Insights
From knowledge to action: AI workflow lessons from Japan for super funds
- Date 21 Oct 2026
- Filed under Insights
Part two in a series exploring the practical application of AI in superannuation
In the previous article in this series, Beyond chatbots: Prioritising the AI opportunities that matter most for super funds, I explored why some of the greatest AI opportunities lie in the operational workflows behind member and employer interactions.
Particular attention was given to how AI is helping super fund staff access specialist knowledge more efficiently, with an example drawn from NRI’s own work with Brighter Super.
But NRI’s work in Japan also points to further opportunities. Namely, putting that knowledge to work in resolving operational questions and coordinating the steps in a member’s request.
In this article, we’ll explore how a better understanding of the processes behind these requests can help funds identify where AI investment can deliver the greatest value.
The work behind the enquiry
Let’s begin by looking at a common scenario.
A member calls to ask why a contribution has not appeared in their account. While the consultant might be able to explain the delay, resolving it can still require an operations team to investigate a mismatch, obtain missing information or follow up with the employer.
So, much of the work that determines the member’s experience takes place after the conversation.
For consultants and operations teams alike, the challenge often lies in bringing together the right information and determining what needs to happen next. I would therefore start with the workflow behind the enquiry and ask where staff spend time investigating, waiting for information or passing work between teams.
Here, recent NRI work in Japan offers two useful perspectives: an asset-management support service that makes accumulated knowledge easier to use, and insurance research exploring how AI could help coordinate the process itself.1,4
A lesson from Japan’s asset-management operations
In November 2024, NRI announced NAV Calculation AI Support for the help desk of T-STAR/TX, its investment-management back-office solution.
The service uses retrieval-augmented generation to prepare draft answers from around 20 years of enquiries, curated past responses and user manuals. In proof-of-concept tests of this response-support system, NRI reported an average time saving of more than 50%.1
Additionally, the service addresses the operational questions that arise during net asset value calculation. By placing support within the help desk, it makes specialist knowledge available at a point where other work may be waiting on an answer. That is a useful example of how an investment in back-office support can help the wider service operate more effectively.
For super funds, the parallel lies in the investigation behind an unallocated contribution, an account update or a payment enquiry. Across these different products and regulatory settings, staff need to find and interpret the right information before the next step can proceed. Such work deserves a place alongside member-facing applications when funds assess AI opportunities.
Bringing specialist knowledge into everyday service
Brighter Super provides a useful point of comparison from Australia.
As explored in the previous article, NRI’s implementation of Sparkie leveraged Salesforce Agentforce to help contact centre consultants retrieve information from insurance documentation, knowledge articles and other sources.2
During testing across hundreds of scenarios, the system achieved close to 100% retrieval accuracy. NRI also reported early signs of consultants resolving more enquiries without escalating to team leaders or subject-matter experts.2
Like Japan’s NAV support service, Sparkie demonstrates how AI can bring specialist knowledge to the point where staff need it. Both examples show the value of improving access to information, but finding the right answer is not always enough to resolve an enquiry. It also requires coordinating work across teams, obtaining missing details and addressing issues that arise during the process.
Extending AI into the workflow
Once the information is found, a case may still depend on several further actions. NRI’s March 2026 research on agentic AI examines how large language models (LLMs) can work alongside tools and other systems to plan tasks, carry out actions and check progress. This extends their role from helping someone find an answer to helping coordinate the work that follows.3
In a prototype for insurance claims intake, described by NRI in March 2026:
- An agentic system retrieved policy information, asked the policyholder for missing details and generated the required process.
- One test introduced a name mismatch following marriage, which required a change to the policy record.
- The system recognised the additional requirement and replanned the workflow to incorporate it, combining AI with other techniques to manage dependencies between the steps.4
For super funds, this suggests a useful area for testing: a claim or account request that depends on a separate member-detail issue being resolved. AI could help connect those tasks while retaining the context of the original request.
Given the prototype’s response-time constraints, post-call investigation and document preparation may offer more practical early applications than live conversation. The value would come from helping the case progress through the necessary checks and actions.4
Choosing where to invest
The examples from Japan offer a useful starting point for super funds assessing where to invest in AI.
For me, the priority would be workflows where delays materially affect members or employers, and where investigation, specialist intervention or hand-offs create recurring effort.
Employer contributions provide a useful example. Beyond identifying exceptions and suggesting remedies, AI could help coordinate the steps needed to resolve them, passing cases between teams with the relevant history attached and helping identify the next action when new information arrives.
That does not mean every step needs to involve an LLM. Existing rules can continue to check formats, amounts and mandatory fields, with the LLM handling the interpretation of language, retrieval of knowledge and preparation of responses. A review of the workflow may also reveal opportunities for better data, clearer procedures or conventional automation.
Choosing among these approaches is part of the investment decision.
A clearly defined part of a workflow can also be a sensible first investment, provided the fund has access to the necessary information and a baseline against which to measure improvement.
The assessment should follow a case through to resolution. Measures such as exception age, repeat contacts, hand-offs and first-time-right completion will help show whether faster individual tasks translate into better service, alongside handling time and member experience.
NRI’s agentic-AI research makes a similar case for measuring the flow of work across teams. Integration, running and human-review costs belong in the business case too, distinguishing additional service capacity from savings that reduce expenditure.3
As AI takes on more of the process, permissions should reflect the work being delegated. Reading a record, drafting a response, updating details and sending a communication each require a considered level of authority. NRI’s research on intent governance provides a useful basis for agreeing what may proceed automatically, where approval is needed and when a person should take over, supported by a record of the actions taken and their basis.5
Applying global experience to Australian super
Funds bring a deep understanding of their own operations. NRI’s contribution is to help translate that knowledge into an achievable AI investment, drawing on Pragma’s service journey mapping, the AI Workflow Lab’s opportunity assessment, Data & AI foundations and delivery experience such as Brighter Super. Together, these capabilities support the work of identifying a bottleneck, selecting an appropriate intervention and testing its operational value.
The Japanese examples add practical evidence and research that can inform those decisions. Applying them well requires local superannuation expertise: understanding which lessons are relevant, how they fit a fund’s systems and service model, and what needs to be adapted. The starting point remains the member and employer journey, with technology chosen to support the improvement the fund is trying to achieve.
What success should look like
The strongest starting point is a workflow that matters to members and employers, has a well-understood source of delay and offers a result the fund can measure. In some cases, better access to knowledge will be enough. In others, the greater value will come from connecting that knowledge to the next action.
That keeps the investment focused on the job the member needs done. A consultant can answer with greater confidence, and the operations team has a better chance of resolving the issue behind the enquiry. The two reinforce one another. For a member waiting on a contribution or a claim, the benefit is a clearer response backed by progress on their request.
Those are the outcomes that should guide a workflow-led AI strategy.
References
- Nomura Research Institute (2024). Launch of “NAV Calculation AI Support” to shorten net asset value calculation time using generative AI. 5 November. Announcement.
Note: The reported time saving relates to response support in the proof of concept, rather than the full NAV calculation process. The service supports operational enquiries; it does not make investment decisions or calculate fund prices.
- NRI Australia & New Zealand (2026). Brighter Super and NRI deliver a first for Australian super with Agentforce-powered AI. 18 September. Media release.
Note: The accuracy figure concerns information retrieval across the reported test scenarios, rather than end-to-end service accuracy. Observations about fewer escalations are early implementation findings reported by NRI.
- Nomura Research Institute (2026). Agentic AI: Current State and Implementation Issues. 407th NRI Media Forum, 24 March, pp. 4, 18–19. Presentation.
- Kawana, T. (2026). Testing Insurance Agentic AI towards End-to-End AI Execution. Financial Information Technology Focus, March, pp. 12–13. NRI.
Note: The work described is prototype validation of claims intake and workflow coordination, rather than a production claims service or autonomous claims adjudication. The article identifies response-time, cost and security issues requiring further work.
- Gondoh, A. (2026). Intent Governance in the Age of AI Agents. Financial Information Technology Focus, August, pp. 6–7. NRI.
