Organisations hold enormous amounts of knowledge spread across documents, emails, wikis and the minds of employees. AI makes it possible to structure and make that knowledge searchable. Here is how that works in practice.
A consultancy of 60 staff had a knowledge problem: years of expertise were spread across thousands of documents, email threads and Confluence pages that nobody maintained. New employees learned by searching, experienced staff were repeatedly asked the same questions. The solution: an AI-driven knowledge base.
Knowledge problems in organisations typically have the same causes. Knowledge is created but not stored in a structured way. Documents become outdated without anyone flagging this. Internal system searches return too many results of varying relevance. And the most valuable knowledge is not in documents, but in the minds of experienced employees.
The result: employees spend an average of one and a half hours per day searching for information. They ask colleagues questions those colleagues have already answered before. And when experienced employees leave, their knowledge goes with them.
The consultancy opted for a phased approach in three steps.
Phase 1: Inventory and prioritise knowledge sources Not all documents are equally valuable. The first step was an inventory of all knowledge sources and determining priority based on usefulness and currency. Outdated documents were marked as "to archive". Current, relevant documents were selected as source material.
Phase 2: Build the knowledge base with RAG Based on the selected documents, a Retrieval-Augmented Generation (RAG) system was set up. This is an architecture where an AI model answers questions by retrieving relevant pieces from a document set and using them as the basis for the answer.
The system could answer questions like: "What is our approach to change management projects?" or "Which client cases do we have in the logistics sector?" Answers were based on actual internal documents, not generic AI knowledge.
Phase 3: Integrate access and use into work processes A knowledge base that nobody uses is worthless. The third step was integrating it into daily work processes: the knowledge base became accessible via an internal chat interface, connected to the existing communication environment (Microsoft Teams).
The knowledge base performed less well for implicit knowledge: the unwritten rules, the culture-specific approaches and the nuances that experts had built up over the years but never written down. That knowledge could not be digitalised without interviews and active knowledge documentation.
Keeping content current also remained a challenge. Documents in the knowledge base became outdated, and the system had no way to automatically flag when information was no longer accurate. A periodic review process was necessary.
A knowledge base is not a one-time project. After the initial setup, structural maintenance is needed: adding new documents, removing outdated information, incorporating user feedback. The firm appointed a knowledge manager, an existing employee with a part-time role, to coordinate this maintenance.
AI-driven internal knowledge management works well for making existing documentation searchable and reducing repetitive knowledge questions. The initial setup requires investment in structuring and prioritising source material. Maintenance is structurally needed.
Mach8 helps organisations set up AI knowledge bases and internal automation. View our AI agents service or get in touch.
We help you go from strategy to implementation. Schedule a no-obligation call.
Schedule a call