Run it with less chaos · Every business
AI for Small Service Businesses: Where It Helps and Where It Doesn't
AI tools are good at first drafts and routine writing and poor at facts they were never given, so the useful question for an owner is which jobs in the week are safe to hand over and which still need a person.
At a glance
- Sensible AI use means giving an AI tool the jobs where a wrong first draft costs little and a person checks the result, such as writing, summarizing and planning, while keeping it away from prices, promises and anything a customer will rely on without a second look.
- In the Federal Reserve's 2025 survey of 6,525 small employer firms, 46% said the business or its employees use AI, and accuracy was the most common problem users reported, named by 46% of them.
- In a preregistered experiment with 758 consultants, AI users finished 12.2% more tasks and worked 25.1% faster on tasks the tool handled well, yet were 19 percentage points less likely to reach a correct answer on a task outside its abilities.
- For an illustrative owner who spends five hours a week on writing, a quarter of that time saved, less review time and the subscription, is worth about $2,160 a year.
- Start with one writing job you repeat every week, keep a person on every message that leaves the building and measure the hours before buying more.
Part 1 | Most small firms are already using it for writing
AI arrived in small businesses through the keyboard rather than the front desk. The Federal Reserve Banks' Small Business Credit Survey, fielded in the fall of 2025, found that 46% of employer firms said the business or its employees used AI, another 15% planned to start within a year and a third had no plans; among users, 83% used it for writing or marketing, while 29% used it for customer service.
The Census Bureau's biweekly Business Trends and Outlook Survey, which asks whether a business used AI in any business function, puts the national rate lower, at 19.8% in the period ending May 3, 2026, and finds that use rose among firms with 20 or more employees while it did not change significantly among firms with fewer than 20. The two surveys ask different questions of different samples, and the gap between them is itself informative, since it suggests that many owners and staff use AI on their own for a draft or an email well before the business adopts it as a system. For an owner the first instinct is often Everyone else is using AI, so we are behind; the surveys suggest instead that most small firms are at the same early stage, using it to write.
Part 2 | The line between help and harm is uneven
The strongest evidence on what AI does at work comes from two field studies. In a study of 5,179 customer support agents published in the Quarterly Journal of Economics, access to an AI assistant raised issues resolved per hour by 14% on average and by 34% for novice and lower-skilled agents, with little effect on the most experienced, and customers' sentiment in the chats improved. The tool worked as a coach that carried the best agents' habits to the newest ones, which is a useful picture for a small business where a new front-desk hire learns by copying whoever trained them.
The second study, run with 758 consultants at a large consulting firm, gave some participants AI on realistic tasks and found that the line between what the tool does well and badly is jagged rather than straight. On tasks inside the tool's abilities, AI users completed 12.2% more tasks, 25.1% more quickly, with more than 40% higher quality than a control group; on a task chosen to sit just outside those abilities, they were 19 percentage points less likely to produce a correct answer, because the tool's answer sounded right and they accepted it.
For a service business the pattern sorts tasks into two groups. Drafting a reminder, rewriting a service description, summarizing a long supplier email or outlining a training checklist sits inside the frontier, because a person can see at a glance whether the result is right; quoting a price, stating a warranty term, answering a medical or legal question, or telling a customer what the schedule allows sits outside it, because the tool does not know your price list, your calendar or your policies unless someone gives them to it, and a confident wrong answer costs more than no answer.
| Job | Good fit for AI | Keep with a person |
|---|---|---|
| Messages to customers | First drafts of reminders, thank-yous and follow-ups in your own voice | Sending, and anything about price, timing or a complaint |
| Marketing | Post ideas, rewrites of service pages, captions | Claims about results, before-and-after promises |
| Estimates and quotes | Tidying the wording of scope notes | Prices, quantities, warranty terms |
| Office work | Summaries of long emails, meeting notes, checklist outlines | Payroll, tax, contracts |
| Customer questions | Drafting FAQ answers from your own written policies | Live answers on health, legal or safety questions |
Part 3 | Count the hours before paying for more
The value of AI in a small business shows up as hours, so the first test is a simple time count: hours a week spent on the task × share of time saved, minus the time spent checking, × weeks worked × what an hour of your time is worth, minus the subscription. The share saved should be modest, since the 25% speed gain in the consulting study came from tasks chosen to suit the tool.
| Hours a week on writing (messages, posts, descriptions) (Assumption) | 5 |
|---|---|
| Share of drafting time saved (Assumption, near the 25.1% speed gain in the consulting experiment) | 25% |
| Hours saved a week (Calculated (5 × 25%)) | 1.25 |
| Hours a week spent checking drafts (Assumption) | 0.25 |
| Net hours saved a year (Calculated ((1.25 − 0.25) × 48 weeks)) | 48 |
| Value of an owner's hour (Assumption) | $50 |
| Subscription (Assumption ($20 a month)) | $240 a year |
| Net value a year (Calculated (48 × $50 − $240)) | $2,160 |
Illustrative example; replace every assumption with your own numbers.
The figure is small next to the claims made for AI, and that is the point of running it: an owner who saves an hour a week and spends it on follow-up with past customers has a better return than one who buys three tools to automate work that took twenty minutes. Users in the Federal Reserve survey reported the same pattern, with 71% saying AI raised productivity and 31% saying it raised sales, while the vast majority saw no change in labor costs.
How you compare
- Adoption: 46% of small employer firms in the 2025 Federal Reserve survey said they or their staff use AI, and about half of those were still experimenting; the Census Bureau's broader measure was 19.8% in spring 2026.
- Problems: accuracy (46%) and fitting tools to the business (43%) were the top problems reported by users, and over half of firms not using AI said it did not apply to their business.
- Your own baseline: write down the hours you spend each week on writing, summarizing and planning, then track them for a month after you start, so the decision to pay for more rests on your own numbers.
Questions to ask this week
Which writing job do I repeat every week? Reminders, follow-ups, post captions and service descriptions are the usual candidates.
Who checks a draft before a customer sees it? Name the person and the step, even if that person is you.
What facts does the tool need from us? Write down prices, policies and hours in one document, so drafts start from your rules instead of guesses.
What will we never hand over? Prices, promises, complaints and health, legal or safety answers belong on that list.
The first step is to pick one weekly writing task, time it for a week without AI and a week with it, and keep the tool only if the hours saved are real.
About this research: This article draws on the Federal Reserve Banks' 2025 Small Business Credit Survey of 6,525 employer firms (a convenience sample), the Census Bureau's Business Trends and Outlook Survey, a field study of 5,179 customer support agents published in the Quarterly Journal of Economics and a preregistered experiment with 758 consultants published as a Harvard Business School working paper. The worked example is illustrative.
Sources
- Federal Reserve Banks. 2026 Report on Employer Firms: Findings from the 2025 Small Business Credit Survey. March 2026. 6,525 small employer firms.
- U.S. Census Bureau. Large Firms With at Least 20 Employees Biggest AI Users. May 2026. Business Trends and Outlook Survey.
- Brynjolfsson, E., Li, D. and Raymond, L. Generative AI at Work. Quarterly Journal of Economics 140(2), 2025. 5,179 customer support agents.
- Dell'Acqua, F., McFowland, E., Mollick, E., Lifshitz-Assaf, H., Kellogg, K., Rajendran, S., Krayer, L., Candelon, F. and Lakhani, K. Navigating the Jagged Technological Frontier. Harvard Business School Working Paper 24-013, 2023. 758 consultants.
Related reading: The Five Kinds of Software a Small Service Business Needs (and What to Cancel) · How Much Small Businesses Spend on Software They Don't Use · Where Small Business Owners' Time Goes, and How to Take It Back · Using AI to Draft Customer Messages That Still Sound Like You · Should a Small Business Put a Chatbot on Its Website?
Work with osNOVA on using AI where it saves real hours
We help owner-run service businesses decide which weekly jobs AI can take on, write down the prices, policies and voice the tools need, and set the review step that keeps every customer message accurate, so the hours saved go back into fuller weeks and more regulars.
Email hello@osnovaops.com about your business
Or leave your email and we will reach out
Tell us what kind of business you run and what you want more of.
Not ready to talk? osNOVA runs free live events for owners.