AI for small business: where it pays off and where it doesn’t
Updated:
AI pays off in a small business where someone spends hours every week on text-heavy routine work: answering the same customer questions, copying invoices and orders from email into other systems, writing quotes and follow-ups, turning notes into records. It doesn’t pay off for work that happens rarely, depends on judgement and relationships, or that a plain rule already handles. Start with the AI already built into the software you pay for and one general assistant on a business plan. Automate only when one task takes enough time to be worth connecting your systems, and keep a person checking the results.
How many small businesses use AI?
Fewer than the headlines suggest. In the EU, 17% of companies with 10 to 49 people used AI in 2025, against 30% of companies with 50 to 249 and 55% of companies with 250 or more. The survey doesn’t cover companies with fewer than ten people.
Source:
Eurostat
Among companies using AI, the most common purposes were marketing or sales (35%) and organising administration and management (31%).
Where does AI pay off in a small business?
In work that repeats, arrives as text and can be checked quickly. These are the usual places:
| Work | What AI does | What stays with you |
|---|---|---|
| Customer emails and messages | Sorts them, answers the common questions and drafts replies from your own information | Complaints, exceptions and anything that commits the business |
| Supplier invoices and receipts | Reads the PDF and fills in supplier, date, amounts and VAT for your accounting system | Checking mismatches and approving payment |
| Orders sent by email | Pulls out products and quantities, matches them to your codes and enters the order | Orders it flags as unclear |
| Quotes and follow-ups | Drafts the quote from the enquiry and your price list, and reminds you who hasn’t replied | The price, the discount and the relationship |
| Marketing text | First drafts of product descriptions, posts and newsletters | What you say, and whether it is true |
| Calls and meetings | Summarises the notes and lists the next steps | Deciding what happens next |
| Finding things | Answers questions from your contracts, manuals and past emails, with the source linked | Checking the source before acting on it |
To see whether a task is worth it, multiply how often it happens by how long it takes. Twenty emails a day at three minutes each is an hour a day. Ten invoices a month at two minutes each is not worth building anything for. More cases by department are in these AI automation examples.
Where AI doesn’t pay off
- Rare work. A task done twice a year won’t repay the time spent setting it up.
- Work that is all judgement. Pricing a tricky job, handling an upset regular customer or choosing a supplier. AI can prepare the facts; the decision stays yours.
- Work a rule already does. If every case is identical, a plain rule in your existing software is cheaper and more predictable than AI.
- Work on paper or by phone. AI needs the input in digital form. Digitising comes first, and sometimes that alone solves the problem.
- Work you can’t check. If a mistake would only show up weeks later, at a cost, keep a person doing it.
Which AI tools does a small business need?
There is no single best AI for a small business. There are four levels, and most small businesses only need the first two:
| Level | Good for | What it costs | What you need |
|---|---|---|---|
| AI features in software you already use | Drafting and summarising inside your email, office suite, helpdesk, accounting or CRM | Often included, or an add-on per user | Nothing new. Check what your current tools already offer |
| A general AI assistant (ChatGPT, Claude, Gemini, Copilot) | Writing, summarising, research, one-off analysis | A monthly fee per user | A business plan if company or customer data goes in |
| An automation with an AI step | Moving work between systems without copying by hand: invoices into accounting, enquiries into the CRM | Platform fee plus AI usage per case, plus setup | Systems with an API or integration, and someone to set it up and test it |
| A custom integration or AI agent | Work where cases vary and the AI has to look things up and act, such as customer service for a webshop | A one-off build plus running costs | A task big enough to justify it, and clear limits on what the AI may do |
Move up a level only when the one below can’t do the job. Buying several AI tools at once usually means paying for overlap. For the automation level, n8n vs Make vs Zapier compares three common platforms on cost and where your data goes.
Is ChatGPT enough for a small business?
For writing, summarising and answering questions, a general assistant such as ChatGPT is often enough. Two limits matter. First, the plan: personal plans and business plans treat your data differently, so company or customer data belongs only on a business plan with a data processing agreement. GDPR and AI compares OpenAI’s plans and lists the checks. Second, the copying: an assistant answers inside a chat window, so someone still has to paste the email in and the result out. When that pasting is the time-consuming part, you need automation, not a better chatbot.
AI automation for small business: what to automate first
The first automation is usually the inbox or the invoices: the work where the same kind of text arrives every day and ends up in another system. Pick the task that happens often, has digital input, can be checked in seconds, and where a mistake is caught before it costs. Let the AI prepare and a person approve, then give it more freedom only where the results show it is right. The AI readiness checklist helps you test a task against those questions before you spend anything.
Do small businesses need AI agents?
Usually not as a first step. An agent decides its own steps and takes actions in your systems, which suits work where every case needs a slightly different route. For a small business, the clearest case is customer service with real actions behind it, such as a webshop’s order and return questions. LeoAI, the customer service agent we built for webshops, has handled 8,400+ conversations, 94% of them with no person needed (source:
leoai.hu). For most other small business tasks, a fixed workflow with an AI step is simpler and cheaper. These AI agent examples show where the line falls.
What holds small businesses back, and how to get past it
Eurostat asked EU companies that had considered AI why they didn’t use it. The most common reasons in 2025 were lack of relevant expertise (71%), unclear legal consequences (53%) and data protection and privacy concerns (49%)
Eurostat. Each has a practical answer:
- Expertise. You don’t need an AI specialist to start. Pick one task, try it with a tool you already have, and bring in help only for connecting systems.
- Legal consequences. The EU
AI Act puts its heaviest rules on high-risk uses, such as AI that filters job applications or evaluates staff (Annex III). Drafting replies, sorting email and reading invoices are not on that list. This is general information, not legal advice. - Data protection. Use business plans, sign the data processing agreement, and send only the data the task needs. The checks are in GDPR and AI.
How to start with a small budget
- List the tasks that eat your week. Note how often each happens and how long it takes.
- Try the tools you already have. Switch on the AI features in your email, office suite or helpdesk and use them on one of those tasks for two weeks.
- Add one general assistant on a business plan, for the writing and summarising those features don’t cover.
- Automate one task only if it still takes hours after that, and run it as a pilot with a person checking every result.
- Keep it if the numbers hold. Compare the time and the mistakes with how it ran before, then pick the next task.
The full sequence, from choosing the process to rolling it out, is in how to implement AI in a business.
What vSites does
We help smaller companies find the tasks worth handing to AI and build what they need. Through AI consulting we go through your processes and tell you in writing, with a fixed-price plan, where AI pays off and where it doesn’t. If it does, we build the AI automation around the systems you already use, and the developer you talk to is the one who builds it. If a tool you already pay for does the job, we say so.