CPF Board uses AI to improve call centre services, prioritise help for less tech-savvy users

SINGAPORE – The CPF Board has been using artificial intelligence (AI) since March to automatically transcribe and summarise calls made to its helpline, allowing human call agents more time to take calls and assist those who need more help.

With the aid of local tech firm NCS, it is piloting a feature that can identify callers who are less tech-savvy and prioritise them, enabling them to reach a call agent faster.

Previously, call agents had to transcribe and summarise every call they handled after the conversation ended.

“Now they become editors instead – they still need to go through the transcript to ensure that it is correct,” said Mr Ng Hock Keong, the CPF Board’s deputy chief executive of infocomm technology and digital services.

Speaking to The Straits Times on the sidelines of NCS Impact 2025, held at the Sands Expo and Convention Centre on July 10, he said the impetus for the project was to leverage AI to better serve Singaporeans.

In 2024, the Central Provident Fund call centre managed 622,000 calls.

“The cognitive load on the call agents will also be different because they don’t have to worry about how to summarise a call later,” said Mr Ng, adding that this means agents will be able to concentrate better on explaining CPF’s policies or handling a case.

The AI system was built using a large language model from cloud computing company Amazon Web Services. Its transcription function is currently available in English, Chinese and Malay, with Tamil in the pipeline.

In June, new functions were introduced to the system.

One of the new features automatically evaluates a call’s service quality, based on indicators such as the caller’s language, tone and volume. A rating is given for every call, and supervisors can also review the ratings to ensure fairness.

This feature helps cut the time needed to audit the service quality of calls, which can be a labour-intensive task, Mr Ng said.

Another new feature identifies and prioritises callers who could be less tech-savvy, allowing them to get help faster.

Using data such as a person’s profile and their past interactions with the statutory board, for example whether they have accessed digital services recently, the AI system is able to flag if self-help options are enough, or if they require more direct assistance.

“We want to bridge the digital divide for those who are less tech-savvy, like the elderly.

“It’s very important for old folks to be able to continue to be functional in the digital world,” Mr Ng said.

“With government services going digital now, we must be able to give them a service channel where they feel like they’re being served.”