Artificial Intelligence
Most AI projects do not fail because of the model. They fail because of permissions, interfaces, server performance and the question of where the data ends up. We have been looking after IT infrastructure in Cologne since 1992 and develop our own software – both from a single source, so none of these questions is left open.
The starting point
Private accounts, private devices, company data in third-party chat windows: that is what AI looks like in most SMEs today. A ban without an alternative only moves it to the phone in someone's pocket – it does not make it any more visible.
The alternative is not a major project. It starts with one task that regularly costs time, and a test with your real data. Only once that test produces a figure is money invested. And we will explicitly make a recommendation even when the result is: it is not worth it for this.
What sets us apart from a pure software company: we do not just build the application, we also run the server it runs on – the same company that looks after your firewall, your backup and your phone system.
Our AI services
Most businesses start with the AI chat because it reaches the largest part of the workforce. Agents pay off once a process is clearly described and happens often enough.
A chat interface under your name, on your server, with access to the documents you release. Answers with source references, permissions per group, sign-in with your existing accounts. Setup, connection of the first knowledge sources and staff induction are included.
See the AI chatNot just answering, but processing tasks: read the enquiry, look it up in the ERP system, create a draft, present it to a person for approval. The MCP server is the only point at which the AI reaches your systems – with permissions per role and a complete log.
See AI agentsReading receipts, assigning cases, making twenty years of files searchable – built into the screen where the work happens anyway. With a review threshold: anything the AI cannot recognise with confidence is put to a person instead of being posted.
See examplesIntroducing Microsoft 365 Copilot without it finding files nobody meant to share. We check the sharing settings first, set up a pilot group and provide an AI policy and proof of training under Article 4 of the EU AI Act.
See governanceScope of services
This table is here because otherwise you would ask about it in the first meeting. If you state a clear boundary, you do not need the word “full service”.
Decision guide
The three options are not mutually exclusive. For most businesses, the sensible combination is Copilot for everyday Office work and your own AI chat for everything that should not leave the building.
Process
Four steps, and you can stop after any one of them. After the second, you know whether it works – and you have paid a fixed price for that, not an open-ended bill.
We look at which task in your business really costs time. The result is an assessment of whether AI would change anything – even if the answer is no. The initial consultation is free of charge.
A clearly scoped test with your real data, measured against a metric agreed in advance: accuracy per field, share of questions with a usable answer, time per task. At a fixed price, so the decision does not depend on the risk.
One group works with it while we measure and make improvements. Only then does management decide on the rollout – based on experience rather than expectation.
Updates, model changes, monitoring and one contact person – under the same managed services contract as the rest of your IT. After twelve months a language model is no longer state of the art, so switching belongs in the contract, not in a follow-up project.
To be honest
For many tasks the benefit can now be demonstrated; for others the technology is simply the wrong tool. We tell you before the quotation, not at acceptance.
What we advise against
Calculations – language models do not calculate, they estimate; figures belong in the specialist application. Decisions without approval, especially anything that moves money or affects people. Legal and tax advice without expert review. A rollout before permissions have been cleaned up: an AI will find every file that has been shared carelessly. And a project without a metric agreed in advance – what cannot be measured cannot be signed off.
Frequently asked questions
Not with the AI chat on your own server: the model and documents sit on a server that we run for you. If you also want to use a model from an external provider – say, because it is better for a particular task – we tell you beforehand which data would go there, and you decide.
No. Your documents are searched and passed to the model as context; they are not trained into the model. The model itself remains unchanged. That is exactly why you can switch models at any time without losing knowledge.
That depends on the size of the model. Smaller models run on existing servers; larger ones need a suitable graphics card. In the feasibility test we measure this against your task and tell you before you buy anything. As a rule, we do not replace anything that still works.
It will, occasionally. That is why we build every application so that the source is shown and a person approves every step that changes anything. An AI whose output nobody can check has no place in a business.
The initial consultation is free of charge; the feasibility test is at a fixed price. After that, the price depends on the number of users, the number of data sources and whether we provide the server. We give you the range in the consultation – it becomes reliable after the test, not before.
Yes. We are not tied to any model and we do not sell one. The task decides which one is used – and, for confidential documents, the question of where it runs.
The question is not the number of employees but the number of repetitions. A process that happens twenty times a day pays off in a business with fifteen employees just as it does in one with three hundred. If it happens twice a month, the setup is not worth it.
Further reading
Service
Custom software, interfaces and apps – where the AI gets built in.
ReadService
Second opinions, strategy and GDPR – including on other providers' AI offerings.
ReadGuide
The question of location, which also determines how an AI is run.
ReadGuide
What EU regulation requires of the software in your business.
ReadService
Licences, setup and support – the foundation for Copilot.
ReadFree initial consultation
Tell us which work in your business regularly eats up time. We will tell you whether artificial intelligence would change anything, roughly what it would cost and where to start.
+49 221 984300-0Switchboard and support hotline
[email protected]Reply within 4 hours on working days
Robert-Perthel-Straße 7250739 Köln – Bilderstöckchen
Mon–Fri 9 am–6 pmEmergency support outside these hours by arrangement
This website works without cookies, without tracking and without advertising networks. For one optional feature we embed content from an external provider. Doing so transfers your IP address to that provider – which only happens if you allow it.
