What does AI cost? Every price in one place

It depends on how far you go. A licence to work with AI yourself costs a few dozen per user per month. Training your team with NinA costs € 4,500 per half day. Finding out what AI delivers costs € 6,000. And a dedicated team that builds AI into your own systems costs € 4,500 to € 12,000 per month.

So the question isn't just what AI costs, but what it delivers. That's why, on every project, we first calculate how much time and money a process costs today. We only start when the saving is at least twice the investment.

Below is everything: our prices in tables, what the cost of AI is made up of, what well-known tools cost, and a worked example with payback time.

An AI model doesn't charge per question but per piece of text: a token. A token is often part of a word. You pay for what goes in (input) and what comes out (output), per million tokens.

A worked example with Claude Sonnet 5.5, which according to Anthropic's price list (October 2026) costs $ 2 per million input tokens and $ 10 per million output tokens (check the current price with Anthropic). An order arrives as an email with a PDF. Together with the instructions, that's about 5,000 input tokens. AI turns it into an order: about 1,000 output tokens.

• Input: 5,000 tokens × $ 2 per million = $ 0.01
• Output: 1,000 tokens × $ 10 per million = $ 0.01
• Total: roughly 2 US cents per order

At 1,000 orders a month, that's about $ 20 in model usage. Set that against an employee who would otherwise spend dozens of hours a month retyping. So model usage is rarely the big cost. Building, maintenance and your people's time are.

An agent that plans its own steps often goes through several rounds and so uses more tokens per task. Even then it's usually cents per task, not euros. Using an expensive model for a simple task is a waste, though. That's why we choose the model that's good enough for each step, and test that in our benchmarks.

An example, not a client case. A business has three processes with lots of manual work. Labour cost: € 45 per hour. We calculate the way we do in the Quick Scan: three numbers per step (duration, headcount, frequency), and we look at the expected scenario, never the most favourable one.

What the processes cost today

1. 01Order entry: 40 minutes × 2 people × 80 times a month = 1,280 hours a year, or € 57,600. Expected: AI takes over 60 percent. Saving: € 34,560 a year.
2. 02Customer questions: 6 minutes × 1 employee × 3,000 questions a month = 3,600 hours a year, or € 162,000. Expected: 50 percent. Saving: € 81,000 a year.
3. 03Quotes: 4 hours × 1 account manager × 20 quotes a month = 960 hours a year, or € 43,200. Expected: from 4 hours to 1, 75 percent. Saving: € 32,400 a year.

Together: € 147,960 saved a year, once all three processes are running.

What it costs

• AI Partnership package S, including Quick Scan: € 60,000 in year one
• Silver maintenance contract for what's live: € 5,340 a year
• Total: € 65,340

The outcome

The saving is almost 2.5 times the partnership investment. That's above our 2x guarantee threshold, so this is one we'd take on. The saving per month is € 12,330. That means the year-one costs are covered by just over five months of savings.

One caveat: processes go live one at a time. In the Quick Scan we plan what runs each quarter, so you can see when the saving really starts coming in.

Because what's built also needs to be maintained. One-off builds often stall after delivery: the builder is gone, nobody manages it, and after a model update or a change to the ERP the process grinds to a halt. Then you've paid for something that no longer works.

That's why we now only build within a partnership. Building, training, hosting and maintenance then sit with one team, with a roadmap and a guaranteed return. For an indication of what having an AI agent built costs in hours, see the questions at the bottom of this page.

If you only want knowledge or only advice, you can buy that separately: a talk, a workshop or the Quick Scan.