Case Study
Pioneering
AI-led knowledge management
Overview
Imagine a global company with thousands of support agents trying to help millions and millions of users every minute of every day. To do their jobs, agents rely on an internal "source-of-truth" encyclopedia about the company's products, policies, and workflows (known as a Knowledge Base). However, the Knowledge Base is disorganized, outdated, and built in a way that is hard for AI to ingest.
This project was about completely rebuilding the Knowledge Base to make it both "AI-ready" and fully accurate. Over 1,300 resources were optimized, resulting in a significant jump in retrieval accuracy and ensuring a seamless experience for both human agents and automated systems.
The
Challenge
The
Approach
The existing internal support database was fragmented and relied on "AI-unfriendly" formatting, which meant chatbot responses were often inaccurate and low quality. Information was often hidden inside complex tables, nested menus, and alert boxes that both human agents and automated systems struggled to navigate. Overall, this disorganization and inaccuracy led to inefficiencies and a lack of trust in the system, as information retrieval systems often missed crucial context or surfaced incorrect answers.
As the strategic lead, I directed the comprehensive redesign of over 1,300 internal resources to create a centralized "single source of truth". To solve these systemic issues, I developed a Universal Content Framework designed for both human predictability and machine ingestion. My strategy focused on:
-
Defining Intent-Driven Content Models: I identified three core intents—product education, branching agent workflows, and issue diagnosis—to ensure every article served a clear, singular goal.
-
Flattening Information Architecture: I moved away from "AI-unfriendly" formats by replacing nested menus with clear, sequential headers (H2 and H3).
-
Optimizing for Retrieval-Augmented Generation (RAG): By restructuring content into semantically related "chunks," I ensured that AI systems could accurately parse and retrieve information without being overwhelmed by data noise.
The
Implementation
I orchestrated the hands-on execution of this overhaul across a high-impact team. Key tactical steps included:
-
Content Redesign: Directed the rewrite of 1,300+ articles, successfully launching modernized resources for frontline support agents in distinct phases.
-
Establishing Operational Infrastructure: I authored a standardized onboarding guide for content leads and defined vendor roles to maintain a "quality-first" workflow and long-term content ownership.
-
Technical Security Tagging: I implemented a tagging strategy using specific HTML tags to redact sensitive internal process steps, preventing confidential data from leaking into external-facing AI responses.
-
Data-Driven Refinement: I applied research-based chunk sizes (approximately 100–300 words) to articles to strike the optimal balance between providing enough context for the AI and maintaining high retrieval precision.
The
Results
This project prepared the global support organization for large-scale AI integration while maintaining 100% business continuity during organizational shifts. Technical evaluations of the redesigned content showed:
-
Enhanced Retrieval: The AI's ability to find and surface the correct article improved from 39% to 57%.
-
Improved Content Quality: High-helpfulness ratings rose from 34% to 52%, while poor-quality answers and "hallucinations" were significantly reduced.
-
Operational Scale: Established a repeatable operational model that allows the team to iterate and build upon it as AI advances.
Key Takeaways
Systemic Infrastructure for Operational Scale
This initiative demonstrated my ability to lead the large-scale transformation of fragmented legacy data into a centralized, machine-readable asset library. By overseeing the redesign of over 1,300 technical articles, I transitioned the support organization from a disorganized support model to a "single source of truth" architecture. This structural overhaul ensured 100% business continuity during a period of significant leadership transitions and organizational shifts.
Quantifiable ROI on AI Performance
The business value of "AI-ready" content was proven through rigorous technical testing. By flattening the information architecture and optimizing content for AI retrieval systems, the automated system's ability to fetch and surface relevant support articles improved from 39% to 57%. This directly enhanced support quality, with high-helpfulness ratings from agents increasing from 34% to 52%.
Sustainable Governance & Workflow Oversight
To ensure long-term scalability, I developed a comprehensive standardized onboarding guide and established a "quality-first" workflow for support agents. This created a repeatable operational framework that moved beyond manual content maintenance, allowing the business to maintain content integrity and accuracy at scale.