The insurer keeps complex and sensitive cases with staff as automation expands across customer service.
Singlife is expanding artificial intelligence across customer service whilst setting limits on which enquiries can be handled without human intervention.
Its email agent now handles more than a fifth of enquiries and has helped reduce handling time by 30%.
Romil Sharma, Group Head of Technology and Operations at Singlife, said the system assesses customer intent, urgency, and context before deciding whether an enquiry can be answered automatically, reviewed by staff, or escalated.
Straightforward enquiries may be handled by the system or checked by customer service teams, whilst complex or sensitive cases are routed to representatives.
Sharma said policy endorsement requests such as coverage extensions may require premium calculations and customer-specific assessment, making human judgment necessary.
Singlife also tracks more than response speed when evaluating the system.
“Ultimately, success is about striking the right balance between efficiency and quality because a faster response only counts as progress if it gives the customer a better experience,” Sharma said.
The insurer assesses whether enquiries are understood and routed correctly, whether responses are useful, and whether customers need to follow up or seek clarification. Feedback from customer service teams is also used to refine content, tone, and detail.
Before AI can respond directly to sensitive insurance enquiries, Sharma said insurers need confidence in accuracy and clear escalation rules for cases requiring human expertise.
He added that governance must cover oversight, controls, customer data protection, and safeguards against harmful or misleading responses.
Accountability must also remain with the insurer.
“You can outsource services and capabilities, but you cannot outsource responsibility and accountability,” Sharma said.
If AI provides an incorrect answer, the insurer must own the outcome, correct the issue, and prevent the same mistake from recurring.