Want to know more?
Are you interested in this project? Or do you have one just like it? Get in touch. We’d love to tell you more about it.
Data insights in minutes, not days: Scaling secure AI in superannuation
Superannuation
How a leading industry fund built a reusable Model Context Protocol (MCP) foundation to connect AI to live enterprise data safely
The Australian superannuation sector operates under intense regulatory requirements, where fund performance, data security and member outcomes are constantly scrutinised. Maintaining compliance while innovating and adopting new technologies can be a significant challenge.
One of Australia’s largest funds wanted to accelerate its AI adoption beyond a basic corporate chatbot. Key teams were eager to use AI for real-time insights, but doing so safely required a secure, well-governed way to connect a Large Language Model (LLM) to internal data stores.
The super fund engaged Equal Experts to implement its Model Context Protocol (MCP) foundations. After running targeted spikes to determine the best transport, technology and architecture, we built a production-ready infrastructure blueprint. We validated the approach with implementations for two teams, iterating our approach to improve LLM accuracy through the addition of organisation-specific context, MCP resources and prompt templates.
By the end of the three-month engagement, the superannuation fund had an approved, scalable MCP pattern, two production-ready implementations and the confidence to scale AI responsibly. It enables staff to generate real, data-driven insights in minutes, rather than days, transforming how the fund uses data while ensuring security and governance remain intact.
Outcomes
Insights in minutes
Reduced data turnaround times from 5 days
Production-ready foundations
Two core services tested and verified by business end-users
Risk mitgation
A clear, secure pathway with scalable governance and guardrails
About the client
Our client is a leading industry super fund in Australia supporting thousands of Australians with their superannuation. As one of the country’s largest funds, it manages tens of billions in retirement savings, focusing on delivering strong long-term returns and modern digital experiences for its large national membership base.
Industry
Superannuation
Organisation size
Large superannuation fund
Location
Australia
Length of project
3 months
Challenge
Moving AI beyond a basic corporate chatbot
Our future-focused client recognised the transformation potential of AI and was an early adopter of the technology. After successfully launching its own internal chatbot, which used corporate documents to answer common employee questions around HR or policies, several teams within the organisation saw far greater potential for AI adoption.
Teams across the organisation, including those within member services and investments, rely on up-to-date data to make crucial decisions but often cannot query the raw data themselves. Instead, they relied on a data team to access operational insights. Typically, a team would make a request and wait up to a week for the creation of a bespoke solution to access the information required, with multiple iterations often necessary to achieve a useful result.
Seeing an opportunity to eliminate this bottleneck and free up the data team for more strategic work, the super fund looked at how to embed AI directly into this day-to-day workflow. These new use-cases relied on access to real-time data from across the organisation’s data stores, with strict governance and data security implications to be considered.
Model Context Protocol (MCP), a new standard for connecting AI assistants to live data systems, offered a potential solution. However, as an emerging technology and entirely new to the super fund, the organisation lacked the internal delivery capability to realise this ambition safely.
Rather than building small, limited capabilities for a few isolated AI trailblazers, our client wanted a foundation that would enable future scale. By creating the foundations in the right way from the outset, subsequent teams would be able to follow the same approved architectural pattern and build their own MCP services easily, securely and quickly.
Equal Experts was engaged to design the MCP server pattern and demonstrate the value of the approach through two early implementations for the investment and member teams.
Solution
Designing a secure, scalable MCP architecture
With MCP being a relatively new and rapidly evolving standard, offering significant flexibility in implementation, we began the engagement with a series of spikes. These validated the best technical approaches for the superannuation fund’s specific organisational context and regulatory compliance requirements.
We started by assessing several transport options, including Server-Sent Events, STDIO and Streamable HTTP, against the fund’s strict security requirements, eliminating options that conflicted with network segregation policies or that were already being phased out. Streamable HTTP was ultimately selected, as it allowed for secure, centralised routing, though it still required custom engineering to adapt standard open-source models.
Further spikes identified AWS AgentCore Gateway, a fully managed MCP service, as the right technology for the fund. This option used a programming model already widely used and understood across the organisation, helping to reduce adoption barriers and the amount of internal upskilling required.
Maintaining vital security and regulatory compliance
Operating within the Australian superannuation landscape meant security could not be compromised for accelerated innovation. We addressed three critical technical hurdles in the design of the MCP architecture:
- Identity and access management: We implemented 3-legged OAuth within the fund’s existing role-based access controls, ensuring correct user-level data restrictions throughout the data querying process.
- Data sovereignty: We engineered the architecture to ensure that no corporate credentials or Personally Identifiable Information (PII) were ever transmitted to external servers.
- Centralised governance: Individual teams could create decentralised functions under a centralised MCP structure to maintain governance, data security and data sovereignty across the organisation.
Improving LLM accuracy and building user trust
We conducted user testing to verify the accuracy of the answers generated by the LLM. Initial testing revealed that while the technical pipeline functioned, the results were not always accurate. The LLM occasionally guessed which tables to use, incorrectly joined unrelated data together and confidently presented inaccurate results. While experienced users could often identify an issue, they were concerned errors could be missed when under pressure or tight deadlines. Blindly trusting these outputs could lead to severe operational risks.
With a solid baseline to build on, we introduced two advanced MCP features designed to improve accuracy:
- Organisation-specific context and resources: We treated the LLM like an industry-knowledgeable new employee. By feeding it the fund’s specific internal onboarding materials, glossaries, and documentation, we drastically improved its contextual accuracy.
- Prompt templates: We embedded specific instructions requiring the LLM to explicitly display its reasoning and the exact database query performed within its response. This allowed the team to quickly review and verify the logic behind an output before acting on it.
Results
Valuable insights in minutes, not days
In just three months, the superannuation fund established a foundational MCP pattern ready for scale across the entire organisation, enabling future teams to build new services easily, securely and quickly. We also demonstrated the value of the approach with two production-ready servers, tested and verified by the end-users.
The tangible impacts for the fund include:
- Accelerated insight generation: Previously, generating complex, data-driven insights required lodging tickets with the data team, taking up to five days. With the new MCP infrastructure, business users can safely query live systems and generate accurate insights themselves in minutes.
- Secure by design: The foundational MCP pattern is a secure, well-governed way to connect an LLM to internal data, meeting relevant security and regulatory compliance requirements.
- Reduced risk of shadow IT: By providing a fast, secure, and pre-approved architectural blueprint, individual business units don’t need to bypass IT governance to innovate.
- Optimised resource allocation: The fund’s data engineering teams are no longer a bottleneck for insights information, freeing them to focus on high-value data modelling and long-term strategic projects.
- Internal capability uplift: Through close collaboration with Equal Experts’ consultants, the fund’s internal teams gained the practical skills and confidence required to maintain, govern, and scale their AI ecosystem completely independently.
Conclusion
By focusing on robust architectural foundations rather than isolated one-off projects, the superannuation fund successfully bridged the gap between rapid AI innovation and ongoing business value. The implementation of approved MCP server foundations provides a reusable framework for the entire enterprise, not just a few select trailblazing teams.
Crucially, the collaborative delivery model ensured that the fund’s internal teams developed new skills and knowledge throughout the process. The organisation transitioned from initial hesitation around live AI data access to having the skills, governance structures, and confidence required to scale their AI roadmap independently, ensuring long-term agility and compliance.