Governance / Service knowledge
An AI assistant is only as good as the knowledge it is given.
Experienced service engineers retire, and what they know leaves with them. AI assistants are meant to bring that knowledge to the next engineer at a machine in the field. That only works if someone decides which knowledge the assistant may trust. I lead that work for the division’s service knowledge.
My roleSince August 2026 I also lead AI knowledge governance at ABB Motion, alongside my role as Software Design Engineer, AI. I define the rules and the evidence; the systems that carry them are decided with architecture and IT.
The knowledge comes from articles written over many years, from product data and, more and more, from AI. Each kind needs a different amount of checking.
The principles
- Trusted means there is enough evidence for what the knowledge is used for. Its status stays visible and can change.
- Evidence comes from several places: automatic checks, cases where it helped, a review when a case reopens, and test questions with answers an expert has checked. None of them proves correctness on its own.
- Knowledge written by AI gets at least one human check before its use can count as evidence.
- A review is small, one item and one or two questions, so an expert can answer it in minutes.
- The checking depends on the kind of knowledge and its risk. A parameter is compared with its source; a safety instruction needs more than one person.
- An outdated article keeps its history and is no longer recommended.
The framework is a working draft inside ABB. This page shows its principles only: no systems, names, numbers or results.