You pay twice: once to build and then every month to run it. At 5DAYS an Automation Sprint starts from €2,950, a Business Agent from €7,950 and a Custom AI System from €15,000, excluding VAT. Running costs depend on volume and include AI usage, software and monitoring, from €699 a month for Agent Care.
What does it cost to build at 5DAYS?
We publish our starting prices. All exclude VAT.
| Package | From | Built for |
|---|---|---|
| Automation Sprint | €2,950 | One defined workflow, with agreed connections |
| Business Agent | €7,950 | Complete processes, with AI reasoning, logging and approval moments |
| Custom AI System | €15,000 | Complex systems that become part of how the business runs |
Custom agents start with a technical Discovery of €950. In it we define goals, connections, risks, feasibility and the exact scope before we write the final quote.
What moves the build price?
- Number of systems. And how well each one's API lets you read and write.
- Variety in the input. Fixed forms are simple. Free text in any format isn't.
- The cost of a mistake. More risk means more checks, approvals and testing.
- Exceptions. The main path is rarely the hard part. The other paths are.
- Data quality. Work that has to be cleaned before it can be used adds time.
What are the running costs?
Four things keep costing money after launch. Ask for each of them to be specified before you agree.
| Cost | How it's charged | What drives it |
|---|---|---|
| AI model usage | By the amount of text, in tokens | Volume, length of input, choice of model |
| Automation platforms | A plan, often with usage limits | Runs or steps per month |
| Other software and APIs | Licences or per request | Your suppliers' terms |
| Care and monitoring | A monthly fee or your own time | How critical the system is |
AI usage is the one people find hardest to estimate. Model providers usually charge per token, a small unit of text. Anthropic's price list, for example, prices input and output tokens separately, with different prices per model. Prices change, so check the current pages rather than trusting any figure in an article.
Automation platforms count differently again: Zapier counts tasks and n8n counts executions.
At 5DAYS, Agent Care starts from €699 a month and covers monitoring, checks on failed tasks and technical maintenance. External AI, API and usage costs are charged separately. The exact work depends on the complexity of the system.
Why can usage costs surprise you?
Because an agent's cost isn't one question and one answer. Anthropic notes that agentic systems often trade latency and cost for better task performance. An agent can make several model calls per task, and it often re-reads its context each time. Long documents and long mail threads add up. A loop without a stopping condition can keep going.
To estimate it, run the following with real data, not a demo:
- Tasks per month, from your own measurement
- Average number of model calls per task
- Average size of input and output, in tokens
- Current price of the model, times a margin for retries and growth
Ways to keep it down: use plain rules wherever they work, use a smaller model for simple steps like sorting, shorten what goes in, set a spending limit and watch the cost per task. Providers also offer options such as caching and batch processing, which can help depending on the work.
How do you judge if it's worth it?
Compare against what you've measured, not against a vendor's claim. Count the hours spent on this process today, what those hours cost you and what errors or delays cost. Then set that against the build price plus a year of running costs. Be honest about the time you'll still spend checking. If the sum only works with optimistic assumptions, pick another process or begin with a plain automation.
