AI tools cannot entirely automate knowledge management, mainly because they are prone to hallucination, lack the tacit knowledge of humans, and struggle with source content when it’s fragmented and stale. The other obstacle to AI automation, in our experience, is organisational inertia and mistrust.
With well-designed constraints however, and staff involvement in the way enterprise knowledge is structured, AI tools can play a powerful supporting role in surfacing and organising relevant knowledge at scale, making context-aware recommendations, and maintaining data over time in a changing world.
What is knowledge management?
Modern knowledge management involves capturing and structuring an organisation’s knowledge in a way that AI systems can reason over, at scale. It also involves validating and updating knowledge, rather than leaving it to gather dust, which is a sure way to undermine trust in an AI’s outputs, even when it’s explainable.
Knowledge capture
Capturing enterprise knowledge ranges from assembling documentation such as scattered or siloed spreadsheets, to mining the implicit knowledge that sits in human experts’ heads (knowledge elicitation). While AI can help with these processes, including obtaining consensus around key definitions, it cannot entirely supplant the role of humans in knowledge capture. This is because so much valuable knowledge is tacit and, by definition, impossible to express in words.
Aside from providing the raw material for AI inference and automation, knowledge capture also enables expertise – of the non-tacit kind anyway – to be retained by an organisation when key staff depart.
Structuring and updating knowledge
Once knowledge has been explicitly recorded and brought together as a single source of truth, it needs to be verified by humans-in-the-loop. These people – both experts and users themselves preferably – must also agree on what important concepts mean in their different contexts. This addresses the problem of polysemy, which is a much harder challenge for AI than for humans.
Context is much harder for AI than for humans.
Humans also need to define the relationships between these concepts, and potentially explore how they can interoperate with other systems. AI tools, including Ontology as a service (OaaS) platforms, can help here by suggesting candidate concepts from the wider domain, and by drawing on public ontologies, using ontology standards.
With structured, machine-readable knowledge, AI can then explain its actions or decisions, based on the enterprise’s own rules, should the need for an audit arise.
What holds back AI automation of knowledge management?
Implementation costs and difficulties in measuring value-added can impede automation of knowledge management. Trust issues are also common, especially among technical staff who’ve had their fingers burned in the past by AI pilots that were too unreliable to scale. Unless there is a sense of ownership over AI, and a means to audit its outputs, enterprise leaders simply don’t have enough confidence to scale automation across more important tasks.
Tech staff who’ve had fingers burned by AI pilots in the past, tend to be wary the second time.
These concerns were raised in a 2026 study by Evangelista and Rizvi, which reviewed the benefits, implementation challenges, and ethical concerns associated with AI-driven knowledge management systems across industries.
How do I get measurable value out of knowledge management?
When properly deployed, AI tools can play a central role in operationalising knowledge; in other words, they can translate stored knowledge into trustworthy decisions and action at scale. The process of drawing out implicit knowledge from existing documentation (inference), and providing accurate, explainable answers to natural language queries, is in our view a fundamental requirement of trustworthy agentic AI.
And it all starts with structured and well-maintained knowledge.