challenge
Healthcare teams create a lot of useful material: reports, decks, research notes, interview summaries, screenshots, strategy documents and project decisions. Much of it stays tied to the project where it was created, or to the people who know how to interpret it. That becomes a problem when the material is used by AI. A healthcare report may include a strong insight, but it may also include participant quotes, client-specific detail, outdated context, clinical assumptions, policy uncertainty, restricted claims or unclear usage rights. An AI system can find and summarise the material quickly, but speed is unsafe when the system does not know what the material is allowed to become.
Client
AKQA
Format
Platform
ACTION
I designed the healthcare knowledge model for the AKQA Knowledge OS to control that movement from source material to reusable knowledge. The model gives AI a useful role: extract, summarise, classify, compare and propose. It gives the system a governance role: register sources, track ownership, record permissions, apply review states and enforce usage rules. It keeps approval, interpretation and accountability with human reviewers. In this model, documents remain sources. AI creates candidate knowledge objects. Governance decides whether those objects can become reusable memory. Only approved, current, permissioned and caveated knowledge can be served back into future work. The goal was to make healthcare knowledge reusable without losing control of source, sensitivity, permission, confidence, limits or review needs.
RESULTS
The result is a system with a simple accountability model: - AI proposes. - The system tracks. - Humans approve. - Only governed memory is served. For AKQA, this creates a stronger foundation for reusing project learning across teams and markets. The model does not treat every document as knowledge. It makes knowledge reusable only when the rules, permissions, caveats and limits are explicit. It separated source material, evidence, candidate knowledge, approved knowledge and restricted knowledge. This matters because a healthcare insight cannot be reused only because it looks relevant. It also needs a clear source, permission, confidence level, usage boundary and review state. The model shows where AI can help and where it must stop.
LEADERSHIP LENS
This work needed design beyond the interface. The main design problem was the logic of the system: what it can store, what it can serve, what it must restrict, what it can claim and when it needs human review. AI knowledge work often focuses on retrieval quality, summary quality and the assistant experience. Those are useful areas, but healthcare needs stronger controls before retrieval happens. A system that retrieves the wrong material with confidence can create risk faster than a system that retrieves slowly. I shaped the model around three principles. Clarity Documents, candidate objects, governed memory and approved use need separate states. Teams should not confuse storage, relevance or similarity with approved knowledge. Control Permission, sensitivity, confidence, review status, lifecycle state and usage limits need to be defined before knowledge can be served. Care The model needs to protect human judgement, participant sensitivity, client confidentiality, clinical boundaries and the limits of AI inference. The value of the work is not only faster access to knowledge. It helps the organisation know what can be trusted, what needs review, what must stay restricted and what should no longer be used.
