Editorial
Research Report
Mind The Agentic Action Gap: Stop Losing Money With AI Agents
Every enterprise GenAI conversation eventually arrives at retrieval-augmented generation (RAG). The concept sounds straightforward: connect your large language model to your proprietary data so it answers questions about your business rather than generating responses from the internet’s general knowledge.
The appeal is real. But so is the gap between the concept and enterprise reality. RAG does not fix bad content. It amplifies it. If your content is disorganized, inconsistent or poorly structured, RAG will confidently serve up that disorder as legitimate answers. The AI does not know it is wrong. It sounds authoritative while delivering operationally useless results.
As Part 1 of this series established, the integration complexity of enterprise AI grows exponentially, not linearly. This article examines the foundational layer that manages that complexity: information architecture. There is no AI without IA.
Table of Contents
- The Retrieval Problem That Determines Everything Else
- What 'Context' Means in Enterprise Retrieval
- The Prompt as Metadata Container
- Is-ness and About-ness: The Classification Framework
- Use Cases Drive Architecture Requirements
- Building the Foundation: A Practical Sequence
- The Competitive Implication
The Retrieval Problem That Determines Everything Else
RAG operates in three stages.
- The system interprets the user’s query.
- It retrieves relevant content from enterprise data.
- It uses that content to generate a response.
Most organizations focus their investment on stage three (generation) while underinvesting in stage two (retrieval).
The math is unforgiving. If your retrieval accuracy is 70%, meaning 30% of the time the system pulls the wrong documents, your overall accuracy ceiling is 70%. It does not matter how sophisticated your language model is. The system is working with wrong inputs, and no amount of generative sophistication compensates for retrieving the wrong source material.
HBR Analytic Services research found that 39% of organizations cite data issues as their top challenge in scaling GenAI, and more than half rated their data foundation readiness at five or lower on a 10-point scale. These are not technology problems. They are information architecture problems that surface as AI failures.
Related Article: No Agents Without Architecture: Why Enterprise AI Fails Before It Starts
What 'Context' Means in Enterprise Retrieval
Context is the most overused and least operationalized word in enterprise AI. Vendors invoke it to describe semantic understanding and vector embeddings. But context for enterprise retrieval is far more specific, and far more dependent on work that has nothing to do with AI technology.
Consider three questions that a field service technician might ask an AI assistant:
- "How do I set up my modem?" Which modem? The HM 2900 series? Model 2960 or 2970? Residential or commercial installation? The AI needs to disambiguate, and it can only do that if the content is structured to support disambiguation.
- "Where is the installation guide?" For which product? Which version? Is the user looking for a quick-start card or a comprehensive manual? Self-installation or professional setup?
- "What does error code 50 mean?" For which product? Error code 50 might mean something entirely different on the 2960 versus the 2970. During installation, normal operation or a specific troubleshooting scenario?
Each question sounds simple. Each is impossible to answer without context that must be built into the content architecture before the AI encounters it. The AI does not understand these distinctions. Your information architecture instructs it.
The Prompt as Metadata Container
Here is a concept that transforms how you think about GenAI retrieval: every user prompt is a metadata container.
When a field service technician types, "I need to install a model 2960 modem, but I am receiving error code 50. Please provide the troubleshooting steps to complete my installation," they have embedded several metadata signals in the query:
- Content type needed: troubleshooting guide, installation instructions
- Product line: HM 2900 Series
- Model: 2960
- Error code: 50
- Process stage: installation
The AI uses these signals to retrieve relevant content. But the AI can only find content that has been tagged with matching metadata. If troubleshooting guides are not tagged by product, model and error code, the system cannot match the query to the correct answer.
This is why organizations with excellent AI models still get poor results. The model understands what the user is asking. The content is not structured to provide the answer. The user’s question contains implicit metadata. The content must contain explicit metadata that matches. Information architecture builds the bridge.
Is-ness and About-ness: The Classification Framework
For over two decades, EIS has used a framework that provides the starting point for content architecture: is-ness and about-ness. Originally developed in the context of product data management and unified commerce, this framework applies directly to the enterprise AI challenge.
Is-ness answers: what type of content is this? If you handed it to someone, what would they call it? A service procedure. A product specification. A troubleshooting guide. A policy document. An installation manual.
About-ness answers: if you had a thousand documents of this type, how would you tell them apart? Which product does it cover? Which process does it support? Which audience is it for? When was it last validated? What conditions trigger its use?
Is-ness gives you content types. About-ness gives you the metadata schema for each type. Together, they create the classification structure that enables precise retrieval.
Most organizations have is-ness roughly figured out. They know a troubleshooting guide when they see one. Few have about-ness systematically defined. That is why their AI retrieves five troubleshooting guides when it should retrieve one.
Use Cases Drive Architecture Requirements
Information architecture meets business strategy at the use case level. A simple formula defines the connection: "As a [role], I need to [action] so I can [outcome]."
For field service:
- "As a field service technician, I need to locate troubleshooting guides so I can service equipment quickly with minimal downtime."
- "As a field service technician, I need to know what spare parts are available so I can procure the necessary parts for my service call."
Each use case defines what metadata the content requires. The first tells you that troubleshooting guides must be findable by equipment type, symptom and error code. The second tells you that parts information must be linked to equipment, with availability status and procurement paths.
Multiply this process across dozens of use cases and you have identified your information architecture requirements, derived not from theoretical taxonomy exercises but from the actual work your AI needs to support.
This is also how you build test cases for RAG performance. Each use case becomes a question you can ask the system. If the AI cannot answer it accurately, you know exactly which content or metadata gap to fix.
Webinar
Apr
27
Where Contact Center AI Is Actually Headed
A straight read on where the contact center platform market is going and what it means for your next technology decision.
Register
Webinar
May
12
How Marketing Teams Can Take Control of the Web Experience
See how Veo and Kvalifik put marketing in the driver's seat, shipping faster and growing qualified bookings by 150%.
Register
Webinar
May
14
Two Audiences, One WordPress. Is Yours Ready?
The practical steps to making your WordPress portfolio ready for both human visitors and AI agents.
Register
Webinar
May
19
From Content Sprawl to Competitive Advantage: A Knowledge Operations Roadmap
How automation, AI and a centralized content platform turn knowledge fragmentation into a scalable competitive advantage.
Register
Webinar
May
20
From AI Investment to CX Results: What Enterprise Leaders Need to Know
Move beyond experiments. See how top enterprises scale AI for CX results.
Register
Webinar
On demand
AI for Your DXP: Connect What You Have, Transform How You Work
Most AI strategies stop at the platform—but work happens elsewhere. Bring intelligence into Teams, email, tickets and CRM.
Watch Now
Building the Foundation: A Practical Sequence
Information architecture at enterprise scale sounds daunting. It does not have to be. The approach that works in practice is progressive enhancement, building metadata and structure incrementally rather than attempting comprehensive coverage from the start.
- Start with content types and core metadata. Identify the five to seven metadata fields that matter most for your priority use cases. Content type. Product or domain. Audience. Authority level (draft, approved, retired). Effective date. A competent content contributor can apply these fields in under two minutes per document.
- Use AI to accelerate enrichment. Once a core metadata foundation exists, AI can suggest additional values: topic extraction, audience inference, entity recognition, relationship proposals. Human reviewers validate AI suggestions, creating a flywheel that is an order of magnitude faster than manual tagging alone.
- Let usage patterns refine the model. Track which queries succeed and which fail. Identify co-access patterns that suggest relationships your metadata has not yet captured. Feed user feedback into the enrichment process. The content model becomes a living system that improves with every interaction.
This sequence acknowledges organizational reality. Content owners have limited capacity. Governance processes take time to mature. Starting simple and iterating beats starting comprehensive and stalling.
Related Article: The Pilot Paradox: Why Enterprise AI Complexity Grows Exponentially
The Competitive Implication
Organizations that build strong information architecture will not just have better GenAI. They will have built infrastructure that supports every future AI application: better search, personalization, analytics and eventually agent-based systems. The metadata foundation normalizes meaning across the enterprise, enabling integration between systems that were never designed to work together.
The AI landscape will continue to evolve rapidly. Models will improve and new capabilities will emerge. But the requirement for structured, well-architected content will remain constant. The organizations doing this work now are building a foundation that lagging competitors will struggle to replicate.
There is no AI without IA. The organizations that act on this will lead.
This is Part 2 of a three-part VKTR series on scaling enterprise AI. Part 1, "The Pilot Paradox," examined why the integration complexity of enterprise AI grows exponentially. Part 3, "Governance That Enables Iteration," addresses the feedback loops and operating model that sustain AI quality at scale.
Learn how you can join our contributor community.