Three Assets, One Unsolved Problem

Enterprises have spent decades building three distinct categories of intelligence assets, often without thinking of them that way.

The first is knowledge: the curated, deliberate capture of human expertise. Consider your company policies, processes, taxonomies, best practices and institutional memory. This is what your most experienced people know, the hard-won institutional intelligence that took years to build.

The second is content: the living output of your organization's daily work. Content like proposals, reports, presentations and contracts all fit this as they’re created continuously by end users without much governance. However, it’s full of implicit intelligence, such as decisions made, patterns established and lessons learned.

The third is data: the structured transactional record of your operations. This is the layer most organizations have heavily invested in, and the one they logically assume AI runs on. It might be your customer records, financial history, project metrics or operational logs.

The uncomfortable truth is that AI needs all three. And most enterprises have not prepared any of them for what AI actually requires.​

The Missing Layer

Here's what I've observed across organizations: investments in knowledge, content and data are almost always treated in isolation. Each has segregated ownership, its own governance model, its own tools and its own definition of "ready."

What none of them has built is the layer that connects them—the semantic and context layer that tells AI not just what the data says, but what it means. How that customer record relates to an open project. Which policy governs that transaction. Which subject matter expert owns that domain. How a proposal written two years ago connects to the engagement currently underway.

​Without that connective tissue, AI doesn't reason. It pattern-matches on fragments. And when it can't find the context it needs, it fills the gap—with confident-sounding answers that are partially or entirely wrong.

We've seen this play out directly. A professional services firm deployed an internal generative AI chatbot to help employees get fast answers. The model was capable, but the knowledge feeding it was scattered across tools: unstructured, unlabeled and missing the context the AI needed to reason accurately. The result was hallucinations, fabricated client details and inconsistent answers. As the system hallucinated, employees lose trust, defaulting back to manual work and forcing a multi-million-dollar investment to sit largely unused.

​The model wasn't the problem. The intelligence foundation beneath it was.

Why This Keeps Happening

According to S&P Global research, 42% of enterprises abandoned most of their AI initiatives in 2025. The average organization scrapped nearly half of its AI proof-of-concepts before they reached production. Why? Well, citing 43% of organizations surveyed, Gartner has found that data quality is one of the top obstacles to AI success.

I'd push that framing further. Data quality is part of it. But the truth is the knowledge, content and data layers are each unready in distinct ways and the semantic layer connecting them doesn’t exist.

Curated information in a well-maintained system is valuable but not sufficient. If the AI model can't access the context around this information, that value stays locked, degrading AI output rather than improving it. The failure isn’t in the AI model; it’s the readiness of context-rich knowledge, content and data that fuels it.

Four Steps To Start Getting It Right

Organizations making true progress on AI readiness are focusing on pragmatic, sequenced investments that build their intelligence foundation incrementally. Here's how the best of them approach it:

Take Inventory Before You Invest Further

Before adding another AI tool, understand what you're actually working with. Map out your three intelligence assets and understand where knowledge lives, what content you run on and where critical data resides. Most enterprises discover significant gaps and overlaps they weren't aware of. That visibility is the starting point for everything else.

Assign Ownership Across All Three Asset Classes

Knowledge, content and data each need a designated owner: someone accountable for the quality, governance and AI-readiness of each asset. In my experience, this is the highest leverage move an organization can make as content is where ownership is most consistently absent, often defaulting to end users who lack training and appropriate incentive.

Build The Semantic And Context Layer Deliberately

Most organizations skip this work because it doesn’t create an impressive demo, but it’s crucial to AI performance. Enrich and classify your information with metadata reflecting relationships, roles and relevance to help AI better understand what it’s reading. This is an ongoing discipline, compounding in value as your AI capabilities grow.

Connect Before You Expand

The instinct when AI underperforms is to add more tools, models and data sources. Instead, connect what you already have. Most organizations sit on plenty of assets which would dramatically improve AI performance if made visible, structured and linked through a coherent intelligence layer. Get that right before expanding.

Solving the chief problems with AI comes from fixing its consumption, not its capability. The key is not investing in the newest AI tool, but by starting an honest assessment of the intelligence foundation your organization is asking it to run on.

As AI progresses, the organizations getting this right won’t just have better outputs, they’ll have a durable advantage in the market. After all, the intelligence layer they build doesn’t just serve today’s AI tools; it becomes the solid foundation for whatever comes next.

Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?