Sean Nathaniel is CEO of Upland, a leader in AI-powered knowledge and content management software trusted by 1,100+ enterprises worldwide.
The symptoms look different depending on where you sit. Data and technology leaders approved AI investments, assuming their data foundation was enough. Contact center and IT support leaders see it more directly: Agents get AI-assisted answers that cite superseded policies. Self-service tools contradict themselves. Customers escalate because the bot couldn’t synthesize across related issues.
Same problem, different view. The AI isn’t broken. The knowledge foundation beneath it isn’t ready.
In a prior article, I argued that enterprises have been feeding AI three distinct asset classes—data, content and knowledge—without preparing any of them for AI use or building the context layer that connects them. This piece focuses on one of those assets specifically, and why it keeps getting left out of the readiness conversation entirely: knowledge.
Why Data Alone Falls Short
Data readiness investments were rational. Structured data is the layer most organizations know how to govern, and it matters. But here’s what it can’t tell AI: what it means. A customer record tells AI what happened. It doesn’t tell AI how that relationship was built, which business rules govern the account or what institutional judgment shaped the last major decision about that engagement.
That context lives in curated organizational knowledge—the policies that govern how work gets done, the processes that encode institutional learning, the taxonomies that define how the organization thinks about its domains and the expert directories that map who knows what. These assets are the organization’s understanding of itself, made explicit. They give AI something data can’t: the context to interpret information, not just retrieve it.
When organizations conflate data readiness with knowledge readiness, the AI can access the records but not the judgment behind them. So, it improvises, and confident-sounding improvisation at AI speed and scale is where hallucinations come from.
The Knowledge Readiness Gap
Most knowledge management systems were designed for humans, not AI. A person types a query, browses results, reads an article and applies judgment. AI can’t do that. It needs the semantic layer that human readers supply instinctively: which policy supersedes which, which expert’s guidance applies in which situations and how one knowledge asset relates to another. Without it, AI sees an unstructured pile of text rather than a coherent map of how the organization thinks.
I’ve seen this play out in contact centers directly. An organization deploys agent-assist AI on top of a well-maintained knowledge base. Agents trust it for simple questions. But when a situation requires applying a recently updated procedure or synthesizing across two related policies, the AI blends old and new guidance into something neither version actually said. The knowledge is there. What’s missing is the semantic layer that tells AI which article applies, when it applies and which version is current. The agent gets a confident, wrong answer, and the customer pays for it.
Knowledge bases also rot. Policies get updated, but old versions stick around. Expert directories go stale. Without governance built with AI in mind, even a well-constructed knowledge system becomes a liability.
Four Steps To Making Knowledge AI-Ready
Existing knowledge assets don’t need to be rebuilt. They need to be enriched, connected and governed with AI consumption in mind.
1. Audit what you have before assuming it’s ready.
Most organizations overestimate their knowledge systems’ AI-readiness. Assess the current state: How current is the content? Is metadata applied consistently? Are relationships between assets explicitly mapped, or implicit and invisible to AI? The audit almost always reveals a knowledge base more valuable than anyone realized, and less ready than anyone assumed.
2. Enrich with the semantic layer that AI needs.
The highest-impact investment is metadata enrichment: adding structured context that tells AI how to interpret and relate knowledge assets. Tag content with roles, domains, relationships and hierarchy. Map which policy governs which process, and which expert owns which domain. This is a discipline problem, not a technology problem. The tools exist; the organizational commitment usually doesn’t.
3. Connect knowledge to the content and data around it.
A policy connects to the contracts it governs. An expert directory connects to the projects that the person contributed to. A process document connects to the operational data that measures whether it’s working. Making those links explicit transforms a knowledge base from a reference library into something AI can reason across, not just retrieve from.
4. Govern for AI, not just for humans.
AI doesn’t browse; it ingests. Stale content that a human recognizes as outdated gets treated as current. Conflicting entries that a human resolves through judgment will confuse an AI trying to synthesize. That means regular review cycles, explicit retirement of superseded content and ongoing validation of the semantic layer. This is what the knowledge-centered success (KCS) methodology was built for—the workflow and accountability structure that keeps knowledge current at scale. Organizations that have adopted KCS are better positioned for AI readiness than those governing knowledge informally.
The Foundation That Your Data Work Was Missing
Your data readiness work wasn’t wasted. But it was incomplete. Data can be reacquired. The accumulated expertise of an organization—the hard-won understanding of how to do the work, what the rules are and what the exceptions mean—took years to build and can’t be quickly reconstructed if left inaccessible.
When that knowledge is structured with the context AI needs, connected to the data and content around it and governed to stay current, AI stops reflecting what it can infer and starts reflecting what your organization actually knows. That’s the difference between AI that employees trust and AI they ignore, and it starts with recognizing that data readiness was only half the job.
Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?
