Surveys show Singapore consumers want AI service tools to remember their history, explain their limits and hand them to a human without forcing them to repeat themselves.
Singapore businesses rushing to deploy AI in customer service are being told by the market to slow down and design for trust, not just automation. A 2026 Twilio survey cited in the lead article found that 78% of Singapore consumers had abandoned an AI agent after it failed to identify them, while only 17% described their AI assistants as very effective at remembering who they were and what they had bought. The same survey suggested that Singaporeans believe AI already handles 49% of service interactions and expect that share to rise to 59% next year.
The message for customer experience leaders is less about whether AI belongs in service and more about how it is introduced. Gartner said in an early 2026 survey that 87% of customers believe companies using generative AI in customer service must still provide access to a human agent. That finding aligns with Gartner’s advice that firms should not force generative AI to be the first stop for every enquiry, particularly when the issue is complicated or emotionally charged.
The strongest customer ratings tend to come from systems that recognise returning users quickly and visibly. In practice, that means linking the assistant to existing customer records, order history and support tickets so the conversation starts with context rather than from zero. The lead article’s review of Twilio’s data also suggests that when AI cannot identify a customer, many people either switch to a human, abandon the exchange or leave the brand altogether. The design challenge is therefore largely one of integration, not simply model quality.
A second requirement is the ability to reach a person without friction. Singapore respondents told Twilio they were more comfortable with AI when a human handover was available at any point, and the article argues that the option should be obvious from the first exchange. That preference also fits broader customer behaviour: if people know they are not trapped inside an automated loop, they are often more willing to let the system handle routine work first.
The handover itself matters just as much as the escape route. Five9’s 2026 research, based on surveys of 3,000 consumers and 600 business leaders, found a large gap between what organisations believe and what customers experience: 96% of customer experience leaders said their companies preserved context during AI-to-human transfers, while 83% of consumers said they still had to repeat information at least sometimes. The practical answer is to pass a concise structured summary to the agent, not a raw transcript, and to route the case to someone who can actually resolve it.
Transparency and control are the other two pillars. The lead article notes that 67% of Singapore respondents wanted brands to be upfront about using AI, which is consistent with Singapore’s governance direction. IMDA launched its Model AI Governance Framework for Agentic AI in January 2026 and updated it in May with additional case studies and best practice guidance, including recommendations that organisations make clear what an agent can do, what data it can access and where human accountability remains. The same logic applies to actions that carry risk: the article says 54% of Singapore consumers would feel more comfortable if AI required human approval before taking steps such as refunds, cancellations or contract changes.
For smaller firms, the recommended starting point is practical rather than ambitious. The lead article argues that most service teams can identify a relatively small number of high-volume enquiries, connect the AI to the systems that hold the relevant data, define when escalation should happen and launch on a single channel before expanding. Gartner has also warned that the cost of generative AI resolutions may exceed US$3 by 2030, which means the business case depends on resolving routine queries efficiently while handing complex work to people with full context. In that model, good AI service is not a replacement for human support, but a controlled first layer that customers are willing to use.
Disclaimer: This content is intended for informational purposes only. Readers are advised to exercise their own judgement, conduct due diligence, or consult a qualified expert before acting on any information provided.
