AI that goes into production. Not into a presentation.
Most organisations have seen an AI pilot by now. Far fewer run AI in daily operation. The difference is rarely the model. It is data access, permission logic, error handling, cost control, and the question of what happens when the model returns nonsense.
We have run AI features in production in our own product for years, with paying customers, real failure cases and real invoices. That is a different craft from building a prototype.
What we actually do
Integrating language models into existing systems Document processing, classification, summarisation, extraction of structured data from unstructured text. Connected to your systems, with your permissions.
Search across your own knowledge (RAG) Making your manuals, contracts, tickets and minutes searchable and answerable — with source references, so answers stay verifiable, and with respect for existing access rights.
Process automation Automating recurring decisions and processing steps, with defined escalation to a person wherever there is uncertainty.
Operating AI and keeping the cost in check Token budgets, caching, model choice per use case, quality measurement in live operation, fallback strategies. An AI feature without cost monitoring is an open invoice.
AI governance and the EU AI Act Classifying your use cases against the risk categories of the EU AI Act, documentation duties, transparency requirements.
Data protection is not a side condition
For many of our clients — particularly in public administration and financial services — the decisive question is not what AI can do. It is where the data goes.
What we advise against
A chatbot on your website that knows nothing your FAQ does not already say. AI-generated content that nobody proofreads. And any project whose justification is that everyone else is doing it. That is why our potential assessment regularly also lists the ideas we do not recommend, with reasons.
Getting started: AI Potential Assessment
One week, fixed price. At the end you have a prioritised list of concrete use cases with effort estimates, expected benefit, a data protection classification and a clear recommendation on where to start.
We offer three operating models
| Model | Data | Suitable for |
|---|---|---|
| EU cloud | Processing in EU data centres, data processing agreement in place | most use cases |
| On-premise | Models run in your data centre, no data leaves the building | highly sensitive data, public sector, healthcare |
| Hybrid | Uncritical processing in the cloud, sensitive steps locally | cost control with mixed data sensitivity |