When product information, prices, documents, and data sources drift apart, the result is inconsistent guidance, wasted search time, and unreliable AI answers. I bring that information into a current, traceable, and approved state. Employees find what applies more quickly, and AI can answer reliably.
In IT since 1992 · 25 years of those at Daimler / Mercedes-Benz · hands-on BotCore experience · AI Manager (WBS TRAINING) · DEKRA-certified AI Officer
Two situations bring people to me. Either the AI Act requires something that has to be evidenced. Or information across the organisation contradicts itself and AI is expected to answer reliably anyway. There is a suitable entry point for both.
What the AI Act requires, and how an organisation sets its own rules.
Leadership wants to get the topic moving, but nobody knows where to start. A talk without technical jargon: what AI actually delivers in an organisation of this size today, where it is of no use, and what a first step looks like. For a leadership circle, a staff meeting, or a customer event.
Since February 2025 the AI Act has required staff to understand the AI they use. Since 2 August 2026 it must also be evident when an AI is answering. One day on site, after which you hold a training record, an AI inventory, a draft usage policy, and named responsibilities.
Everyone does it differently, and there is no rule to fall back on. Approved tools, prohibited data, review duties, disclosure, and an escalation path for incidents. Agreed with leadership, data protection, and, where applicable, the works council. The outcome is an approved document that actually applies.
AI assistants such as Microsoft Copilot, agents, and chatbots already work on company data, often with permissions nobody has reviewed. In ten working days you get a verifiable picture: which AI is in use, who is accountable for it, what can be evidenced, and what needs to happen in the next 90 days.
Details on governance readinessWhen information contradicts itself and AI is expected to answer reliably on top of it.
Which sources exist, which are current, and which are authoritative? Where do product details, prices, or document versions conflict? Who may approve binding content? The check identifies the most important errors and risks, organizes the sources, and defines a sensible first pilot.
The pilot implements one defined area of knowledge. It includes automated ingestion, clear sources and permissions, AI answers with citations, and approvals based on risk. The result can be tested by both business and IT.
If the pilot works, further knowledge areas, sources, and use cases can follow. I work with IT leadership, architecture, and business owners to define quality criteria, responsibilities, and the operating model.
The aim is to manage knowledge reliably, from the original source to the approved answer.
Marketing, sales, product owners, and IT work with the same approved information.
New employees find current answers and their sources instead of piecing knowledge together.
Outdated versions, conflicting prices, and unclear statements become visible, reviewed, and deliberately archived.
Answers cite approved sources, show uncertainty, and can be tested systematically.
These four points guide my planning and implementation.
A plausible AI answer alone is of little use. It needs an approved source, a clear audience, and a known date or version.
Documents are not transferred into spreadsheets one by one. Ingestion, OCR, classification, version checks, fact extraction, and conflict detection are automated wherever it is sensible.
AI identifies anomalies and prepares decisions. Binding product statements, prices, and confidential content are reviewed and approved by designated knowledge owners.
A knowledge system includes sources, interfaces, permissions, approvals, archiving, and quality control. These parts need to work as one process.
In IT since 1992, 25 years of those at Daimler and Mercedes-Benz across changing projects and roles. Responsibility for business-critical systems and hands-on work with AI knowledge platforms.
Tell me briefly which documents, systems, or departments provide different answers. In an initial conversation, we will determine whether an AI Knowledge Readiness Check is the right next step.