Designing data for humans and AI: The knowledge engineering imperative

As organizations move from experimenting with artificial intelligence to deploying agentic AI at enterprise scale, many are encountering a shared constraint. Data foundations built for decades of human interpretation were not designed for machines that must reason, decide, and act autonomously.

Dashboards, spreadsheets, and reports rely on human judgement to supply context and meaning. AI agents cannot make those same assumptions. To operate effectively, they require information that is structured, semantically rich, and grounded in business definitions.

This challenge is accelerating. With nearly 80 percent of enterprise information considered unstructured¹, AI agents struggle to interpret meaning, relationships, and intent at scale, limiting organizations’ ability to move beyond pilots into sustained value creation.

Knowledge engineering — the discipline of capturing and structuring an organization’s information so machines can interpret and use it effectively — can provide the answer. 

A paper published by KPMG in the US, The knowledge engineering imperative, examines why this challenge has become central to enterprise AI strategies and why knowledge engineering is foundational to scaling agentic AI with confidence. The summary below highlights selected themes explored in greater depth in the full report.

As AI models improve and become more accessible, technical capability alone is no longer the primary differentiator. For many organizations, scaling enterprise AI now depends on whether AI agents understand business context well enough to apply judgment.

Humans derive meaning through experience, shared language, and institutional knowledge. AI agents require that meaning to be made explicit through structure, relationships, and rules.

Without this semantic foundation, AI systems may generate technically valid responses that appear plausible but are not dependable for enterprise decision-making.

Knowledge engineering represents a shift from making data available to enabling informed action. Rather than focusing only on retrieval or reporting, it allows AI agents to interpret information in line with organizational intent.

An AI assistant may be tasked with recommending the best supplier for a new product line. Although it can easily retrieve supplier data, without understanding how the organization defines terms such as “preferred supplier,” how risk tolerance varies by region, or how exceptions are handled, it cannot reliably make decisions.

By making business nuances machine readable, knowledge engineering enables AI agents to reason across information, assess implications, and support decisions with greater consistency and transparency.

The research identifies three core knowledge engineering elements that work together to make organizational knowledge explicit and usable by AI agents.

• The ontology creates a common enterprise language so both humans and AI systems interpret terms consistently, enabling accurate reasoning and decision-making.
• The knowledge graph connects data and meaning across the organization, enabling AI agents to traverse relationships, infer context, and support more complex, enterprise-wide reasoning.
• The semantic layer provides a shared business view of data by connecting technical data to consistent enterprise definitions, helping to ensure that humans and AI agents interpret information the same way across systems.

Together, these elements help AI agents move beyond pattern recognition toward decisions grounded in how the organization actually operates.

Knowledge engineering is not about perfecting data or one-time architecture decisions. It is a strategic foundation that supports trust, explainability, and scalability as AI adoption deepens.

Traditional approaches to data management often focus on completeness and accuracy in isolation. Knowledge engineering shifts the focus to help ensure that the right information, grounded in business meaning, is available at the right moment to support decisions.

According to the report, organizations that treat knowledge engineering as an enterprise capability are typically better positioned to move from disconnected AI use cases toward coordinated, AI-enabled workflows.

The full paper explores the architectural considerations, implementation priorities, and organizational implications required to establish knowledge engineering as a durable foundation for agentic AI.

^{1.  CDO Magazine, “Unstructured Data: The Hidden Bottleneck in Enterprise AI Adoption,” CDOMagazine.tech, March 27, 2025.}