What does AI automation cost?

Assessment and potential analysis run 5,000 to 15,000 EUR, a first productive system for a bounded process 25,000 to 80,000 EUR. Operations are billed as a monthly flat rate. Offers far above that should be split into partial deliveries you can accept individually.

Do you build on n8n, Make, or custom development?

On whichever tool the case needs. Standard tools are fast and cheap for simple connections. Where load, data protection or complex logic enter, we build custom systems. In both cases we take over operations.

Do our data have to go into a US cloud?

No. We host in German and European datacenters, or fully on-premise at your site, with no outside connection at all if you want. Which of the three paths is right depends on how sensitive the data being processed is.

How fast is a first process automated?

After the assessment, a clearly bounded process takes four to eight weeks to production. Anything that takes much longer is scoped too large and should be split.

What happens if a model vendor changes prices or retires a model?

Then we switch the model. Our systems are built so the model is replaceable, and ongoing operations sit with us. Exactly this case is why operations are part of the offer.

Service · AI automation

AI automation for the Mittelstand.

We automate business processes with AI where it pays off, and we tell you where it does not. Built and operated by one German engineering team. Your data stays in the EU, or entirely inside your own building if you prefer.

Talk to an engineer

What we automate

AI process automation pays off wherever people make structured decisions over unstructured inputs: reading texts, checking documents, classifying cases, writing replies. The fields where this pays off most often in the Mittelstand:

Document processing
quotes, invoices, delivery notes and contracts read, checked and booked into your ERP automatically.
Email and inquiry handling
incoming messages classified, standard cases answered, the rest routed to the right desk.
Quote generation
drafts from product data, price lists and previous quotes. Sales reviews instead of typing.
Invoice checking
matched against orders and contracts; a human only sees the deviations.
Customer service
AI agents that answer standard questions and document the handover to humans cleanly.
Processes around SAP and ERP
automation at the edges of the system, through interfaces instead of modifications.

AI agents or classic process automation

Not every process needs a language model. If a decision can be written as twenty clear if-then rules, classic rule-based automation is cheaper, faster and easier to document for an audit. Intelligent automation therefore means, for us: the cheapest mechanism that reliably solves the case wins.

AI comes in where inputs are unstructured: free text, PDFs in shifting layouts, emails, contracts. These cases dominate in the Mittelstand, and this is where AI agents carry productive load today. We build both and have no reason to sell you the more expensive tool.

Where automation pays off, and where it does not

Whether a process pays off comes down to a single product: frequency times time saved per case. A task that comes up daily and costs a quarter of an hour each time carries itself quickly. A task that happens twelve times a year does not pay back through repetition, however tedious it is. The full calculation with numbers is in automating business processes.

Two further factors decide not whether to automate but how. The more stable the rules, the more can be solved without AI. And the more expensive a mistake, the more a human review step belongs in the flow, which consumes part of the saving again. Processes with high error costs we therefore automate up to the decision, not through it.

Three cases where we regularly advise against: the data does not exist but lives in heads and inboxes, then the first project is a data project. The process changes fundamentally every quarter, then you would be automating a moving target. And there is no internal owner, then nobody will use the system after launch. A vendor who sells you something anyway in these cases is selling you a problem.

Operations are part of the offer

An AI system is not a machine you set up once. Model vendors ship new versions, change prices, retire models. Someone has to notice, evaluate and follow up, or your automation stands still within eighteen months.

We run what we build: monitoring, updates, backups, on-call with defined response times. Billing is a monthly flat rate, not per seat. Your costs do not grow with your headcount.

How a project runs

Assessment: two to four weeks. The result is a written document with prioritized use cases, estimated effort and a recommendation where to start. Not a slide deck.

Pilot: one clearly bounded process, four to eight weeks to production. Afterwards you decide based on measured numbers whether we continue. Operations and further development run as a flat rate, in weekly increments you can click through.

Common questions about AI automation.

What does AI automation cost?: Assessment and potential analysis run 5,000 to 15,000 EUR, a first productive system for a bounded process 25,000 to 80,000 EUR. Operations are billed as a monthly flat rate. Offers far above that should be split into partial deliveries you can accept individually.
Do you build on n8n, Make, or custom development?: On whichever tool the case needs. Standard tools are fast and cheap for simple connections. Where load, data protection or complex logic enter, we build custom systems. In both cases we take over operations.
Do our data have to go into a US cloud?: No. We host in German and European datacenters, or fully on-premise at your site, with no outside connection at all if you want. Which of the three paths is right depends on how sensitive the data being processed is.
How fast is a first process automated?: After the assessment, a clearly bounded process takes four to eight weeks to production. Anything that takes much longer is scoped too large and should be split.
What happens if a model vendor changes prices or retires a model?: Then we switch the model. Our systems are built so the model is replaceable, and ongoing operations sit with us. Exactly this case is why operations are part of the offer.