No access to the operational data
Without a connection to ERP and business applications the answer stays generic — and therefore worthless.
AI where it
takes work off your hands.
The value of AI does not appear in a chat window but in the process: when documents are pre-sorted, enquiries routed and data from business systems brought together reliably. That takes its own infrastructure and clear boundaries.
Request a conversationYour own infrastructure
Models and data under your control
Where it runs
In your infrastructure or at a provider of your choice
Measurable
Every use case with a target figure
Starting point
AI is already in the building. It has not reached the process.
In many companies AI is already in use – individually, through private accounts, around IT. What is missing is the decision behind it: which content may leave the building, which models run where, who has access to what, and how AI reaches the systems where the actual work happens. Without that, AI stays a chat window next to the process rather than part of it.
Unclear processes
If it is not defined when a case is decided how, AI cannot decide it either.
No handling of errors
Without a review step and escalation to a person, productive use is not possible.
Data protection as an afterthought
Permissions, logging and data retention decide whether a solution may go into operation at all.
How we work
How we go about it.
Assess use cases
We collect candidates, estimate effort and effect, and pick the ones that hold up.
AI infrastructure
Central access to models with permissions, logging and cost control — instead of many individual subscriptions.
Connection to your systems
Through interfaces and an MCP server, AI gets controlled access to exactly the data a case needs.
A person in the loop
Suggestion, review, approval: decisions with consequences stay with your people.
Sequence
In what order.
Phase 1
Review
Use cases, data situation, legal framework.
Phase 2
Prototype
One case on real data, measurable against the current workflow.
Phase 3
Production
Infrastructure, permissions, monitoring and rollout in the department.
After that
Expansion
Further cases on the foundation already in place.
Outcome
An AI environment that can actually be run.
You get a basis for decisions and, if you want it, the implementation: which models run where, how access and logging are governed, which data leaves the building and which does not. Plus the connection to the systems where work happens. Swapping a model stays possible without rewriting the workflows.
- An assessed list of use cases
- AI infrastructure set up
- Connection to ERP and business applications
- Permission, logging and cost control
- Review and escalation paths
- Training for the department and for development
Frequently asked
What clients ask first.
- How long until we see a usable result?
- An assessment with a recommendation takes a few weeks. A first production workflow takes longer, because a dependable statement about quality only comes from your own data – which is why we begin with a trial on real transactions rather than a demonstration.
- Does our data have to leave the building?
- Not necessarily. Open models can run on your own hardware, in which case processing stays entirely with you. Where a vendor model is clearly better for the task, we put in writing which content goes there before implementation. We do not agree to your data being used to train third-party models.
- What does running it ourselves cost compared to using a vendor?
- Running it yourself shifts ongoing usage cost into one-off hardware and internal effort. The volume at which that pays off depends on the tasks and cannot be stated in general. We work it through for your case with the volumes actually expected, before anything is purchased.
- How is the work billed?
- Assessment and concept are bounded and can be quoted at a fixed price. Setup and integration are billed for time, because they depend on your existing infrastructure. Ongoing model and operating costs are stated separately so that the cost of running it stays visible.
- What do we need to provide internally?
- A contact in IT, real example cases – ideally including the difficult ones – and a willingness to define what AI may decide on its own and what goes for approval. That boundary is a business decision, not a technical one.
- Who operates and maintains the solution afterwards?
- Models and providers change faster than conventional software. We therefore build so that a model can be swapped without rewriting the workflows. We take on operation and monitoring of output quality on request – and equally the handover to you.