20 SEP 2026 — Singlife says AI agents have cut average handling time by 30 per cent and halved new-hire training. The launch announcement a year ago published no handling time, no training time, and no headcount.
The percentages are real numbers from the company. What they are percentages of has not been made public.
What is actually deployed
Two things, and they do different jobs. The General Insurance Email Agent reads incoming customer email, judges the nature and urgency of each request, and routes it. Buddy is internal. It supports customer-service executives with real-time answers drawn from Singlife's own product manuals, training guides and FAQs.
Both run on Salesforce's Agentforce platform with Data Cloud underneath. The launch was announced on 14 October 2025, with Singlife described as the first insurer in Singapore to adopt the platform.
Romil Sharma, Singlife's group head of technology and operations, said at the time that AI was becoming a key part of how the business is run. Arun Kumar Parameswaran, Salesforce's executive vice-president and managing director for South and Southeast Asia, framed it as unlocking service efficiency while enhancing human connection.
Deflection and augmentation are different
The reported results mix two categories that are usually worth separating.
The email agent handles more than 20 per cent of enquiries arriving by email. That is deflection: work no longer reaches a person. Buddy's 30 per cent reduction in average handling time is augmentation. The same person does the same job faster because an assistant retrieves the answer.
The two have different implications. Deflection changes how many staff are needed. Augmentation changes what each of them can get through, and it usually depends on the assistant being right often enough that checking it costs less than looking it up.
The one figure that is checkable
Among the claims, the training number is the most concrete. New-hire onboarding is said to have gone from four weeks to two.
That is stated as an absolute, not as a percentage of an unpublished quantity. It is the only result here a reader can hold onto. It is also the one most exposed to a simple alternative explanation. Training programmes shorten for many reasons, and a year is long enough for a curriculum to have been redesigned alongside the tool.
TechNode Global, which reported the results, states plainly that the performance figures were supplied by Salesforce and Singlife and have not been independently audited.
What a baseline would have cost
Publishing an average handling time in October 2025 would have taken a sentence. The company evidently had the figure, because it can now report a percentage change against it.
We reported this week on a much larger AI adoption claim with no external verification. There the pattern was an unaudited number. This one is narrower and more fixable. A single published baseline at launch would have turned this year's announcement from a claim into a measurement.
The figures may well be accurate. They are simply not checkable, and that was a choice made at launch rather than a limitation of the technology.
What happens next
Singlife intends to take Buddy beyond internal assistance and turn it into a customer-facing agent, letting policyholders retrieve policy information and complete routine self-service requests.
That is a materially different risk profile. If an internal assistant gives a service executive a wrong answer, a trained human sits between it and the customer. A customer-facing agent has no such buffer, and in insurance the answers concern what a policy does and does not cover.
If that ships, the number to watch is not the percentage improvement but the agent's refusal rate. How often it declines to answer, and what happens to the customer when it does.
