How can AI help small businesses? Mostly by taking reading and writing work off people. It answers routine customer questions, drafts replies and proposals, reads invoices and orders, and finds answers in your own files. Those uses work today. In the setups we see, most start with an off-the-shelf tool costing a small team a few hundred euros a month. A custom build makes sense when the work runs through your own systems, and starts around €10,000 per feature. AI helps much less when the process keeps changing, the data is missing, or a simple rule would do the job.
Adoption is still uneven. In Eurostat's 2025 survey of EU firms with 10 or more employees, 17% of small ones used AI, against 55% of large ones. Among companies already using it, marketing or sales was the most common purpose, at 35%. Business administration came next, at 31%. The answer differs by function. So this guide goes function by function: what works, what does not, and roughly what it costs.
How AI can help a small business, function by function
The cost of AI for a small business depends mostly on one choice: a tool or a build. Tool costs below are what small teams typically pay in the setups we see. Build costs are the ranges we publish for our own work.
| Function | What AI does well | Off-the-shelf or custom | Rough cost |
|---|---|---|---|
| Customer support | Answers routine questions from your help content, drafts replies for agents, tags and routes tickets | Off-the-shelf first, inside your helpdesk. Custom when answers need live order or account data. | Tool: an add-on to your helpdesk plan. Custom: €10,000 to €25,000 for one feature. |
| Sales and CRM | Turns calls into CRM updates, drafts follow-ups and proposals, cleans and enriches records | Your CRM's built-in AI first. Custom when the work spans CRM, ERP and email. | Tool: an add-on to your CRM plan. Custom: €25,000 to €50,000 for several connected features. |
| Finance and accounting | Reads invoices and receipts, matches them to orders, codes expenses, drafts payment reminders | Often built into accounting software. Custom when documents vary and must be checked against your ERP. | Tool: often included. Custom: €10,000 to €25,000 for one flow, €25,000 to €80,000 for an agent. |
| Operations and documents | Reads orders, delivery notes, contracts and forms, then fills your system of record | Custom more often, because both the documents and the destination system are yours | €10,000 to €25,000 for one document flow, €25,000 to €50,000 for several |
| Internal knowledge | Answers staff questions from manuals, policies and past work, with a link to the source | Off-the-shelf for one clean shared drive. Custom when permissions, many sources or EU-only processing matter. | Tool: a per-user subscription. Custom: €10,000 to €25,000 for one source, more with several. |
Most AI tools for small business are add-ons to software you already use. That is the right place to start, because the tool already sees your data. A build earns its price when the work crosses systems or the output has to be checked against your own rules.
Customer support
Support is where most small businesses feel AI first, because the work is language in and language out. An assistant can answer routine questions from your help content and policies, and hand over to a person when it is unsure. It can draft replies for your team to approve, tag and route tickets, and summarize long threads. If you sell across EU borders, it also drafts replies in languages your team handles slowly.
The best evidence we know of comes from a 2023 NBER study of a customer support operation. AI assistance raised issues resolved per hour by 14% on average. For novice and low-skilled agents the gain was 34%, while experienced agents saw minimal change. For a small business, the gain is largest where new or seasonal staff answer questions your experienced people already know.
What fails is the assistant that guesses. If the assistant cannot see your order system, it will still answer questions about orders, plausibly and wrongly. That is the point where off-the-shelf stops and a custom integration starts. One legal rule applies as well. Since 2 August 2026, the EU AI Act requires that people are told when they are talking to an AI. A support assistant is usually limited risk, and that disclosure is the main duty.
AI customer support for small business teams usually starts inside the helpdesk, as an add-on to the plan you already pay for. A custom assistant that reads order status, account data or warranty records from your systems is typically one AI feature. With us that is €10,000 to €25,000.
Sales and CRM
Small sales teams lose hours to admin rather than to selling, and admin is what AI handles well. It turns call notes into CRM updates, drafts follow-up emails in your tone, and drafts proposals from your past ones. It cleans duplicate records and fills in missing company details. Marketing drafts belong here too. Marketing or sales is the most common purpose in the Eurostat data.
Prediction on thin data works less well. Lead scoring needs history. With a few hundred closed deals, a model has little to learn that your sales lead does not already know. Fully automated outbound at volume is the other trap. It is cheap to send and expensive for your reputation.
Start with the AI built into your CRM, because it already sees your records. Custom work makes sense when the job spans the CRM, the ERP and the inbox. Drafting a quote that needs current stock levels and the customer's price list is a good example. Several connected features like that usually land in the €25,000 to €50,000 band.
Finance and accounting
AI for small business accounting is mostly about documents that someone currently retypes. Invoices, receipts, bank statements and remittance advice all arrive in different layouts. Current models read them well. They extract supplier, amounts, VAT and line items, match them to purchase orders, and suggest the expense category. They also draft payment reminders that sound like a person wrote them.
One boundary matters under the EU AI Act. Using AI to assess the creditworthiness of individual people is high risk, with obligations from 2 December 2027. Reading supplier invoices is not, and neither is drafting a reminder to a business customer.
Many accounting packages already include some document reading, and for standard invoices it is often enough. A custom flow makes sense when documents vary widely or must be checked against your ERP, and typically costs €10,000 to €25,000. An agent that reads, matches and books invoices for a person to approve is a bigger build, usually €25,000 to €80,000.
Operations and documents
In our view this is where custom builds pay back most clearly. The documents and the destination system are both specific to you. A typical case: orders arrive by email as PDFs in five layouts, and someone retypes each one into the ERP. A model can read every layout, map the fields, flag mismatches and leave a draft order for a person to confirm.
The same pattern covers delivery notes, supplier certificates and work orders. It also covers checking documents against a list, such as whether a contract contains the clauses your template requires. None of this needs a model that reasons for minutes. It needs reliable extraction, validation in code and a review queue.
It fails when the process itself is unsettled. If three people handle the same document three different ways, automating it locks in the confusion. Agree on one way first, then automate that.
Off-the-shelf document tools exist and suit standard formats. Once the output has to land in your own ERP under your own rules, it is a build. One document flow is usually €10,000 to €25,000. Several flows with a shared review queue sit around €25,000 to €50,000.
Internal knowledge
Every small business has two or three people who know where everything is. An internal knowledge assistant makes that knowledge searchable. Staff ask in plain language and get an answer from manuals, policies, price lists and past projects, linked to the source. It helps most during onboarding, when new people do not yet know who to ask.
The answers can only be as good as the documents. Three versions of the same price list produce confident answers from the wrong one. Permissions are the other trap. An assistant connected to the whole shared drive will happily answer questions from HR files if nobody restricted them.
There is a data question too. Where documents are processed matters under GDPR, and the major providers offer EU region processing. If your files hold personal data, ask where processing happens and under which data processing agreement before you connect anything.
For one clean shared drive, an off-the-shelf assistant on a per-user subscription is a sensible start. Custom search over your documents is usually €10,000 to €25,000 for one source. Several sources with permissions that mirror your existing access rights usually move into the €25,000 to €50,000 band.
When AI does not help a small business
AI is the wrong tool more often than vendors admit. These are the cases where we tell people to stop, or to spend the money elsewhere.
- The process keeps changing. Automating a moving target means rebuilding every quarter. Stabilize the process first.
- A rule or a query already gets the right answer. A model adds cost, latency and doubt to something that worked.
- Nobody can say what a correct output looks like. Without that, there is no way to test the system and no finish line.
- The data is not there. If the knowledge lives on paper or in two people's heads, the first project is capturing it.
- The volume is too low. A task that takes two hours a week is about 100 hours a year. A €15,000 build to remove it costs €150 per saved hour in year one.
- The decision is about a person and errors are costly. Hiring and credit decisions about individuals are high risk under the EU AI Act, with obligations from 2 December 2027.
- The whole budget is under about €10,000. Buy a good tool, train the team on it, and clean the data you will need later.
The volume test is the one people skip. Multiply hours saved per week by 50, then by what an hour of that person costs you. If the result does not clearly beat the build price over two years, buy a tool or wait.
How to pick your first AI use case
Eurostat also asked companies that had considered AI but not adopted it what stopped them. Lack of relevant expertise came first, at 71%. Unclear legal consequences came second, at 53%. You do not need an expert to make the first choice, though. You need a list and an honest afternoon.
- 01List the tasks where people read or write the same kind of thing every week: tickets, invoices, orders, proposals, internal questions.
- 02Estimate the hours per week for each, and note who does them.
- 03Check that the data exists in digital form and that someone can say what a correct result looks like.
- 04Check whether a tool you already pay for does it. Often it does, and the first project is switching it on properly.
- 05Pick one task with real volume, a clear correct answer and a person reviewing the output. Measure the hours before and after.
Good AI use cases for small business share three traits: weekly volume, digital data and a correct answer someone can check. Support replies and invoice extraction are the most common first picks for that reason.
If the list produces something worth more than a tool, outside help starts to earn its fee. Our two week readiness assessment, €9,500 to €14,000, runs this exercise against your real data and ends with one costed first build. If the total budget is under €25,000, skip the assessment and buy a tool. We would rather say that up front.