Is Your Data GenAI Ready for Data Executives

Key Takeaways

• Context is Crucial for AI Reliability: AI hallucinations primarily occur when systems lack sufficient business context and consistent definitions, making proactive knowledge and information management essential.
• Semantic Layers and Context Graphs: While semantic layers standardise what terms mean, organisations must also build “context graphs” to capture the historical timelines and rationale behind specific business decisions.
• Transitioning Tacit Knowledge: Because AI cannot autonomously deduce hidden human agendas or unwritten rules, enterprises must urgently extract and document tacit knowledge from Subject Matter Experts before institutional knowledge is lost.
• Implementing “Headless BI” Architectures: To prevent autonomous agents from misinterpreting raw relational schemas, organisations must route AI queries through governed semantic APIs instead of allowing direct access to SQL databases.
• Defining Atomic Data Quality: Preventing AI from encountering contradictory metrics across different platforms requires defining data quality dimensions—such as completeness and accuracy—at a meticulous, atomic level.
• Enforcing Governance and Disambiguation: Robust AI governance relies on resolving terminological ambiguity by applying strict qualifiers to data attributes, ensuring terms like “churn” are explicitly defined by their specific domain.
• Elevating Data via Knowledge Graphs: Although relational databases are needed for structural integrity, layering a knowledge graph over them attaches essential, time-sensitive metadata directly to data nodes so AI can accurately resolve ambiguous entities.

Webinar Details

Title: Is Your Data GenAI Ready for Data Executives
Date: 2026-02-26
Presenter: Howard Diesel
Meetup Group: African Data Management Community
Write-up Author: Howard Diesel

What does Achieving Artificial Intelligence entail?

Achieving artificial intelligence (AI) readiness demands rigorous data management and strict business alignment. While generative AI is frequently criticised for “hallucinating,” human personnel have historically exhibited similar behaviours by presenting contradictory figures lacking standardised context—a phenomenon previously termed a lack of a “single source of truth”.

AI hallucinates primarily because organisations fail to provide it with sufficient context. To mitigate this, enterprises must look beyond mere data architecture and prioritise information and knowledge management, which structures the underlying business meaning of the enterprise’s data.

Figure 1 Is Your Data Gen-AI Ready?

What are Context Graphs and How Can They Help Inconsistent Semantics?

Enterprises frequently suffer from inconsistent semantics, where a single term such as “region” can denote a geographic area in one system but a sales territory in another. This misalignment fundamentally undermines AI reliability.

To address this, organisations are constructing semantic layers—ontologies situated atop graph databases—to universally define terminology. However, while a semantic layer outlines definitions, it lacks a “context graph”. The context graph introduces crucial historical and temporal dimensions, elucidating how and why specific business policies and decisions were formulated at given points in time.

Figure 2 Universal Semantic Layer for Confident AI

How is Capturing Tacit Knowledge Hindered?

Capturing institutional context is complicated by human factors, such as concealed agendas or decisions driven by dominant personalities, which a standard knowledge graph cannot autonomously detect. Consequently, organisations must actively extract tacit knowledge from Subject Matter Experts (SMEs) and translate it into explicit, documented guidance to instruct AI systems.

This process mirrors expert systems, where human specialists continuously validate or correct machine decisions to establish accurate operational precedents. To sustain this capability, enterprises must formally reinstate the “Knowledge Manager” role to ensure the systematic collection and preservation of vital operational context.

Figure 3 Module 6: Knowledge Management & AI Governance

What is AI-Ready Enterprise Architecture?

Establishing an AI-ready enterprise requires the comprehensive alignment of business, information, data, technology, governance, and knowledge architectures. A significant emerging risk is the deployment of “agentic AI”—autonomous agents that frequently encounter a “context wall”. These agents cannot independently deduce unwritten business rules or derive profound contextual meaning directly from relational database schemas.

Furthermore, allowing agents to query isolated data domains often yields conflicting terminology. The prescribed architectural solution is “headless BI,” which mandates that AI agents access information exclusively through a governed enterprise semantic layer rather than executing direct SQL queries against raw databases.

Figure 4 From Strategy to Execution

Figure 5 Course Agenda (4+1 Days)

Figure 6 Learning Objectives

Figure 7 The Context War

Figure 8 The Arc of the Argument

Figure 9 Welcome to the Third Innings: The Era of Scrutiny

What is the Context Gap, and how does it Create Data Quality Challenges?

The “context gap” is vividly illustrated when disparate systems calculate identical metrics differently. For instance, querying a revenue figure might yield “sales minus returns” in Tableau, incorporate discount deductions in Power BI, or simply aggregate all sales in Python. Such discrepancies are critical vulnerabilities for AI integration.

Trustworthy AI necessitates precise, atomic-level definitions of data quality dimensions, such as completeness and accuracy, meticulously tailored to specific operational contexts. Implementing standardised vocabularies, such as conformed dimensions, enables dynamic, runtime evaluations of data quality, ensuring that AI agents evaluate fitness-for-use before executing data products.

Figure 10 The “Context Gap” Breaks Agentic AI

Figure 11 The Bridge: Defining the Semantic Layer

Figure 12 The Dialectic: Automated Intelligence Vs. Human Nuance

What is the Illusion of AI Magic and How Can Semantic Layers Help?

A pervasive misconception among “AI optimists” is that Large Language Models (LLMs) can spontaneously synthesise accurate business context simply by ingesting corporate communications and raw schemas. In reality, AI inherently misses the rationale behind historical data anomalies—such as excluding a customer cohort due to a failed promotion—because SQL cannot articulate these hidden business rules.

Consequently, enterprises must construct standardised semantic layers to govern AI interactions. Similar to historical mandates requiring personnel to utilise Business Intelligence (BI) tools rather than raw database queries, AI agents must be compelled to access data exclusively via governed semantic APIs to prevent interpretive errors.

Figure 13 Thesis: AI Accelerates the Assembly

Figure 14 Antithesis: the ‘Tacit Knowledge’ Wall

Figure 15 The Verdict: Semantics are Solvable; Context is Infinite

Figure 16 The Semantic Layer is the API for Agents

Figure 17 The Hybrid Model: Human-Guided, AI-Accelerated

Figure 18 The Landscape: Semantic Spheres of Influence

Figure 19 The 2026 Outlook: the Tooling Gap

How can Humans Apply Better Guidance for AI?

Strategically, enterprises must recognise that AI cannot autonomously generate customised organisational ontologies, nor can standard industry data models adequately substitute for highly specific internal terminologies. The critical imperative is to proactively extract tacit knowledge from subject matter experts and formalise it into explicit documentation before institutional attrition occurs.

Data professionals must enforce stringent architectural controls by implementing governed APIs and semantic layers, explicitly shielding autonomous agents from direct database access. Meeting AI at the correct architectural juncture with precise context is essential for reliable automation.

Figure 20 Strategic Takeaways

Figure 21 The API Gateway

Figure 22 AI Agents Require a “Shared Substrate” to Avoid Hallucination and Metric Drift

What does AI Accountability, Governance, and Resolving Ambiguity look like?

Unlike human employees, who are subject to traditional Human Resources governance and accountability structures, autonomous AI agents currently operate without inherent behavioural controls. Emerging regulatory frameworks, such as the EU AI Act, are compelling enterprises to implement rigorous AI governance and operational literacy programs.

A fundamental governance requirement is the resolution of terminological ambiguity; for example, distinguishing “customer churn” from “employee attrition”. This is achieved by appending strict qualifiers to data attributes in accordance with ISO standards, thereby forcing user prompts and AI logic to establish precise domain context before retrieving sensitive or complex information.

How will Entity Resolution and Graph Databases help AI Agents to Deliver Real-Time Operational Value?

For AI agents to deliver real-time operational value, they must seamlessly execute complex entity resolution, canonicalising disparate identities (e.g., distinguishing multiple iterations of “Robert Smith”). Traditional relational databases are insufficient for this task, as they trap data in static tables, frequently leading to overloaded structural meaning and a profound lack of historical context.

Conversely, a knowledge graph explicitly attaches enriched metadata, ontologies, and temporal dimensions directly to data nodes. While relational databases remain essential for enforcing structural referential integrity, layering a knowledge graph atop them supplies the indispensable contextual “colour” required for accurate AI decision-making.

Figure 23 The ‘Sweet Spot’: Entity Resolution and the Thesaurus

Figure 24 The Technology Underpinning: RDF & Knowledge Graphs

Figure 25 Lever 2: the Context Layer (the Decision Graph)