Singapore’s Central Provident Fund (CPF) Board is integrating “everyday AI” into how its employees work, seeking to modernise service delivery without compromising the trust and reliability members expect.
“Our AI strategy is about empowering every CPF officer to be AI-ready… to enhance productivity, deliver better services and improve the employee experience,” says Soh Tse Min, CPF Board’s group director for customer relations, in an interview with The Edge Singapore at Salesforce’s Dreamforce conference.
CPF Board’s approach moves beyond traditional centralised AI teams. It has established a three-tier strategy: basic AI literacy for all staff, advanced skills for a core group, and specialist expertise at the top. The goal is to democratise AI use while maintaining depth of expertise.
Digital foundation
The AI-ready workforce strategy required foundational infrastructure changes. CPF Board previously ran separate on-premise systems for case and knowledge management, forcing customer service agents to toggle between platforms and slowing response times. Those systems also faced end-of-life risks typical of on-premise deployments.
In July 2024, the statutory board rolled out the Next-generation Integrated Customer Ecosystem (NICE) 2.0, a cloud-based platform built on Salesforce Service and Experience Cloud that unifies customer relationship management, knowledge management, and the omni-channel portal. Soh shares that Salesforce was selected for its scalability, modularity, out-of-the-box features, integration with whole-of-government systems, and low-code/no-code approach.
See also: MAS consults on guidelines for AI risk management
NICE 2.0 has delivered measurable improvements. The statutory board’s FAQ system now handles 1.2 million monthly queries, and has improved First Contact Resolution from 86.6% to 88.6% within the first year. The platform processes approximately 34,400 service cases per month, enabling more personalised, cross-channel responses.
Eight months after NICE 2.0’s launch, CPF Board added AWS Connect (voice engagement channel) to the platform for its contact centre services. Customer service agents now focus on customer conversations while call summaries are generated automatically. The system surfaces context, showing what information a customer accessed online before calling, allowing more targeted responses. The new contact centre also allows identity verification, which enables authenticated customers to receive detailed information securely.
Additionally, CPF Board utilises Salesforce’s Einstein Bot to handle inquiries on WhatsApp. The bot uses structured conversation flows to guide customers through predefined options and interpret free-text queries to surface relevant FAQs before routing complex cases to human agents, who can view the full chat history. Operating around the clock, the bots manage routine queries, freeing staff to focus on cases that require judgment and empathy.
See also: DBS boosts corporate client support with generative AI-powered virtual assistant
Trust and human oversight
As the trustee of members’ retirement savings, CPF Board maintains what it calls “comprehensive guardrails” through a responsible AI framework and risk-based approach that balances innovation with protection.
“Trust is very valuable to us, and we need to maintain it. That’s why our generative AI use cases are internal-facing and require human-in-the-loop,” says Soh.
She adds that the statutory board has identified six high-value areas for generative AI, including writing and drafting assistance as well as knowledge management. The focus remains on boosting productivity and efficiency while preserving accuracy and reliability.
Two in-house tools illustrate this – draftCentral helps officers compose replies to complex appeals, while navCentral retrieves relevant data from CPF Board’s repositories. Both require human oversight, a deliberate safeguard reflecting the organisation’s trustee role.
Cultural shift
During NICE 2.0’s rollout, CPF Board insisted on minimal customisation, resisting the urge to recreate bespoke features from legacy systems. In line with that, it consolidated thousands of FAQs into a single knowledge base and standardised response times across all service types.
To stay ahead of the latest tech trends, click here for DigitalEdge Section
Close coordination between operations and technology made those changes possible. Soh and her IT counterpart co-chaired monthly steering committees with equal responsibility — while she managed business stakeholders, he led technical teams. “We knew from day one we were in this together. For ops-tech integration, the partnership between business and tech teams is critical. We did that very well,” she claims.
Data-driven decision-making helped align teams, too. She adds: “We made use of data to give us a sense of whether we can cut or streamline a process. When you lay out the facts on the table, people are more accepting of the outcome.”
Besides that, senior management support, dedicated change management resources, and open feedback loops underpinned the transition. “The real work starts after the implementation process. We’re still taking in employee feedback and assessing them. It’s a work in progress, up to today.”
Eyeing AI agents next
According to Salesforce, AI agents are the natural evolution of generative AI. While generative AI generates content and insights, AI agents act on those outputs — executing tasks, retrieving information, and collaborating across systems with minimal human input.
Not resting on its laurels, the CPF Board is exploring the use of AI agents. Early pilots focus on internal productivity, but Soh believes such tools could make digital services more inclusive. “I do see the potential in terms of them… making digital services more accessible and levelling the playing field between tech-savvy members and those who prefer traditional channels.”
Still, the statutory board recognises that greater use of technology must not come at the expense of the trust and human touch that members expect. Its next challenge is scaling generative AI and deploying AI agents while preserving that balance. “[While we believe new technology can be] an enabler and we’d like to try them, we also recognise the need to preserve the reliability and accuracy that members trust us for,” says Soh. “At the end of the day, it’s about striking a delicate balance.”
