What does AI cost for an SME? What drives the price

"What will it cost?" is usually the first question a managing director or COO asks us, and rightly so. Stories about AI range from a simple subscription to large transformation programmes. For an SME with 20 to 250 employees, the reality is a lot more predictable than you might think, once you know where the cost comes from.

In this article we explain what drives the price of an AI project, how long it takes and how to judge for yourself whether it's worth it.

First: what are you actually buying?

"AI" is not a product. What you have built almost always falls into one of these categories:

• An AI co-pilot: an assistant that knows your own documents, procedures and tools, which your team can question in Teams, Slack or a web app.
• Process automation: an agent or workflow that takes over recurring work, such as processing invoices, handling intake or reporting.
• An AI inbox: an agent that sorts your support mailbox, summarises, drafts replies and routes to the right person.
• Custom AI software: a complete application built for you, with AI woven into how it works.

The cost differs a lot per category, because the amount of integration and build work differs a lot.

Timelines

For a defined scope, these are realistic timelines with a specialised team:

What	Timeline
AI audit & roadmap	2 weeks
Process automation	2–4 weeks
AI inbox	3–4 weeks
Lead orchestrator	3–5 weeks
AI co-pilot	4–6 weeks
Custom AI software	8–16 weeks

A few notes:

• The timeline says a lot about the price. An automation that connects two systems is faster and takes less work than one that connects four systems with exception rules.
• Ask for a fixed price upfront for defined work. Open-ended hourly billing is rarely a good idea for an SME on an AI project.
• If your scope changes week by week, a team at a day rate often works better than a fixed project.

Where do the costs go?

In an AI project, most of the budget does not go to the AI itself. The language models from OpenAI, Anthropic or Google are a building block you pay for per use. The real work is in:

1. Integration with your existing tools. Your CRM, ERP, accounting software or helpdesk. More systems means more work. This is usually the biggest cost.
2. Making your data usable. Documents scattered across folders and mailboxes need to be structured and made searchable.
3. Reliability. A demo is quick to build. A system that also works correctly on a busy Monday, with logging and an audit trail, takes more.
4. Permissions and security. Who may see which information? A co-pilot that shows salary data to everyone is not what you want.

And after delivery?

Expect two kinds of recurring costs:

• Usage costs of the AI model and hosting. These depend on volume. A good partner makes them clear upfront.
• Maintenance and support, if you want it. You can also take over the code and documentation completely and run it yourself. A good partner doesn't build in lock-in.

How do you know it's worth it?

Calculate it in hours. Take the process you want to automate and ask three questions:

• How many hours per week does your team spend on it?
• What does an hour of those people cost, all-in?
• How many errors or delays does it cause today?

Multiply the hours by the hourly cost and by 52 weeks, and you know what that process costs you today, every year. That's the number you weigh an automation against. For one of our clients in logistics, a set of AI agents delivered around €120,000 in annual savings.

Where do you start?

Start small, with one process that takes a lot of time and is clearly defined. Not sure which process that is? A short audit is the fastest way to find out: after two weeks you know where AI delivers the most.

See our full offer, or tell us which process is eating your team's time.