Turning ten million pages of leadership expertise into a learning tool users can trust.

Overview

Decades of frameworks, assessments and facilitation guides. Ten million pages of it. And almost none of it reachable when a learner actually needed an answer. The body of expertise was the firm's greatest asset — a global leadership development organisation serving Fortune 500 clients and individual learners alike. It was also, in practice, locked inside documents.

Conversational Q&A over a proprietary corpus, every answer cited back to source. No hallucinated frameworks, no orphaned claims.

Challenge  | Ten million pages is an asset — and a liability.

The firm's content library had grown over decades. The same leadership concept might appear in a 1990s framework, a 2010s revision, a case study, and a facilitator's guide — four documents, overlapping but not identical, each written for a different purpose. A learner got all four back with no signal about which to trust.

• Overlapping content, no hierarchy. The same concept lived across multiple document types — framework, case study, facilitation guide — each authoritative for a different reason. Standard retrieval treated them as interchangeable text.
• Format authority mattered. A competency model is trustworthy because of its research methodology. A facilitation guide because of who wrote it and how it's used in the room. The retrieval layer had to weight these differently.
• Every answer had to be defensible. Any answer given to a Fortune 500 learning director had to be cited, traceable, and grounded in a passage the firm had written. Not 'probably right' — defensible.

The work  | An agentic retrieval layer built on the firm's own content.

We worked through the entire archive, removed duplicates, organised material by topic and authority, and gave every passage a clear lineage back to its original source. On top of that sits an agentic retrieval layer that plans its own search — deciding which document types to pull, weighing them by the authority their format carries, and running follow-up retrievals when an answer needs more than one source to stand up.

Outcomes  | Pilot-ready, with the metrics to defend it.

Customer-facing app, ingestion infrastructure, and the benchmarks the client uses to measure their own retrieval quality going forward.

About the client

Decades of expertise locked inside documents. We built an agentic retrieval system for a global leadership development firm that answers a learner's question in conversation, plans its own search across decades of research, and shows the exact page behind every answer.