AI for a small business looks narrower in practice than in vendor marketing. Choose one repetitive job that costs a few hours a week, put a tool in front of it, and check a month later whether those hours came back. Strategy documents can wait.
This guide covers how to pick that first task, which tool categories are worth paying for, realistic costs as of 2026, and the risks that hit small firms hardest — a small team has fewer people to catch a mistake before a customer sees it.
The usual way to waste money here is subscribing to a platform and then hunting for something to use it on. Work in the other order. List the jobs your team repeats weekly, then score each one on three questions: how often it happens, how similar the cases are, and how bad a wrong answer would be.
Tasks that score well are frequent and forgiving — answering the same five customer questions, drafting replies to routine enquiries, summarising a weekly export, translating product listings. Each produces a draft a person fixes in a minute.
Tasks that score badly are infrequent, unusual or consequential. Anything ending in a legal, tax or credit decision belongs with a person, and anything a customer sees before an employee reads it needs a review step. The pattern matches our guide to AI automation for small businesses: automate the boring middle, keep judgement at the edges.
Most small business needs fall into five categories. The costs below are typical monthly figures as of 2026 — they change often, so confirm current pricing on the vendor's own page before you budget.
| Category | Typical job | Typical cost as of 2026 |
|---|---|---|
| General chat assistant | Drafting text, summarising documents, internal questions | Free tier; roughly $20-$60 per user per month |
| Customer service chatbot | Repeated questions on your site or messaging channel | Roughly $50-$300 per month, by conversation volume |
| Workflow automation with AI steps | Moving data between tools, with an AI step in the middle | Free tier; roughly $20-$100 per month for a small team |
| Meeting and document summariser | Calls and long files into notes and actions | Roughly $10-$30 per user per month |
| Translation and localisation | Listings, product pages and support replies | Roughly $10-$50 per month for light volume |
One caveat before comparing features: a tool you cannot trace is a tool you cannot check. Read the data terms as carefully as the feature list, and prefer a paid tier that excludes your content from model training.
Four weeks is long enough to learn something and short enough that nobody loses patience. Keep the test small and the measurement boring; name one owner in week 1.
Subscriptions are the visible cost and rarely the largest. Budget four lines. Licences come first: a chat assistant is typically $20-$60 per user per month as of 2026, and seats spread to people who open the tool twice a month. Setup is second — two to five hours per workflow, including the prompt tuning week 3 describes.
Review time is third and the one teams forget: plan on ten to twenty percent of each task's time to check output, more when customers are involved. Fourth is the cost of a mistake — a wrong refund figure or mispriced listing can outweigh a year of subscriptions.
None of these figures are promises, and vendors reprice regularly. Treat any number you read as a planning range and verify it on the official page before committing budget.
Large firms have compliance teams. A five-person business does not, which changes which risks matter.
If a tool touches customer conversations, the governance questions are the same ones we set out for customer service chatbots. The same caution applies to machine translation, quick but uneven on tone — see our comparison of AI translation and human translation.
Two numbers are enough for a first verdict: minutes per item, and the share of items still needing a human rewrite. When minutes fall but rewrites rise, you moved the work rather than removed it — common in the first weeks, usually settled once instructions improve.
Add a third signal when customers are involved: complaints traced to the output, which should not rise. Record all three monthly, next to the subscription cost. A workflow that saves two hours and costs one of review is still a win, as long as you know that is what it is.
References - official documentation and guidance (checked September 2026)
Last updated: 2026-09-14. Features, pricing and data controls change often; verify current details on the official pages above.
You can start on free tiers and pay nothing for the first month. Once a workflow proves useful, expect roughly $20-$60 per user per month for a mainstream chat assistant, and $50-$300 per month for a support chatbot that handles real volume. Those are typical figures as of 2026 and they move often, so treat them as a planning range rather than a quote and confirm on the vendor's own pricing page.
Pick something repetitive, frequent and low-stakes: answering the same customer questions, drafting routine replies, summarising a weekly export, or translating product listings. Measure how long it takes today. Avoid starting with anything that needs a final legal or financial judgement, anything a customer sees before a human reads it, and anything you cannot measure, because you will not be able to tell whether the change helped.
For off-the-shelf assistants and chatbots, no. If you can write a clear email, you can write instructions a tool will follow. Technical help becomes useful when the tool has to connect to your own systems, such as pulling orders from a database or writing back into a CRM. Decide that after a manual test has shown the task is worth automating, not before.
Three matter most. Customer data pasted into the wrong tool breaks privacy promises you may have made. A confident wrong answer reaching a customer or a bank account costs more in a small firm, because fewer people check the work. And seat subscriptions plus setup time add up quietly, so a workflow that is never measured keeps costing money after it has stopped helping.
Compare the same task before and after using two numbers: minutes spent per item, and the share of items that still need a human rewrite. If minutes fall but rewrites rise, you have moved the work rather than removed it. Pick one person who owns the workflow, and set a review date before you subscribe, so the decision to keep or drop it is made on evidence.