The most expensive mistake is the use case bought too early

The typical sequence looks like this: management sees a demo, budget is released, a pilot starts — and four months later it turns out that the data it was all meant to rest on sits in three systems with contradictory versions. The project is not cancelled, it is quietly extended. That is the normal case, not the exception.

The cause is rarely a lack of technical skill. It is that the feasibility question gets asked only once money is committed. A solid initial analysis costs a fraction of a failed pilot and answers three questions before budget flows: is the process worth it at all, is the data in a state that supports it, and is there someone who owns the decision.

What an initial analysis actually examines

The process itself: how often it runs, how long it takes today, who ultimately decides — and how you would recognise after 90 days that it has improved.
The data situation: where the necessary knowledge sits, how current it is, how many duplicates and outdated versions exist. Opening a sample by hand tells you more here than any system documentation.
Permissions and protection classes: who may see what. An AI answer must never reveal more than the person asking would be allowed to see in the source system anyway.
Ownership: is there a named business owner with decision-making authority and available time. Interest alone does not carry a pilot.

If one of these four questions is open, the outcome of the analysis is not “which model” but “close this gap first”.

Since August 2026 the EU AI Act is no longer a future concern

The regulation applies in stages. Since 2 February 2025 prohibited practices are banned and companies must ensure AI literacy among the people operating these systems. Since 2 August 2025 the rules for general-purpose AI models apply, along with the governance and penalty framework. Since 2 August 2026 the main body of the regulation applies; Article 6(1) and its obligations follow on 2 August 2027.

For most mid-sized adopters the decisive question is not whether they operate a high-risk system — that is rare. It is whether they can demonstrate which category their use falls into, and whether the people working with it have the required competence. That classification belongs in the initial analysis, not in a later round of panic.

A roadmap that shows a result after 90 days

We work in three blocks of 30 days. The first decides and secures: scope, baseline, data clearance, architecture decision. In the second an end-to-end prototype is built on the approved data set and measured against a test set frozen in advance. In the third the pilot runs with real users, and at the end there is a decision: continue, refine or stop.

The third block matters most because it explicitly allows for stopping. A pilot that is not permitted to fail is not a pilot but an expensive declaration of intent.

How we support you

Initial analysis with a written assessment: use case evaluation, data sample, permissions and roles, classification under the EU AI Act.
AI roadmap across 30/60/90 days with measurable acceptance criteria instead of milestone prose.
Ongoing guidance during implementation — including when the implementation sits with your own team or a third party.
Second opinion on existing proposals: we check whether a concept holds up before you sign.