Partnership with JADE IT

An AI system has not been introduced just because it is technically in place.

When companies use AI, the discussion usually starts with models, infrastructure, data protection, and data sovereignty. That is justified. If internal documents, customer data, or confidential business knowledge are processed, the technical setup is not a side issue. It must be clear where data is processed, who has access, and how the system is integrated into the existing IT environment.

But the practical value often only becomes clear later: in the day-to-day work of the people who are expected to use the system.

Many difficulties do not arise because an AI system fundamentally fails to work. They arise because the context of use has not been clarified properly. Who is allowed to use which information? Which outputs need to be reviewed? Which departments should start with which use cases? Who maintains internal knowledge sources when documents, processes, or responsibilities change?

This makes AI implementation a question of roles, responsibilities, and processes as well. Technical safeguards remain important, but they are not enough on their own. In established approaches to AI risk management, such as the NIST AI Risk Management Framework, context of use, organizational responsibilities, training, and feedback are explicitly part of the overall picture. For companies, this is a sober but important point: a system can be technically sound and still be poorly introduced.

Training is also less abstract than it often sounds. Most employees do not need to understand model architectures. What matters more is that they know how to work with AI in their own area: when outputs need to be checked, which information may be used, and where the system’s limits are. The OECD’s analysis of the AI skills gap makes a similar distinction between specialized AI skills and broader AI literacy.

Another common pitfall is the gap between a demo and everyday use. In a demonstration, a system can look convincing quickly. In actual operation, what matters is whether it fits existing processes, has enough context, and helps with the users’ actual tasks. Current implementation analyses such as the MIT NANDA report on generative AI adoption in business describe this gap: many initiatives fall short of expectations when they are not well embedded in work processes or miss the practical tasks users actually need to perform. The report is not a universal formula for success, but it points to a problem many companies already know from other software projects: good technology is no substitute for proper implementation.

With AI, poor implementation becomes visible particularly quickly. Language models can produce outputs that seem plausible at first glance. At the same time, quality and reliability depend heavily on which information is included, how questions are asked, which sources are used, and whether outputs are reviewed with the necessary subject-matter knowledge.

For implementation, it is therefore not enough to hand out login details and give a short explanation of the interface. It is also about concrete working practices: Which documents belong in a knowledge base? Which department starts with which use case? Who decides on access rights? And how do employees learn not to take outputs at face value, but to assess them in context?

AI that runs locally or in controlled infrastructure can make data flows easier to contain and control. But it does not automatically answer the organizational questions that arise during a project. These questions need to be addressed deliberately if the system is not merely to be available, but to become useful in practice.

With Airene, IOWIS provides the technical foundation for AI that companies can run under their own control. Airene is designed to make modern AI capabilities usable in the company’s own infrastructure, with a focus on data sovereignty, confidentiality, and controlled data processing.

JADE IT complements this technical side where projects need more structured support: in project management, change management, and user training. This also includes involving the relevant departments, incorporating feedback from day-to-day use, and helping employees adapt to new ways of working.

This means implementation does not stop at technical deployment. It also takes into account the organizational questions that arise later in day-to-day work.

How much support an AI project needs depends on the specific use case. A clearly defined use case with only a few user groups requires a different implementation approach than a project involving several departments, internal knowledge sources, and established processes.

What matters is that technical deployment and organizational implementation are considered together. When employees understand what they can use AI for, where the limits are, and how outputs should be assessed, operation becomes more practical and easier to control.

Working with JADE IT is therefore a natural step for us: Airene creates the foundation for AI that companies can run under their own control. JADE IT helps bring that foundation into day-to-day work in a structured way where implementation, coordination, and training play a larger role.