Case Study: Turning a Scattered AI Knowledge Base into a Living, Cited Playbook for an Industry Body

The situation

IAB Hong Kong’s AI and Technology Committee had real material on AI governance and responsible digital marketing, including contributed research from partners like Google and Meta, but it lived nowhere durable. It was spread across Google Drive folders, forwarded PDFs, one-off decks from past events, and email threads. Every time a committee member or member company needed a governance answer or a stat for a deck, they either remembered where they’d seen it once, or they didn’t, and started from zero.

The core problem was fragmentation, the same failure mode that hits most industry bodies: content gets produced for a single event or briefing, consumed once by whoever was in the room, and then evaporates. Nobody was losing hours a day to this the way an operations team would, but the committee was losing something slower and more expensive: compounding intelligence. Every member was relearning the same lessons independently instead of the industry’s knowledge getting sharper with each contribution.

What the committee actually wanted

Two requirements emerged clearly in scoping, and both shaped the design:

1. Answers had to be trustworthy enough to put in front of member companies under the IAB HK name. Any tool that could invent or misstate AI governance guidance was a liability, not an asset, for a body whose credibility depends on getting compliance content right.
2. The resource had to get better over time, not stay static. A one-time PDF or slide deck was the thing the committee was already drowning in. What they needed was a living surface that absorbed new contributions automatically rather than requiring a manual rebuild every time new research came in.

This meant the engagement couldn’t just ship a searchable document library. It had to behave like a knowledge system with governance built in, not bolted on.

What Humantyze built

Humantyze designed the HK AI Marketing Playbook as a RAG-based knowledge graph rather than a folder of documents, because knowledge in practice is networked, not hierarchical. For example, a regulation connects to a stat, which connects to a case study, which connects to a tool. Folder structures force every idea into one home. A graph lets it surface everywhere it’s relevant.

The resulting system:

• Ingestion: Contributors drop in a PDF, deck, or article, and the AI extracts the concepts, reads the charts, generates the pages, and links them into the graph automatically, replacing manual curation with automated absorption.
• Retrieval: Multilingual embeddings plus hybrid vector-and-keyword search, reranked for relevance, so English and Chinese content are both searchable and the most useful result surfaces first rather than the most recent upload.
• Page structure: Every page follows a fixed format — a TL;DR in three bullets, why it matters for Hong Kong marketers specifically, bold stat callouts contributors can lift straight into a deck, and a “so what” close — so the resource respects the reader’s time instead of assuming they’ll read a wall of text.
• Governed answers: Responses are drawn only from contributed material, with the system explicitly instructed never to invent claims. Every answer cites its source, and every page lists its contributions, so committee admins can curate, review, and back up the knowledge base rather than trusting a black box.
• Guided exploration: A quiz layer surfaces verified questions grounded in the knowledge base, so getting a question wrong and reading the cited answer functions as a briefing, not a gimmick.
• Agent access: An MCP server exposes the Playbook so AI agents and models, not just human browsers, can query the committee’s collective intelligence directly, positioning the resource as infrastructure rather than a webpage.

Design Considerations

The tempting version of this build is a heavier one: a dedicated enterprise search platform, a locked-down internal wiki, or a static published report refreshed annually. Humantyze’s call was the opposite. A heavier stack would have added a maintenance burden and an adoption barrier for a volunteer committee with no dedicated engineering headcount, without adding proportional value. The principle was the same one that guides every Humantyze build: automate the plumbing, and keep the surface as low-friction as possible for the people who have to keep it alive.

Building it as a self-updating graph with automated ingestion meant the committee didn’t need a content operations person to keep the resource current. Contributors hand over what they already have — research, decks, case studies — and the system does the structuring and moderation. That’s the difference between a resource that’s alive a year from now and one that’s already stale by the time it launches.

Outcome and what it signals

The Playbook launched publicly at IAB Hong Kong’s AI and Technology Committee event at Meta’s Hong Kong office, positioned as a live second brain for the industry rather than another one-off report. The immediate win is that AI governance knowledge, previously scattered across formats nobody could reliably search, now lives in one cited, growing surface that gets sharper with every contribution instead of staler with every passing month.

The governance layer — grounded answers only, mandatory citations, admin review — gives the committee a model that scales as trust in the system grows, the same principle Humantyze applies to AI rollouts inside client organisations: don’t ask people to accept full AI autonomy on day one, build the checkpoint in from the start.

The broader lesson for industry bodies and membership organisations sitting on similar knowledge sprawl: the highest-leverage AI intervention is often not a new AI product for members to learn, but a disciplined system that turns the research you already have into one connected, citable, ever-growing view of what the industry knows. Check out the playbook by visiting https://iabhongkong.com and clicking on the “AI & Tech” menu, or go direct at https://iabhkwebbook.humantyze.ai