AI agents & process automation
An agent takes over the repetitive work. People only approve exceptions — and can see why the agent decided what it decided.
When companies call us
- Someone retypes orders from emails into your system every day.
- Customer support keeps answering the same questions.
- Invoices and delivery notes are copied by hand and errors surface in accounting.
- You tried Make or Zapier, but the scenario breaks on every exception.
Classic automation works as long as the input is predictable. The moment an email is phrased differently, an attachment arrives in another format, or an order carries a note saying “and please add…”, the scenario stalls and a human does the work again.
An AI agent differs in one respect: it understands context and can use tools. It reads the email, extracts line items, checks stock, creates the order and drafts a reply. And when it is unsure, it does not invent an answer — it hands the case to a person.
We do not build an agent for everything. We build one for a single process that genuinely hurts, measure the saving, and only then add the next one.
How we work
1. Pick the process with the highest return
We map what repeats most often and count the hours. If no process delivers a return, we tell you and we do not start.
2. Pilot on real data
Within 2–4 weeks the agent runs on your actual emails and documents — still in “propose, human approves” mode.
3. Evaluations instead of impressions
We build a test set from your own cases and measure accuracy. Without it you cannot tell whether the agent is better than last week.
4. Production and gradual release of control
Where the agent is consistently accurate, we let it act. Where it is not, a human stays in the loop. That call is always yours.
What you get
- A running agent wired into your systems (email, ERP, CRM, inventory).
- An approval interface for exceptions — the agent never acts silently where it is unsure.
- A decision log: for every case you can see the steps taken and the data used.
- An evaluation suite so you can verify at any time that a model change did not degrade results.
- Documentation and team training. The code and the data are yours.
Indicative scope
- Process analysis and solution design
- €900 – €1,800
- Pilot on real data (2–4 weeks)
- €3,500 – €7,000
- Production rollout and integrations
- from €6,000
- Operations, monitoring and evaluations
- from €390 / month
Prices are indicative and exclude VAT. We quote exact figures after the analysis — not off the cuff on the first call.
Frequently asked questions
How do I know the agent will not make an expensive mistake?
The agent starts in a mode where it only proposes and a human approves. We release control only where it performs well on the test set over time. For operations involving money or legal consequences we recommend keeping human approval permanently.
Do we have to replace the systems we already use?
No. The agent connects to what you have — via API, database or MCP. If a system has no API, we can also work with file import and export.
Where does our data go?
Wherever you choose. We can run in European regions of the major providers, or deploy a smaller model on your own infrastructure if data must not leave the company. We explain the cost and performance trade-off upfront.
How long before it actually runs?
A first pilot on real data usually takes 2–4 weeks. Production rollout including integrations typically 6–10 weeks from the start.
If you can name the one process that costs you the most hours, we have something to talk about. If you cannot yet, we start with an audit.