Building an AI Knowledge Platform After 7 Years of Failure

The biggest barrier to enterprise knowledge management isn’t technology — it’s trust. Seven years of failed attempts had taught everyone in the organization to ignore the next proposal. The platform that finally worked started by earning credibility, not by building software.

Every new leader who came in took their shot at it. Every one of them left behind a proposal, a pilot program, and a Slack channel that went quiet after three months. Seven years of this.

The problem wasn’t complicated to describe: knowledge was scattered across 30+ systems, 40+ million content pieces, 23 stores, 40 languages, 10 repositories. Teams had competing ownership, different tools, different workflows, different standards. Content authors were toggling between disconnected systems every single day — they called it the “toggle tax.” Nobody could search across everything. Nobody knew if content in one system contradicted content in another. And every AI initiative downstream was building on a foundation nobody trusted.

Why I Started With People, Not Technology

I came in with no predefined scope. Just: “figure out what’s possible.” My first move was the one nobody had tried — I spent weeks doing day-in-the-life sessions with the content authors, editors, and reviewers who actually lived in these systems. Not interviewing them in a conference room. Sitting with them while they worked.

That’s where I learned what was actually broken. It wasn’t a technology problem. It was a trust problem. The real problem is never in the brief. Seven years of failed promises meant nobody believed the next proposal would be different. The listening wasn’t just research — it built the credibility that let me eventually align 2 VPs and 8+ senior stakeholders around a single vision.

What the AI Knowledge Platform Actually Does

I call the architecture a knowledge supply chain — not just a search tool, but a system that actually connects everything. A catalog that indexes content across all 30+ systems, discoverable by people AND machines through keyword, metadata, and semantic matching. Content health agents that autonomously find contradictions, stale content, and broken relationships. Workflow automation for the full content lifecycle — create, review, publish, modify. AI woven throughout: generation, evaluation, aided editing, discovery.

I wrote the strategy from scratch, secured funding, assembled a tiger team, and had a working prototype in 4 weeks. I didn’t just design it on paper — I used AI-assisted development to build high-fidelity mocks with real components from our design systems and packaged the code to serve as a starting point for our engineers. When the team saw a working prototype instead of another slide deck, the energy shifted completely. Then we ran monthly beta releases for 16 weeks with real internal customers testing and giving feedback before shipping to production.

The Results: From 30 Disconnected Systems to One Knowledge Layer

Content defects dropped 63%. Time to publish fell 79%. The platform eliminated 30 million customer contacts that were happening because people couldn’t find the right information — $240 million in cost avoidance.

The platform became the foundation for AI initiatives across the org — the reliable knowledge layer that everything else could build on.

But the number that matters most isn’t on a dashboard. After the launch, a customer who’d been there for all seven years of failed attempts said:

“After 7 years, countless proposals, you used your expertise in people and systems to deliver this platform which solves problems we never thought possible.”

That’s the one I keep.