
In short: AI for small businesses means software that understands texts, receipts or enquiries and turns them into drafts. Don't start with the tool, but with a task that repeats the same way every week and follows clear rules. Check beforehand where the data is stored. And decide who checks every result.
This is where you decide whether and where AI starts in your business. If the task is clear and you want someone to set it up, read AI & Automation for small businesses.
Most questions we hear from business owners start with a tool: “Should we buy ChatGPT?” The better first question is a different one: which work costs us time every week without needing any real skill? This article takes you through three questions, a decision table, the obligations under the GDPR and the EU AI Act, and a starter plan for the first 30 days.
Is AI worth it for small businesses?
For many businesses, yes, if the task is right. According to the Federal Statistical Office (Destatis), around 26% of companies with at least ten employees used AI in 2025; among those with 10 to 49 employees it was 23% (Destatis, ICT survey 2025, published 24 November 2025). Since 2025, Destatis has counted according to the EU definition of an enterprise, so older figures are only partly comparable. In a telephone survey of 604 companies with 20 or more employees, the industry association Bitkom arrives at 36% (Bitkom, 15 September 2025).
Important: solo self-employed people and businesses with fewer than ten employees are not included in either figure. The same logic still applies to you as a solo business. AI tools for the self-employed are only worth it if they make a specific task smaller. So the question is rarely whether AI suits you. The question is where you start.
Where do I start with AI in my business?
With a task, not a licence. This order in five steps has proven itself:
- Find the task: What happens the same way every week?
- Check the data: Where is the information stored, and is it personal data?
- Clarify responsibility: Who checks every result?
- Choose the type of tool: chat assistant, automation, text recognition, or phone and chat bot.
- Test and decide: 30 days with a clear measure, then expand it or drop it.
The first three steps are the real work. They decide whether the first licence pays off. The following sections go through them one by one.
Question 1: What happens the same way every week?
Good candidates repeat themselves, follow rules and involve a lot of text or data. Typical examples are quotes put together from standard blocks, receipts typed up by hand, appointment confirmations, the same five questions by email and enquiries from the contact form.
For one week, keep a tally: which task, how often, how long. At the end you have an honest ranking. Pick the task that comes up often, is annoying and where a mistake is not immediately expensive.
Fictional example: A hair salon in Neukölln answers the same questions every week about prices, opening hours and available appointments. That is a good candidate. Colour advice for a regular customer is not.
Bad candidates are rare special cases, decisions with far-reaching consequences and anything for which you do not yet have a clear rule yourself. AI does not tidy up a chaotic process. It just makes it chaotic faster. If you cannot explain the task to a new temp in five sentences, it is not ready for a machine.
Question 2: Where is the data?
An automation is only as good as its access. If customer data is spread across three Excel files, an inbox and a notebook, the first step is getting organised, not AI.
Check three things:
- Primary system: Which program do you trust when two entries contradict each other?
- Interface: Is there an export or an interface through which data can flow on automatically?
- Personal data: Does it contain names, addresses or even health data? Then the GDPR applies.
Often the biggest gain is not the AI at all. It lies in your programs finally talking to each other, so that nobody types data from one window into another any more.
Question 3: Who stays responsible?
Using AI properly in a business mainly means this: every automated step has a person who checks the result. Every time at first, later with spot checks. For each automation, write down who checks, who notices and fixes errors, and what happens if the tool fails.
For texts to customers, one principle helps: the AI writes the draft, a person sends it. That way tone, facts and commitments stay in your hands. AI assistants sound convincing even when they are wrong. That is why a person always checks prices, deadlines and commitments.
Also record how the tool is used: which template or instruction (the so-called prompt) applies to which task, which building blocks the AI may use and where the checked results are stored. That sounds like bureaucracy, but it is half a page. It makes sure the results do not depend on who happens to be at the computer, and that someone covering during the holidays knows straight away what to do. At the same time, this half page is a good building block for the AI literacy that the AI Act requires (more on this below). And if you introduce a second tool later, you already have a pattern to follow.
Which AI suits which task?
Which AI suits a small business depends on the task, not the brand. That is why the table shows types of tool. Cost range: low means a licence per person per month, medium means a licence plus set-up, project-dependent means set-up plus ongoing costs per use.
| Task | Type of tool | Cost range | Data risk | Who checks |
|---|---|---|---|---|
| Drafting texts, quotes and emails | Chat assistant | low | medium, as soon as customer data is in the text | whoever sends the text |
| Summarising internal notes | Chat assistant | low | low without personal data | whoever uses the note |
| Sorting and forwarding enquiries | Automation | medium | medium | the owner, daily at first |
| Capturing receipts and invoices | Text recognition | medium | high, financial and personal data | bookkeeping before the export |
| Standard questions on the phone or in web chat | Phone or chat bot | project-dependent | high, direct customer contact | spot check every week |
Fictional example: A joinery workshop in Weißensee starts with the chat assistant for quote texts. Receipt capture only comes once the tax adviser has confirmed the export format.
Can I use ChatGPT in my business in line with the GDPR?
Yes, if the subscription, the contract and the way you use it are right. If a provider processes personal data on your behalf, you need a data processing agreement with them under Art. 28 GDPR. Before you buy, ask whether the provider offers such an agreement and where it processes the data.
Five rules that work well in small businesses:
- A business subscription, not a private account. No customer data in consumer tools, meaning not in private accounts or free versions.
- Enter as little data as possible. Only the data the task really needs. For a quote text, “customer, kitchen, 4 m worktop” is often enough instead of a name and address.
- Check training. Look in the settings and terms to see whether the provider uses your input for training, and switch it off where possible.
- Write down the rules. One page is enough: which tools are allowed, which data is off limits, who helps with questions.
- Update your records. Add the tool to your record of processing activities.
There is more on this in the guidance from the German data protection authorities (Datenschutzkonferenz) of 6 May 2024. You will find the obligations in detail under GDPR & EU AI Act.
What does ChatGPT cost for businesses?
The licence is usually the smallest item. Setting it up takes more time. These were the prices on the providers' pages, accessed on 13 November 2026:
- ChatGPT Business: US$25 per person per month, US$20 with annual billing, from two licences (OpenAI Help). OpenAI points out that prices may vary by country and currency.
- Claude Team, standard licence: US$25 per person per month, US$20 with annual billing, for 2 to 150 people (Anthropic). According to the provider, your content is not used for training by default.
- Google Workspace Business Standard with Gemini: €16.20 per person per month on the flexible plan, €13.60 on the annual plan (Google Workspace Help). Gemini is included in Docs, Gmail, Meet and other apps.
For ChatGPT and Claude, the pages show prices in dollars. Check the final price in euros including VAT at checkout before you buy. For an automation, also budget for set-up and the ongoing fees of the tools that connect your programs. This can only be priced reliably once the task is clear.
What does the EU AI Act already require today?
The AI Act also applies to businesses that only use AI. According to the European Commission's timeline (accessed 15 November 2026):
- Since 2 February 2025: the AI literacy obligation (Art. 4).
- Since 2 August 2026: the transparency obligations (Art. 50). From this date, the rules are also enforced, including for AI literacy and transparency.
- 2 December 2026: transition period for certain providers whose systems were already on the market before 2 August 2026. It concerns machine-readable marking under Art. 50(2), so it affects the manufacturers, not you as a user.
In Germany, the KI-Marktüberwachungs- und Innovationsförderungsgesetz (German act on AI market surveillance and innovation) has applied since 29 July 2026. Under it, the market surveillance authority is usually the Bundesnetzagentur (Federal Network Agency).
Do I have to train my team in AI?
Yes, as soon as your team uses AI systems at work. Art. 4 of the AI Act requires you to do your best to ensure sufficient AI literacy. A short, documented training session on possibilities, limits and your rules is a good start. A ban does not help. People will then use ChatGPT anyway, just without rules.
Practical tip: add an AI notice
Art. 50 requires chatbots to be designed so that people notice they are talking to AI (AI Act Service Desk, Art. 50). That is primarily the provider's responsibility. In practice, one sentence in the first message still helps: “You are chatting with an AI assistant.”
If you use AI-generated images, videos or audio recordings that look real, you as the user must disclose that they are artificial. The same applies to texts that inform the public on matters of public interest. Exception: a person has reviewed the text editorially and takes responsibility for it. A line like the one under the headline of this article costs nothing and builds trust. Whether an obligation applies in your case is a matter for legal advice, not for this tip.
Your 30-day starter plan
A four-step roadmap that fits into a month:
- Days 1 to 7, stocktaking: Keep the tally. List your programs and note where which data is stored.
- Days 8 to 10, selection: Choose one task. Define a measure, such as minutes per case. Name the person who checks. Sort out data processing if personal data is involved. Give the team a short training session and record it.
- Days 11 to 28, testing in everyday work: The tool runs with real cases. The person checking reviews every result and notes errors and time.
- Days 29 and 30, decision: expand it, adjust it or consciously drop it. “Not worth it” is also a good result if you know it after 30 days rather than after a year.
If you first want to see where the biggest lever is in your business, a digitalisation roadmap helps.
How do I know whether the pilot is working?
By figures you set beforehand. Three measures are enough for most pilots:
- Time per case: minutes before against minutes after, including checking and rework.
- Errors: How often did the person checking have to correct the content, not just the tone?
- Acceptance: Does the team use the tool without being reminded, or only when someone asks?
If the tool saves time but makes more mistakes, it is not ready yet. If it hardly saves any time but the team likes it, it is often worth taking a second look at the process. And if neither is right, the decision is easy.
Frequently asked questions about AI in small businesses
Which AI is worth it for small businesses?
The one that fits a specific, recurring task. For texts and emails, a chat assistant on a business subscription is often enough. For enquiries, receipts or appointments, an automation between your programs is a better fit. Start with one task and measure for 30 days how much time it saves.
Can I use ChatGPT in my business in line with the GDPR?
That depends on the subscription and how you use it. For personal data you need a data processing agreement under Art. 28 GDPR, and you need to know where the data is processed. Private accounts and free versions do not belong in a business. Only enter the data the task really needs.
Do I have to train my team in AI?
Since 2 February 2025, Art. 4 of the AI Act has required businesses that use AI to do their best to ensure that their staff have sufficient AI literacy. The article does not prescribe a particular course or certificate. A short, documented training session on possibilities, limits and your rules is a good start.
What does AI cost for a small business?
According to the providers' pricing pages (accessed 13 November 2026), a chat assistant on a business subscription costs €13.60 to €16.20 (Google Workspace Business Standard) or US$20 to US$25 (ChatGPT Business, Claude Team) per person per month. Automations and phone bots also involve set-up and ongoing fees. That depends on the licence and the set-up.
How long does an AI pilot take?
Plan for 30 days: one week of stocktaking, a few days of selection, a good two and a half weeks of testing in everyday work, then the decision. Shorter rarely produces reliable figures. Longer is worth it if the task is rare and you see too few cases during the test.
What next?
If you have found a task and want someone to implement it, you will find the process and examples under AI & Automation for small businesses.
This article is not legal advice. For specific legal questions about the AI Act or the GDPR, contact a specialised law firm.
Sources
- Destatis: companies using artificial intelligence (ICT survey 2025, published 24 November 2025), accessed 13 November 2026
- Bitkom: press release on the breakthrough of artificial intelligence (15 September 2025), accessed 13 November 2026
- European Commission, AI Act Service Desk: implementation timeline, accessed 15 November 2026
- European Commission, AI Act Service Desk: Article 50, accessed 13 November 2026
- Regulation (EU) 2024/1689 (AI Act), EUR-Lex, accessed 3 November 2026
- KI-Marktüberwachungs- und Innovationsförderungsgesetz (KI-MIG), BGBl. 2026 I Nr. 223, accessed 13 November 2026
- Regulation (EU) 2016/679 (GDPR), EUR-Lex, accessed 13 November 2026
- Datenschutzkonferenz: guidance on artificial intelligence and data protection (6 May 2024), accessed 13 November 2026
- OpenAI Help: billing and licences in ChatGPT Business, accessed 13 November 2026
- Anthropic: Claude pricing, accessed 13 November 2026
- Google Workspace Admin Help: AI features and pricing of the Business editions, accessed 13 November 2026
As of: 15 November 2026