Case study · State Government Department
Introducing a governed enterprise AI assistant for organisational knowledge
- Client persona
- State Government Department
- Focus areas
- Enterprise AI solutions AustraliaEnterprise AI assistantSecure generative AIAI implementation strategy
Executive summary
Staff in a complex organisation spent significant time finding policies, procedures and answers spread across many repositories. The team defined the business case, use cases, knowledge boundaries and governance for an enterprise AI assistant, treating expected behaviour, knowledge quality and human oversight as core design work rather than afterthoughts.
Representative experience delivered by the MMG Tech team within complex enterprise and public sector environments. Client details are withheld and outcomes are described without unverified metrics.
The challenge
Why it mattered
Information existed but was hard to reach. Content was duplicated, outdated or held in systems with different access rules, so staff relied on colleagues and informal knowledge.
Leadership saw potential in generative AI but also clear risks: incorrect answers, exposure of restricted information and user expectations that the assistant could do more than it should.
- Fragmented and inconsistent knowledge sources
- Unclear boundaries for sensitive information
- High and varied user expectations of AI
- No agreed ownership once live
- Need for responsible AI governance
The solution and architecture
What was done and why
- 01
Business case and use cases
Use cases were identified with staff and prioritised by value, risk and knowledge readiness. Early scope focused on high volume, low risk questions where answers could be traced to an authoritative source.
- 02
Knowledge sources and quality
Candidate repositories were assessed for accuracy, currency and ownership. Content owners were engaged to retire duplicates and fix gaps, because assistant quality depends on the quality of what it can draw on.
- 03
Expected AI behaviour
The assistant's role, tone, limits and refusal behaviour were defined in plain language. It was designed to cite sources, acknowledge uncertainty and direct people to a human when a question fell outside scope.
- 04
Information boundaries and governance
Access rules from source systems were respected so users only received answers from content they were already permitted to see. Governance covered data handling, responsible AI principles, approval of new sources and incident response.
- 05
Adoption and expectations
Communications and training explained what the assistant does well, what it does not do and how to give feedback. Setting expectations early reduced frustration and built trust.
- 06
Operational ownership and monitoring
A named business owner, content stewards and a support model were established. Answer quality, feedback and escalation patterns were reviewed regularly to guide continuous improvement.
Outcome and impact
What changed
- Improved access to organisational knowledge for staff
- A governed approach to responsible AI use
- Clear information boundaries aligned to existing permissions
- Defined ownership and escalation once the assistant is live
- A repeatable approach for adding new use cases
Capabilities demonstrated