How much does AI cost a company? Cost drivers and how to calculate value

What an AI project costs depends less on the AI itself than on three factors: how well your data is prepared, how many systems need to be connected and how big the scope is. According to Bitkom, many companies find AI more expensive than expected, usually because running costs and integration effort were underestimated. Knowing these factors in advance makes planning realistic.

The main cost drivers

1. Data readiness: are documents and data digital, current and findable, or do they need preparing first?
2. Integration: the more systems are connected, such as ERP, CRM or email, the higher the effort.
3. Scope: a clearly defined process costs less than a company-wide programme.
4. Requirements: EU hosting, access control and documentation duties under the AI Act.
5. Standard or custom: an adapted off-the-shelf tool is usually cheaper than a custom build.

Don't forget running costs

• Usage costs for language models, depending on volume.
• Hosting and operations.
• Maintenance: new documents, changed processes, interface updates.
• Training new employees and refreshers.

How to calculate the value

For a first estimate, a simple calculation with your own numbers is enough:

1. Time saved: how many hours per month does the process cost today, and what share can be automated?
2. Value of time: multiply the hours saved by your internal hourly cost.
3. Other effects: fewer errors, faster response times, more revenue from freed-up capacity.
4. Net it out: subtract running costs and divide the investment by the monthly net benefit. That gives the payback period in months.

How to reduce cost risk

• Start with a clearly defined pilot.
• Measure before and after.
• Use existing systems rather than introducing new ones.
• Plan running costs from the start.

In our free AI Scan we give you a written estimate of effort and value for your use case. More on AI strategy & opportunity assessment.