AI Costs for SMEs: Planning a Realistic Budget

In short: What AI costs a mid-sized company depends less on the software than on data preparation, integration into existing systems, operation and your team's time. So plan not only the project but also the monthly running costs. If you first classify the tasks in your company and start with a tightly scoped pilot, you can calculate an AI budget reliably.

This article is part of our guide AI Consulting for SMEs.

"What does AI actually cost?" is one of the first questions managing directors and owners ask. The honest answer: it depends on the task, on your data and on how deeply the AI is embedded in your workflows. That sounds evasive, but it can be broken down into concrete cost blocks that you can plan one by one.

How pressing the question is shows in the Bitkom survey of September 2026 among 603 German companies with 20 or more employees: 45 percent of all companies are spending more on AI in 2026 than in the previous year. At the same time, 63 percent of companies without AI cite costs that are hard to calculate as an obstacle, and even among companies already using AI the figure is 54 percent (Bitkom, 14 Sep 2026). This article helps you reduce that uncertainty.

Where the money in AI projects actually goes

Many budgets are built around the licence: what does the tool cost per month? Reality looks different. According to the same Bitkom survey, AI users name these items as a major cost (Bitkom press release of 14 Sep 2026):

• Infrastructure such as cloud and computing power: 51 percent of AI users
• Data preparation: 50 percent
• Integration into existing systems: 41 percent
• Subscriptions and licences for AI software: 21 percent, on a par with external consulting
• Employee training: 19 percent
• Token consumption: a major cost for 8 percent, a minor one for 11 percent

The message for your planning: software is rarely the largest item. Costs rise where AI meets your existing systems, your data and the way you work.

Indicative values by project type

Note: the ranges shown are indicative values for the German market (our own assessment, as of 2026) and not mysoftwarelab prices. mysoftwarelab provides an individual offer after the company org chart analysis.

The following ranges are our own assessment based on typical project scopes for one-off project costs. They are only an indicative value for a first order of magnitude, not an offer.

Project type	Indicative one-off value	What the amount mainly depends on
AI potential analysis / AI check	€2,500 – €8,000	Number of departments and interviews
AI strategy and roadmap	€5,000 – €25,000	Depth of analysis, number of sites
Chatbot / customer communication	€8,000 – €35,000	CRM connection, knowledge base, channels
Process automation with AI	€15,000 – €60,000	Number of systems, exceptions, approval levels
AI-based document processing	€10,000 – €40,000	Variety of documents, target systems
Demand forecasting / planning	€25,000 – €80,000	Data quality and history
Predictive maintenance	€30,000 – €120,000	Sensors, machine data, hardware

On top come monthly running costs. They are often underestimated because they do not appear in any project quote.

The five cost blocks in detail

1. Analysis and consulting

Before the first euro goes into software, you need to know which tasks are suitable at all. External AI consultants in Germany often charge daily rates; as an indicative value (our own assessment, not mysoftwarelab prices) these lie roughly between €1,200 and €3,500 per day. What matters is not the rate but whether you end up with a prioritised list of concrete tasks, data sources and approval rules. So ask up front what you will have in hand after the consulting days.

2. Data and integration

An AI agent that prepares quotes needs access to item master data, prices and customer data. Every interface to ERP, CRM, accounting or document storage takes time. Clean master data saves more money here than any tool negotiation.

3. Software, licences and model usage

There are three models: ready-made SaaS with a monthly price per user, platforms with usage fees, and custom solutions where you pay for the language models directly via your own API accounts. Check whose account the model costs run on. If they run on your own accounts, you see every euro and keep control over data and provider choice.

4. Operation, maintenance and further development

AI solutions are not a project with an end date. Models change, processes change, new exceptions appear. Plan a fixed monthly budget for monitoring, adjustments and bug fixing instead of negotiating every time.

5. Your team's time and enablement

The least visible item. Your specialists have to explain rules, check results and give approvals, especially in the first weeks. Then there is training: Article 4 of the EU AI Act requires a sufficient level of AI literacy among staff who use AI. What that means in practice is covered in the article on the AI literacy obligation under Article 4.

Example calculation: budget for a first AI agent

Assume a wholesaler with 40 employees wants an AI agent to read incoming e-mail enquiries, find matching items and prepare draft quotes. A person in the sales back office checks and approves. The following figures are illustrative and do not replace a calculation:

Item (example)	Assumption
Analysis, defining rules and checklists	one-off €4,000
Connection to mailbox and ERP, testing	one-off €12,000
Internal time in the first eight weeks	around 60 hours of back office and management
Model and API costs	assumed €150 per month
Operation and adjustments	assumed €600 per month

In this example the first year comes to around €16,000 one-off plus around €9,000 running. Whether it pays off depends on how many hours the back office spends on quotes today and what it does with the time freed up. How to set costs against benefits properly is shown in the article calculating AI ROI.

Typical mistakes in AI budget planning

1. Only budgeting for the licence. Integration, data preparation and operation are usually the larger items, as the Bitkom figures also show.
2. Scoping the pilot too large. Trying to automate "the whole administration" produces an incalculable project. One task, one department, clear metrics.
3. Forgetting running costs. Always plan a monthly amount for model usage, monitoring and adjustments.
4. Ignoring internal time. Without specialists who explain rules and approve results, no AI will be good.
5. Dependence on one vendor. Clarify up front who owns data, rules and prompts and how you could switch.

How to plan your AI budget in five steps

1. Classify tasks: Go through your org chart role by role and mark which tasks AI should take over entirely, which it should prepare and which stay with people. A structured starting point is the AI org chart check.
2. Measure current effort: How many hours per month go into the selected tasks today? Without this figure there is no business case.
3. Separate one-off and running costs: Plan project costs and monthly costs separately, each with a buffer.
4. Pilot with a stop criterion: Decide in advance how you will judge after eight to twelve weeks whether to expand or stop.
5. Look at total costs over three years: Only over several years does it become clear which option is cheaper. Include switching costs and internal maintenance time.

How mysoftwarelab keeps costs transparent

mysoftwarelab, based in Worms, Germany, builds and operates a company cockpit with AI agents for businesses with 5 or more employees, as an external AI department. We do not publish fixed prices because the effort depends on your org chart. After the company org chart analysis, which takes around two weeks, you receive an individual offer in one of the packages KI Starter, KI Partner, KI Transformation or KI Enterprise. AI and API costs run through your own accounts, so you can see the model costs yourself at any time.

Frequently asked questions about AI costs for SMEs

How much does AI cost for a small or mid-sized company?

It depends on the task. A tightly scoped pilot is usually in the low to mid five-figure euro range, plus monthly costs for model usage, operation and adjustments. The figure only becomes reliable once tasks, data sources and interfaces are defined.

Which costs are most often underestimated in AI projects?

Data preparation, integration into existing systems, ongoing operation and your own team's time. In many projects the software licence is not the largest item.

Are the costs of language models like ChatGPT or Claude high?

For typical office tasks in mid-sized companies, the pure model costs are usually manageable. More important is that they run transparently on your own accounts and that you can set limits and monthly budgets.

Is an AI pilot project worthwhile for small businesses?

Yes, if the task occurs frequently, can be described clearly and costs noticeable time today. Start with one task, measure the effort before and after, and then decide whether to expand.

How do I calculate an AI budget without relying on vendor figures?

Measure the current effort of the task in hours, separate one-off and running costs, include internal time and look at total costs over three years. That way you can compare offers.

Conclusion: tasks first, then budget

AI costs become predictable when you start not with the tool but with the tasks. If you classify roles and tasks, measure today's effort and separate one-off from running costs, you can compare offers and start a pilot with a clear stop criterion. The largest items are usually data, integration and operation, not the licence.

Sources

• Bitkom, press release of 14 Sep 2026: For the first time, a majority of companies use AI (German)