotes or content such as product descriptions, summarizing call notes, sorting receipts. Poor first candidates: anything where a wrong answer is costly and hard to spot, such as legal wording, tax filing or medical advice.
I follow a simple rule: if someone does the same task over and over every day, like sending receipts or emails, that task can probably be automated.
Write the task in one sentence and the number it moves: “Reduce the time we spend writing quote follow-ups from five hours a week to two.”
2. Pick one tool
Pick one, not three. Check the software you already use first. If it has an assistant that covers your task, test that. If not, choose one general assistant and use the cheapest plan that lets you test the task.
Before you paste anything real into it, read what the vendor says about your data on that exact plan, because free and paid plans can differ. If the terms are unclear or the page will not load for you, keep customer details out and use made-up examples.
Video · 21 min
Setup Claude Better Than 99% Of People (Full Setup File)
If Claude is the assistant you pick, I show how to set it up with rules, memory and folders so it knows your business from the first chat.
Watch the video →
3. Run a four-week pilot
Give one person ownership, with a set amount of time each week. Use real examples from the last month. Write the instructions you give the tool down, so a colleague can repeat them. The four-week length is a practical suggestion, not a research finding: it is long enough to see the work repeat and short enough to stop.
4. Measure against a baseline
Record the baseline before the pilot starts, then compare. Track hours, how often a person has to fix the draft, and what customers or staff say. A small log in a spreadsheet is enough.
Set the stop rule in advance: “If we do not save at least X hours a week, or if fixes take longer than writing it ourselves, we stop.” Stopping is a result, not a failure.
5. Widen carefully
If the pilot worked, do one of two things: run the same tool on a second, similar task, or give it to a second person. Do not add three new tools at once. Write a short team rule before you widen (see the risks section) and plan some training: see AI training for employees.
Which first use cases give the best payoff for the effort?
For most small businesses, drafting and answering are the best first tasks: little setup, easy to check, quick to measure. The table is a judgement, not a study. The ratings come from how each task is built, and your own business will differ, so test before you trust it.
| First use case | Effort to start | Likely payoff | Cost of a wrong answer | A good first test |
|---|---|---|---|---|
| Draft emails and replies | Low | Medium to high if you write many | Low if you read before sending | Draft ten routine replies, compare time per reply |
| Answer common customer questions | Low to medium | High if the same questions repeat | Medium, because customers see it | Draft answers for your top 10 questions, a person approves each |
| Product descriptions and marketing copy | Low | Medium | Low to medium | Write a month of posts or ten descriptions, edit and compare time |
| Summarize meetings, calls and notes | Low | Medium | Low | Summarize five calls, ask attendees if the summary missed anything |
| Sort receipts and prepare bookkeeping reports | Medium | Medium, and grows with volume | Medium to high, because numbers must be right | Run last month’s receipts alongside your usual process and compare |
| Follow up on leads and quotes | Medium | High if leads go cold today | Medium | Draft follow-ups for open quotes, send them after you review |
| AI phone or chat receptionist | Medium to high | High if you miss calls | Medium to high, because it speaks for you | Try it on after-hours calls only, review every transcript |
Two notes on reading it. First, the “payoff” column depends on volume. A task you do twice a month will not repay the setup. Second, “cost of a wrong answer” is the column most owners skip. A draft a person reads before it goes out is safe. A tool that answers customers on its own is not safe until you have reviewed a few hundred of its answers.
One study speaks to the customer service rows. In a paper by Erik Brynjolfsson, Danielle Li and Lindsey Raymond, published in the Quarterly Journal of Economics in 2025 (NBER working paper 31161, 2023), a generative AI assistant was rolled out to 5,179 customer support agents. Productivity, measured as issues resolved per hour, rose 14% on average. For novice and low-skilled workers it rose 34%, while experienced workers saw minimal impact. That is one company’s setting, so read it as a reason to try an assistant on support work, not as a forecast for your business.
What does a first pilot look like in practice?
Here is an illustration with made-up numbers, not a client result. A six-person home services company spends five hours a week writing follow-ups to open quotes. The owner picks that one task.
- Baseline: five hours a week, and about a third of quotes get no follow-up at all.
- Tool: the assistant already built into the company’s customer management software.
- Pilot: for four weeks, the office manager lets the tool draft each follow-up from the quote details and edits it before sending. She logs the minutes spent per follow-up and how many drafts needed rewriting.
- Stop rule: if drafts need a full rewrite more than one time in four, or total time does not fall below three hours a week, stop.
- Decision: at week four, compare hours, follow-up coverage and any customer replies. If it works, apply the same method to the next task, which could be answering the five questions customers ask most.
The numbers are invented so the structure is clear. The method is what carries over: one task, one owner, one baseline, one stop rule.
Video · 14 min
How I Let Claude Automate My Email Replies (For Gmail & Microsoft)
I show how to connect Claude to Gmail or Microsoft 365 and have it draft replies in your voice, the kind of repeat task a first pilot should pick.
Watch the video →
What risks should you handle before you start?
Three, and each is cheap to deal with before you start.
Your data. Decide what must never go into an AI tool, for example customer payment details, health information, passwords and contracts. Put it in a one-page rule and share it with anyone who will use the tool. The post on AI governance has the longer version, and the AI readiness assessment helps you see where your business stands.
Wrong answers. AI tools state wrong things in a confident tone. Keep a person in the loop for anything a customer, regulator or bank will see. The support study above is a reminder that gains vary by who uses the tool, so check results on your own team rather than assuming.
Hype. Be wary of any course, tool or consultant that promises guaranteed income or savings from AI. On September 25, 2024, the U.S. Federal Trade Commission announced Operation AI Comply, five enforcement actions against deceptive AI claims, and said there is “no AI exemption from the laws on the books.”
If you want a formal reference, the NIST AI Risk Management Framework is voluntary and was released on January 26, 2023, with a generative AI profile added on July 26, 2024. It was written for organizations of all sizes and is more than a five-person business needs. Treat it as a map of the topics, not a to-do list.
Does your team need training?
Yes, but start small. In the Chamber’s survey, 41% of small businesses using AI said they provide AI tools to employees and hope they learn on their own, and 39% offer on-the-job training. The numbers are self-reported by owners. A reasonable reading: tools alone rarely change habits. An hour a week of shared practice on the pilot task, plus the one-page rule, does more than a new subscription. If you would rather have someone teach it live, I run hands-on Claude workshops for teams. See AI training for employees for the longer plan and AI strategy if you want to decide the next three tasks in order.
When should you hire help?
Hire help when the work moves past a pilot. Three signs:
- The task needs the tool to connect to your accounting, customer or scheduling software.
- The task involves customer data and you want someone to check the rules.
- A pilot stalled and nobody has the time to find out why.
When the tool has to connect to your own systems, I build that kind of connection as custom AI apps. Before hiring, read what an ai consultant actually does and, if you are comparing providers, how to choose an AI consulting firm. For the kinds of offers you will see, read artificial intelligence solutions. Ask any provider to start with one small project and a written definition of done, using the same method as your pilot.
What are the next steps?
- Write the one task and the number it should move.
- Choose the tool: your existing software first, one general assistant second.
- Record the baseline and the stop rule, then run four weeks.
- Decide with the numbers: widen, change or stop.
- Write the one-page data rule before a second person starts.
If you want a second opinion on your first task or your pilot plan, start on the consulting page.
Updated November 11, 2026. Sources opened on that date: the U.S. Census Bureau BTOS story, the U.S. Chamber of Commerce Empowering Small Business 2025 page and full report, NBER working paper 31161, the FTC announcement and the NIST AI RMF page.


