Knowledge audit before enterprise AI

Choose a recurring business decision or support task and name the people who need to make it. Ask what they must know, where a wrong answer creates risk and which evidence would make the answer trustworthy. The output is a small set of priority knowledge needs, not an indiscriminate catalogue of every document.

For each priority need, record the source, subject expert, content owner, intended audience and review trigger. Distinguish missing knowledge from existing knowledge that people cannot find. Identify restricted or personal information before changing access. A search or AI layer cannot itself decide who is accountable for a source or whether somebody is entitled to read it.

Take a bounded workflow, such as reusing a technical lesson or answering a recurring operational question. Agree representative questions, expected sources and acceptance criteria with its users. First evaluate whether the information is current, accessible and usable. If AI is in scope, separately evaluate the answers against those sources; a plausible answer is not evidence that the knowledge is correct.

Record a baseline before the pilot: time to find a trusted answer, unanswered questions or coverage of critical knowledge. Choose measures that the team can actually collect. Keep observed results separate from projected benefits. Expand only after ownership and maintenance routines work in the pilot.

At Gazprom Neft, Eugene Victorov developed knowledge mapping and critical knowledge-audit procedures as part of the upstream KM programme. At Arrival, his work included knowledge audits, mapping, a glossary and lessons learned. These are documented KM activities; this guide does not present them as production AI deployments.

• Enterprise Knowledge Management System
• Arrival — United Kingdom
• Knowledge Management Consulting