A customer in Jakarta spends 20 minutes explaining a billing dispute to a chatbot, gets bounced to a human agent, and has to start the story over from scratch. Multiply that across the millions of AI-mediated conversations happening daily across the region, and you get a sense of the trust deficit quietly building beneath Asia Pacific’s AI customer service boom.
New research from customer engagement infrastructure firm Twilio’s 2026 Customer Insights Series puts hard numbers to a problem consumers have long felt anecdotally: AI agents across APAC suffer from what the company calls “AI amnesia”, a tendency to forget who a customer is or what they said moments earlier.
Also Read: Twilio on why AI companies must rethink customer engagement to succeed in Asia Pacific
Seven in 10 APAC consumers say they have abandoned an AI-powered customer service interaction mid-conversation because the system failed to recognise them or lacked context from a prior exchange. Five per cent didn’t just abandon the chat; they walked away from the brand entirely.
For a region where AI adoption in customer service is accelerating on multiple fronts at once, from Singapore’s banking sector to Indonesia’s e-commerce giants to the Philippines’ business process outsourcing industry, that is an expensive gap between ambition and execution.
A confidence gap between brands and their customers
The disconnect starts with perception. Over four in five APAC brands (84 per cent) believe their AI agents are doing a good job recognising returning customers and recalling history. Yet 70 per cent of consumers say they routinely have to start from scratch every time they open a new AI conversation, a gulf between self-assessment and lived experience that should worry any product team that has taken its own dashboards at face value.
The problem doesn’t resolve itself when a human steps in, either. Nearly two-thirds of APAC consumers (65 per cent) report having to repeat themselves or fill in gaps after being handed off from a bot to a human agent, which suggests the underlying issue is systemic rather than a chatbot-specific flaw.
Robert Woolfrey, Twilio’s Asia Pacific and Japan vice-president, put it simply: AI is only as useful as the context fed into it. Brands that connect their customer data properly, he argued, can move past basic chatbots into assistants that actually remember individuals.
The root cause, per Twilio, isn’t the models themselves but fragmented data, a customer’s history scattered across a CRM here, a support ticketing tool there, a separate loyalty app elsewhere. No agent, human or artificial, can assemble a coherent picture from pieces that were never designed to talk to each other.
For Southeast Asian businesses running newer AI agents alongside legacy systems that predate the current AI wave by a decade or more, that fragmentation problem is arguably sharper than in more digitally mature markets. It is also precisely the gap products like the next generation of AI-native CRM are being built to close.
Consumers want disclosure, not just competence
Memory isn’t the only trust issue. The report also surfaces a widening expectations gap around transparency. Seven in 10 consumers believe AI agents should identify themselves clearly at the start of every conversation. Only 22 per cent of APAC businesses currently do this consistently, a gap wide enough to suggest many brands still treat AI disclosure as optional rather than a baseline expectation, despite mounting evidence that the AI trust gap is now a commercial liability, not just an ethical nicety.
Consumers want more than a heads-up, too. Fifty-eight per cent want the option to switch to a human agent on request, 57 per cent want assurance that AI-driven actions require human sign-off before execution, and 51 per cent want visibility into exactly what data an AI system can access. This is a more demanding, more specific version of trust than “does the bot work” — it’s “can I see and control what the bot knows about me.”
Also Read: The AI trust gap: Why SEA startups need proof before they scale
Businesses aren’t entirely ignoring the ask. Some are prioritising explainability features (47 per cent), shorter paths to a human agent (44 per cent), and giving customers more say over whether their data feeds AI model training (39 per cent). But across nearly every trust metric Twilio measured, the gap between what consumers expect and what businesses actually deliver remains wide.
Gemma Calvert, a professor of consumer neuroscience cited in the report, framed the stakes bluntly: when AI is deployed before it can handle the full range of real customer needs, consumers don’t blame the technology in isolation; the failure becomes part of how they experience the brand itself. It’s a useful corrective to the instinct, still common among product teams racing to ship, that a forgetful chatbot is a minor UX bug rather than a brand-trust problem with compounding costs.
The road to AI-to-AI service
Despite the friction, APAC consumers aren’t turning away from AI-driven service — quite the opposite. Sixty-eight per cent say the bots they’ve dealt with have genuinely improved over the past year, and a notable 65 per cent are already comfortable with AI agents from different companies communicating directly with one another to resolve issues, without a human shuttling information back and forth. That comfort level matters: it is the precondition for the kind of agent-to-agent commerce that vendors building payment agents are betting will become the region’s next infrastructure layer.
Everyday task delegation to AI is already mainstream: 86 per cent of consumers are comfortable letting AI schedule appointments, 85 per cent trust it with dinner reservations, another 85 per cent with processing returns, and 82 per cent with picking concert seats. None of these are trivial — each requires a degree of confidence in AI judgement that would have looked premature just a few years ago, and echoes the broader question of whether it’s possible to install real judgement into AI agents at all.
Also Read: The app worked, the product didn’t: Can we install judgement into AI agents?
Businesses are positioning for a future where delegation deepens further. APAC leaders expect AI agents to handle 65 per cent of all customer service interactions by 2027, up sharply from 52 per cent today. And 91 per cent of business leaders say they are actively building AI workflows for proactive support, reaching out before a problem is even reported, rather than waiting for a complaint to land.
What it means for Southeast Asia’s builders
For the region’s fast-growing customer engagement, fintech and e-commerce platforms, the Twilio findings amount to a fairly unambiguous signal: the race to deploy AI agents is outpacing the race to make those agents actually remember customers and disclose themselves honestly. That gap won’t close on its own, and it certainly won’t close by adding another point solution on top of an already fragmented stack.
As AI-to-AI communication moves from novelty toward norm, the startups and enterprises that invest early in unifying fragmented customer data, and in being upfront about when and how AI is being used, stand to capture a trust dividend their competitors are currently leaking away, interaction by frustrating interaction. In a region where word-of-mouth and app-store reviews can make or break a consumer brand overnight, fixing AI’s memory problem may prove just as commercially important as building the AI in the first place.
The post “AI amnesia” is quietly costing Southeast Asian brands their customers appeared first on e27.