The five cost items that always arise
Whether you license a platform, hire a service provider or build it yourself: five cost items always arise. Only who pays for them changes.
- Software or licence. The visible price. It can be charged per seat, per usage or per package.
- Implementation. Someone has to connect data, define terms, build the tool and test it. With build-it-yourself toolkits, your team or a partner does this.
- Operation. Monitoring, updates, backup, recovery, handling of fault reports. This item is forgotten most often, because it only begins after the demo.
- Model usage. Language models are mostly billed by usage, either by the provider directly or via the platform.
- Your own time. The business department, IT, data protection and the works council all contribute. These hours are real, even if no invoice arrives.
Public price examples
The following prices come from the manufacturers’ own pages, checked on 15 September 2026. They are not directly comparable, because they include different things.
- Retool (EU, annual payment, Business): €46 per developer per month plus €14 per internal user per month. Implementation and operation by your team. Comparison with Retool
- n8n (self-hosted, Business, annual payment): €667 per month with 40,000 executions, according to the page for companies with fewer than 100 employees. Comparison with n8n
- Microsoft Copilot Studio (US reference): $200 per month for 25,000 credits with annual payment, or $0.01 per credit. Power Apps, data services and implementation may come on top. Comparison with Copilot Studio
- Palantir Foundry: no public price for enterprise contracts found. Comparison with Foundry
- Onagero: one fixed offer for pilot and operation per workflow, after a demo. Build, operation and maintenance are included. How the price comes about
Per-seat, usage-based or package?
Per-seat prices are predictable but hold back usage: if every additional person costs money, fewer people get access, and the tool becomes less useful.
Usage-based prices are fair when usage fluctuates, but hard to predict. Set caps and warnings before you begin.
Package prices per workflow tie the costs to a result. For that you need a clear boundary between what belongs to the package and what does not. That is exactly why a good pilot contract writes down exclusions.
The three most expensive pitfalls
The free proof of concept. It postpones the budget question and often ends in a demo that nobody operates. A paid, small pilot with an acceptance criterion ends up cheaper. More in the article Setting up an AI pilot project properly.
Unlimited scope. “We can build anything” sounds good and gets expensive. A pilot needs a named workflow.
Forgotten operation. A tool that nobody maintains breaks at the next ERP update. Include operation in your calculation from the start.
How to weigh costs against value
Set the measured value against the total costs, not the hoped-for value. Freed-up hours are not automatically saved salaries. How to calculate this cleanly is explained in the article Measuring AI value. You can estimate your current effort for a recurring task in the effort calculator.