What does AI cost an SME?

What does AI cost an SME?

That depends on what you want to do with AI. A subscription for your team costs an amount per user per month. AI that works inside your own systems costs most when it is built, and after that a monthly amount for usage and maintenance. There are three types of cost:

Standalone amounts for ‘an AI project’ say little about your business. A subscription for five people and AI that enters orders into your ERP are very different things. That is why below you can read, per type, how you pay and what determines the price.

What does an AI subscription per user cost?

You pay for a business subscription to ChatGPT, Copilot or Claude per user per month. With annual billing, the monthly price is usually lower than with monthly billing. Sometimes a basic version is already included in a subscription you have. This is how it works per subscription, as of 22 October 2026:

You work out what you pay in total like this: the number of people using the tool, times the price per user, times twelve months. Pay attention to the currency and tax: OpenAI and Anthropic quote their prices in dollars, Microsoft in euros. Also check whether VAT is already included in the price. The current price is on the provider’s site.

Note two things. Personal subscriptions, such as ChatGPT Plus and Claude Pro, are meant for one person, without central management. Why that matters is explained in Is ChatGPT safe for businesses? And usage sometimes has a limit: with ChatGPT Business you can buy extra usage with credits, and with Claude Enterprise you pay for usage as well as for the seat.

Which tool suits you depends mainly on what you already work in. For that, read ChatGPT, Copilot or Claude: which AI suits your business?

Prices and names of subscriptions change often. Check the price with the provider before you take out a subscription.

What does AI in your own systems cost?

AI in your own systems costs most when it is built, plus a monthly amount. This is not a chat window, but an application that works along in your work. For example: AI reads incoming email, extracts the details, puts them into your system and prepares a draft. An employee checks it and decides.

Such an application has a few parts. A language model reads and writes the text. API integrations fetch data from your systems and put it back. And ordinary software enforces the rules: which step comes next, what goes to a person and what happens if something is not right. So the language model is only one part. A large share of the work is in the rest.

An example: for De Rijplaten Specialist and De Slowfeeder Specialist, two online shops from Ede, I built an automatic quote generator. AI recognises an enquiry in the email, puts the details into the system and proposes a price. After a click on approve, the quote goes to the customer. For the same businesses I also built automatic transport planning. You can read more about this in the case.

If AI takes steps in your systems itself, that is called an AI agent. When that makes sense is explained in An AI agent for your business.

What determines the price of a custom AI application?

The price depends mainly on the work around the AI: how many systems are involved, how good their API is, how many exceptions there are and how much human review you want.

Human review takes build time. Even so, I recommend it for anything that affects money, agreements or people. AI does the groundwork, you decide.

Is it only about two systems sharing data, without AI? Then read what an API integration costs. Is a whole system missing? Then read what custom software costs.

What does using a language model cost?

You usually pay for using a language model per amount of text. That text is measured in tokens: small pieces of words. How that works is explained under the term LLM.

Providers such as Anthropic and OpenAI quote a price per million tokens. There is a price for the text that goes in, such as an email or a PDF. And there is a price for the text that comes out, such as a summary or a draft. The text that comes out is more expensive per token. And a more powerful model costs more per token than a lighter model.

What that costs per month depends on how many emails or documents go through it, how long they are and which model you choose. For a simple task, such as sorting an email, a lighter model is sometimes enough. With me, hosting, subscriptions and maintenance are billed together in one monthly amount. How much that is, is stated in the proposal.

How is the price set when you work with me?

The first conversation is free. After that I carry out a business analysis, and you receive a proposal with a fixed price. That way you know where you stand in advance. This is what you can have me do:

What it costs for you is stated in the proposal. If you continue after the analysis, I offset the analysis against the follow-up work. Also read how your price is set. I do not sell licences and receive no commission from suppliers. If a subscription for your team is enough, I will tell you. Read more about AI in your business or book a conversation.

How do you compare quotes for AI?

Are two quotes far apart? Then they often do not describe the same thing. So do not just look at the amount at the bottom, but put them side by side on these points:

• Which task it solves. Which emails, documents or steps go through it, and which systems are involved?
• What a person checks. Where is the approval step, and what happens if the AI is unsure?
• One-off and monthly together. Work out what you will pay over a few years, for the build and the monthly costs together.
• How use of the language model is charged. Is it included in the monthly amount, or do you pay for it separately? And what happens if more email or documents go through it?
• Fixed price or hours. Is the price fixed, and what falls outside it?
• Testing and explanation. Is it tested with real examples from your own work, and does someone explain to your team how it works?

For every quote, also ask: where does my data go, will we sign a data processing agreement, and who owns the code and the settings if we stop?

Which costs are easy to forget?

The costs that do not appear on an invoice are the easiest to forget. Think of:

• Your own people’s time. For the analysis, the testing and learning to work with something new. And after that for the review: if an employee looks at every draft, that takes time too.
• Cleaning up data. Is the same customer in your system three times? Then that has to be fixed first. With me that is extra work, and I only do it once you agree to its price in writing.
• Training and rules. If your team uses AI, the EU AI Act requires you to take measures that support your people’s AI literacy. A certificate is not needed. Read more in AI literacy is mandatory and AI policy on one page, or about AI training.
• Agreements about data. If an AI service processes personal data for you, you sign a data processing agreement with that party.
• Costs with other suppliers. Sometimes access to your software package’s API comes with a more expensive subscription. Check this before you start.

Is money the biggest barrier to AI for SMEs?

Not according to CBS (Statistics Netherlands). In 2024, CBS asked businesses with 10 or more employed persons that had considered AI but did not use it why they did not. Almost three in four (74.6 per cent) cited a lack of experience. Just over one in five (22.8 per cent) found the costs too high. Privacy and legal consequences were also mentioned more often than costs. These are provisional figures.

Use is growing, though. In 2025, 27 per cent of businesses with 10 to 50 employed persons used AI, compared with 11 per cent in 2023. Among businesses with 50 to 250 employed persons, it went from 20 to 45 per cent. The figures for 2025 are provisional.

Knowing what AI can and cannot do helps you choose with purpose. That is why the first step is often small: a training session for your team, or an analysis of one process.

What can AI do for your business, and what can it not?

AI is good at text-based work that comes up often. It does not replace expertise or a person who decides.

AI is good at this:

• reading and sorting emails and enquiries;
• extracting data from PDFs, such as orders, packing slips and invoices;
• writing a first version of a reply, quote or report;
• summarising long documents;
• finding answers in your own manuals and procedures.

AI cannot do this, or not without help:

• Work without errors. A language model can give an answer that sounds convincing but is wrong. That is called a hallucination. So someone has to check the result.
• Know your prices. A language model does not know your prices and is not a calculator. A price proposal should come from your own price list, via fixed rules. How that works is explained in Making quotes with AI.
• Fix messy data. If your lists are wrong, the result will be wrong too.
• Make decisions that affect people. If it concerns a customer, a job applicant or an employee, a person decides.

When is AI not worth the money?

AI is not worth the money if the work rarely comes up, if every case is different, or if your software package can already do it. A few situations:

• Low volume. If it is about a few emails or documents a week, a good template or a subscription is often enough.
• Many exceptions. Is every job different and does it need its own judgement? Then it remains mainly work for people, and AI helps with the text at most.
• It is already in your package. Some packages already have AI features, such as reading in invoices. Look at that first.
• Your way of working is still changing. Are you switching process or software soon? Then wait until that is settled.

Not sure yet where AI fits for you? Then read You don’t know where to start with AI.

How do three different businesses choose?

Every business makes a different choice. Below are three made-up examples. They are not my clients, and they deliberately contain no amounts.

How do you keep the costs of AI low?

Start with one process, have it looked at first and agree a fixed price. Step by step:

1. Choose one process. Work that comes up often and always goes the same way, such as enquiries, orders or invoices.
2. See what your package can already do. Sometimes the feature is already there, or a subscription is enough.
3. Write down for a few weeks how it goes now. How often does the work come up, how long does it take and where does it go wrong? With those figures from your own business, you can weigh up a price properly.
4. Collect examples. A few ordinary cases and a few difficult ones, such as emails, orders or documents from the past months.
5. Have an analysis done and ask for a fixed price. Also ask what comes on top per month, and what falls outside the price.
6. Start with a first version. Build only the most important part, with a person who checks. Such a first version is called an MVP.
7. Only expand once it works. A next step gets its own proposal.

Frequently asked questions

Sources: CBS: AI monitor 2024, use of AI technology by Dutch businesses (in Dutch) (reasons for not using AI), CBS: Businesses use AI most often for marketing or sales (in Dutch) (12 December 2025), OpenAI: subscriptions, Microsoft: Microsoft 365 Copilot subscriptions (in Dutch), Anthropic: Claude subscriptions and per-token pricing, OpenAI: the API, priced per million tokens, Google: Workspace subscriptions, European Commission: AI literacy, questions and answers, the Dutch Data Protection Authority (Autoriteit Persoonsgegevens): Data processing agreement (in Dutch) and GDPR, Article 28(3) (in Dutch). Consulted on 22 October 2026. Prices of other providers change often: check their site for the current price.