In our view, the usual lists of AI tools for small businesses answer the wrong question: they tell you what to add, not what to skip. A common question is: "How are yall learning the tools or even what tools to use?" Our short answer: keep one general assistant, use the AI already inside the software you pay for, and add one more tool only for the job that takes the biggest share of your week.
By the end you'll have the short list of tool categories worth paying for, the filter for every other tool, and a seven-day cut you can do with what you already pay for.
Keep three kinds of AI tool and make every other one earn its place
For small firms planning to use AI, the top challenge was finding tools to meet business needs (54%); for firms already using it, the top challenges were accuracy (46%) and adapting tools to meet business needs (43%) 1. We read that as a sign that the hard part is choosing, not getting access. The five steps below turn the short answer into a method you can run on your own business.
Step 1: Count the AI you already pay for, including what's inside your software
Do this. List every AI tool and AI add-on the business pays for, plus the AI features now built into software you already have, and next to each write the one job it does, the last time anyone used it and its monthly cost.
It worked when you have one dated list with a job, a last-use date and a monthly cost on every line, and a total at the bottom.
Why. In the Federal Reserve Banks' 2025 Small Business Credit Survey, 46% of small employer firms said the business or its employees were using AI for work 1. The survey told firms that AI tools "can include stand-alone AI applications or features of software you already use" 1. We read that definition as a reminder that the AI you pay for can sit inside tools you bought for something else, so the count starts there.
Step 2: Make one general assistant the base
Do this. Pick one general AI assistant and run the business's first drafts, emails, summaries and planning through it before you try any other AI tool.
It worked when each person who uses AI in the business can name the one assistant they open first, and that work goes through it.
Why. The most common uses among small firms using AI were writing or marketing (83%), individual productivity (61%) and planning or analysis (51%), while 19% used it for coding 1. In our view, one general assistant can cover those three kinds of work, so it belongs at the base rather than as one tool among many. In a 2023 experiment published in Science, Shakked Noy and Whitney Zhang gave writing tasks to 453 college-educated professionals, and with ChatGPT the average time taken fell by 40% and output quality rose by 18% 2.
Step 3: Add a tool only for a job the assistant fell short on, on real work
Do this. Write down the one job that takes the biggest share of your week, give the assistant one real piece of that work, and look at a specialist tool for that job only if the assistant's result falls short.
It worked when every tool on your list beyond the assistant has one job written next to it and the date the assistant fell short on that job.
Why. Our reading of the AI tool lists aimed at owners is that the jobs they return to are few: meeting notes, the inbox, first drafts, answering calls and leads after hours, and replying to reviews. Our rule of thumb: tools change faster than jobs, so choose by the job and the tool becomes easy to swap. Match the tool to the skill it asks of you: if you aren't technical, take the simple option, and pick the flexible one only if you will maintain it.
Step 4: Write the job down before you test a tool on it
Do this. Before you try any AI tool on a job, write that job's steps down as you would for a new hire, and give the tool the same steps.
It worked when the tool was judged on the same written steps a new person would get, and you compared its result with how the job is done today.
Why. In our view, AI speeds up a process you already have; it doesn't replace having one, so a new tool won't fix a job nobody has written down. One person put it this way: "I stopped paying for every new platform and started learning how the systems actually work." Understand each tool's job from start to finish yourself, rather than handing the whole choice to the person on the team who knows tech best. When you get stuck, ask the assistant itself: paste a screenshot and ask what to do next.
Step 5: Decide what each tool may read and send before it gets your files
Do this. For each tool that will touch your inbox, client files or meetings, write one line on what it may read, what it may send, and whether your chats on that plan can be used to train its models.
It worked when every tool on the list has that line, and you've checked it against the terms of the plan you actually pay for.
Why. One person warned: "This is sound advice, but beware of what control you give agents to as these have huge privacy concerns." Since August 2025, users of Claude's Free, Pro and Max plans choose whether their chats can be used to train new models; the Team and Enterprise plans fall under Anthropic's Commercial Terms instead 3. We read that as a reason to check the plan, not just the brand: the same assistant can come under different data terms depending on which plan you pay for.
We think the long tool list answers the wrong question
One person reacted to a list of tools this way: "Gotta say, each one of these tools would be super helpful in their own way for my Team. Totally checking these out to see how to incorporate into our workflow. Rock on!" In our view, a list like that invites you to add, and a list that only adds can't tell you what to drop.
Of the small firms using AI, about half were experimenting with it, 44% had partly built it into their processes and 7% had fully integrated it 1. By late 2024, nearly 40% of the US population aged 18 to 64 used generative AI, and 23% of employed people had used it for work in the previous week, according to Alexander Bick, Adam Blandin and David Deming 4. Respondents in that study reported time savings equal to 1.4% of total work hours 4. In a study of 25,000 Danish workers in 11 occupations, Anders Humlum and Emilie Vestergaard found that AI chatbot users reported average time savings of 2.8% of work hours 5. We read those studies together: about half the small firms using AI are still experimenting, and the time people report saving is small, so in our view adding tools on top of a trial adds cost before it adds time.
In the Danish study, the benefits of AI chatbots, including time savings, were 10% to 40% greater where employers encouraged their use 5. Employer encouragement nearly doubled chatbot use for work by the typical worker, from 47% to 83% 5. In our view, the lesson for an owner is that how you set up and encourage the use of a few tools matters more than how many you buy.
Add each AI tool a list recommends and pay for each one on its own.
One general assistant, the AI already inside your software, and an extra tool only where the assistant fell short on real work.
On our ladder, this article is the move from trying tools as they come up to a small set of tools, each chosen against a real job. The next level is giving every tool an owner and a number.
Surveys count AI use in different ways, and we read the gap as a clue
As of early May 2026, the Census Bureau put the share of businesses using AI in the past two weeks at 19.8% 6. In November 2025 the Census Bureau changed its question from AI use "in producing goods or services" to AI use "in any business function" 6. The Federal Reserve survey from step 1 counted AI inside software firms already use. In that survey, 46% of small employer firms said the business or its employees were using AI for work 1. That survey was fielded from September to November 2025 and drew 6,525 responses from a nationwide convenience sample of employer firms with 1 to 499 employees 1. A US Chamber of Commerce report published in August 2025 says almost 60% of small businesses use AI for business operations, more than double the 2023 share 7. A Federal Reserve note of April 2026 puts AI adoption at about 18% of US firms at the end of 2025 in Census data, against about 41% of the workforce reporting work-related use of generative AI in November 2025 8.
| who was asked | what counted as using AI | share | |
|---|---|---|---|
| US Census Bureau, early May 2026 | US businesses | used AI in the past two weeks | 19.8% |
| Federal Reserve Banks survey, 2025 | small employer firms, convenience sample | using AI for work, including features of software you already use | 46% |
| US Chamber of Commerce, August 2025 | small businesses | use AI for business operations | almost 60% |
| Federal Reserve note, April 2026 (November 2025 data) | the US workforce | work-related use of generative AI | about 41% |
We read the gap partly as a matter of what counts: when AI inside software you already use counts as using AI, more firms are users, so some of the AI in your business may already be paid for. Eurostat found no predominant AI technology among EU enterprises; the most used, text analysis, was used by 11.75% of them 9. In our view, that is one more reason to choose by the job rather than chase a single winning tool.
Four public examples show where AI helps, where it hurts, and what to check
Google folded its AI into plans businesses already paid for
On January 16, 2025, Google said it would include its generative AI in Workspace Business and Enterprise plans without the need to buy an add-on 10. Google's January 2025 example: a Business Standard customer who paid $32 per user per month with the Gemini Business add-on would pay $14, $2 more than Workspace without Gemini 10. We read that as the clearest case for step 1: a business on a plan like that may already pay for AI it hasn't counted. Check what your own plan includes today rather than relying on that example.
AI budgets can sit far from where AI pays back
Reporting on MIT NANDA's 2025 study, Fortune wrote that more than half of generative AI budgets go to sales and marketing tools, while MIT found the biggest return in back-office automation 11. We read that as a reason to look past marketing tools to back-office jobs, such as the inbox, invoices and scheduling, when you pick your one extra tool. Fortune also reported that buying AI tools from specialised vendors and building partnerships succeeded about 67% of the time, while internal builds succeeded only one-third as often 11. We read the second finding as support for buying a tool for a named job over building a complex setup of your own.
The same AI helped on some tasks and hurt on a complex one
In a preregistered experiment with 758 knowledge workers, run with Boston Consulting Group, people using GPT-4 on a complex task chosen to lie outside AI's capabilities were 19 percentage points less likely to produce a correct solution than those without it 12. On tasks within AI's capabilities, a separate group in the same study completed 12.2% more tasks, 25.1% more quickly, with significantly better quality 12. We read that as the reason step 3 asks for a test on real work: a list can't tell you which side of the line your job falls on.
The FTC alleged a "robot lawyer" wasn't tested against a human lawyer
In September 2024 the US Federal Trade Commission announced Operation AI Comply, a law enforcement sweep against companies using AI to deceive consumers 13. FTC Chair Lina M. Khan said: "there is no AI exemption from the laws on the books" 13. The FTC said DoNotPay claimed to offer "the world's first robot lawyer", and its complaint alleges the company did not test whether its chatbot's output was equal to the level of a human lawyer 13. Our rule of thumb: if a tool's main claim can't be checked on one real piece of your work, skip it.
Three worries about a short list deserve a straight answer
The first worry is falling behind. One marketing consultant described what business people tell them: "Every business person I have talked to, long enough, fear and confusion are common themes. How do I take advantage? We don't want to be left behind." The same consultant added: "Quietly, if they only knew that I have the same fears and confusion looking at the pace at which AI is going, they wont ask me." In our view, being behind is measured in jobs done better, not in tools signed up for. Less than 20% of US firms with four or fewer employees reported AI use in the past two weeks; between December 2025 and May 2026, use among firms under 20 employees did not change significantly, while it rose among larger firms 6. In 2025, 19.95% of EU enterprises with 10 or more employees used at least one AI technology, including 17% of small enterprises (10 to 49 people) and 55.03% of large ones 9. We read those numbers as a sign that small firms without much AI yet have company, so there is time to choose a few tools with care.
The trick though is to maintain the culture and experience of our retail operation while still adapting to change.one business president
The second worry is losing what customers come for. In Census data through May 2026, about 14% of US retail businesses said they currently use AI, below the national rate 6. We read that as room to move at your own pace. Keep AI on the work behind the counter, such as drafts, notes and admin, and decide on purpose which parts of the customer's experience stay human.
The third worry is that the tool gets things wrong, or that AI doesn't fit your work at all. Keep a person checking anything a tool sends to a customer, and test it first on work where you already know the right answer. A third of small employer firms had no plans to use AI; of those, more than half said it is not applicable to their business, and 13% said it is too difficult or confusing 1. In our view, the short list still helps a business in that position: it keeps you from paying for tools while you wait.
Within seven days, cut your AI tools to the short list and write down the saving
Run the cut within seven days, starting from the list you made in step 1. Write the monthly total before and after, and keep the dated list and the last tool you cancelled as your record.
the seven-day AI tool cut
You now have a short list, a filter and a seven-day cut
You started with lists that only add tools. The short list of tool categories worth paying for: one general assistant, the AI already inside your software, and one more tool only for the job that takes the biggest share of your week. The filter for every other tool: five questions, from a written job to a written line on what the tool may read and send. A seven-day cut you can do with what you already pay for: the checklist above, with a total before and after.
Read before: where AI helps an owner who isn't technical. Read next: Are the AI Agents in Your Business Useful, or Just Busy? for telling a useful AI agent from a busy one, and How to Build Systems in Your Business That Your Team Actually Uses for writing a job down as a list your team uses.
See where your business stands on choosing technology: take the Scorecard.
Sources
- 2026 Report on Employer Firms (2025 Small Business Credit Survey) · Federal Reserve Banks, March 2026
- Experimental evidence on the productivity effects of generative artificial intelligence · Shakked Noy and Whitney Zhang, Science, 2023-07-14
- Updates to Consumer Terms and Privacy Policy · Anthropic, 2025-08-28
- The Rapid Adoption of Generative AI · Alexander Bick, Adam Blandin and David J. Deming, NBER Working Paper 32966, revised February 2025
- Large Language Models, Small Labor Market Effects · Anders Humlum and Emilie Vestergaard, Becker Friedman Institute Working Paper 2025-56, May 2025
- Large Firms With at Least 20 Employees Biggest AI Users · US Census Bureau, America Counts, 2026-05-26
- 2025 Empowering Small Business Report · US Chamber of Commerce, 2025-08-18
- Monitoring AI Adoption in the U.S. Economy · Board of Governors of the Federal Reserve System, FEDS Notes, 2026-04-03
- Use of artificial intelligence in enterprises · Eurostat Statistics Explained, December 2025
- Google Workspace enables the future of AI-powered work for every business · Google Workspace blog, 2025-01-16
- MIT report: 95% of generative AI pilots at companies are failing · Fortune, 2025-08-18
- Navigating the Jagged Technological Frontier · Dell'Acqua et al., Harvard Business School Working Paper 24-013, 2023-09-22
- FTC Announces Crackdown on Deceptive AI Claims and Schemes · US Federal Trade Commission, 2024-09-25
