Case study · AI engineering
Ask the knowledge base anything, and it never exposes a Chair.
CHAIRS Global works with Chairs and governance leaders on complex board, leadership and organisational questions, backed by a knowledge base of proprietary research, interviews and case studies. We built AskCHAIRS, the agent that answers governance questions from that knowledge, names its public sources, and draws on confidential interviews without ever identifying an individual Chair.
What it changes: The knowledge stops depending on knowing where it lives or who to ask. A user asks a governance question in plain words and gets an answer grounded in CHAIRS Global’s own material, with public sources named and confidential insight used safely.
- Client
- CHAIRS Global
- Sector
- Governance, board and leadership advisory
- Status
- Live in production, as AskCHAIRS
- What we built
- A governance AI agent on CHAIRS Global’s knowledge base
- Knowledge
- Research, interviews, case studies and public sources
- Environment
- Deployed on AWS, in an account they own
- Grounded retrieval
- Knowledge base design
- Public source attribution
- Confidential interview handling
- Voice calibration
- Persona testing
- Answer or clarify
7
Chair personas stress tested
2
Knowledge tiers, public and confidential
0
Individual Chairs identified
Live
AskCHAIRS, in production
The ask
CHAIRS Global had a substantial body of governance knowledge spread across research, interviews, case studies and public sources. The requirement was never a generic question and answer chatbot. AskCHAIRS had to use CHAIRS Global’s own knowledge, not generic AI knowledge alone, attribute public sources clearly, use confidential interview material without naming individual Chairs, and hold a consistent voice across very different users.
A generic assistant would not know which sources should inform an answer, how to attribute them, how to use sensitive interview material safely, or when to answer directly rather than ask a clarifying question first. That called for a knowledge and behaviour layer built around CHAIRS Global, not a chatbot pointed at a folder.
The hard part: two kinds of knowledge, one had to stay anonymous
Public gets named, people do not
Public sources are attributed by name, so a user can see where an answer came from. But interviews and case studies carry sensitive insight tied to real people. Anything drawn from that material is attributed to the CHAIRS Global research and interview process, never to a named Chair.
Grounded, and dependable across sessions
During testing the same kind of question could behave differently from one session to the next. We tightened the instructions and retrieval so answers stayed dependable, and taught the agent when to answer directly and when to ask a clarifying question first.
What we built
Grounded answers
Questions are answered from CHAIRS Global’s own research, interviews and case studies, not from generic AI knowledge alone.
Public sources, named
When an answer uses a public document, the agent names the source, so the origin of the information is clear to the user.
Confidential insight, no names
Interview and case study knowledge is used and attributed to the research process, without exposing the identity of any individual Chair.
One consistent voice
Tone, structure and brevity were calibrated to CHAIRS Global’s own approach, with tables and checklists where they help and less unnecessary hedging.
Answers, or asks
The agent judges whether it has enough context to answer, or whether it should ask one clarifying question first, so users neither restate themselves nor get a confident guess.
A structured knowledge base
Documents were ingested with their metadata and organised into partitions, including a dedicated public data set, so the agent retrieves from the right material.
How it was hardened
CHAIRS Global did not treat the first working version as finished. They tested it as seven different Chairs, then fed back on tone, hedging, source attribution, clarification timing and formatting, and we iterated on each round. They have since asked for more worked examples to be added to the knowledge base.
Tested as seven Chairs
- The Aspiring Chair
- The New Chair
- The Established Chair
- The Senior Chair
And the harder rooms
- The Chair Under External Pressure
- The Chair in Complex Terrain
- The CEO Succession Focused Chair
The line that matters most
The whole build turns on one distinction: public knowledge is named, and confidential knowledge is used but never attributed to a person. That is what lets CHAIRS Global open its research to an AI experience without opening the people behind it.
The job transfers
Anyone with knowledge worth protecting
Professional and membership bodies
Research and policy institutions
Advisory and consulting firms
Anyone mixing public and confidential
Firms with interview or client material
Teams holding sensitive case studies
Any knowledge base with a confidentiality line
More ai engineering work
AI engineering · Life insurance · Australia
They stopped choosing which calls to hear
Life Insurance Direct sells life cover on the phone. We built the system that reads every one of those calls, scores both sides, and hands the team a review queue in risk order instead of a sample.
- 9,218 in four months, both sides
- Calls read
- Live, in their own cloud
- Status
Next
Tell us which part of this looks like your problem.
We will walk you through how it was built, who built it, and what it took to keep it running afterwards.