When a flight booking goes wrong, a food delivery arrives incomplete, or an online purchase disappoints, customers naturally turn to the support channels they expect will help. Yet across Malaysian companies—from airlines to e-commerce platforms—they increasingly encounter the same problem: an artificial intelligence chatbot that cannot solve their issue and seems determined to keep them from speaking to a real person.
The Malaysia Cyber Consumer Association has documented a sharp rise in complaints about AI-powered customer support systems over recent years, with a particularly troubling pattern emerging. These chatbots are being hard-coded to recognise only predetermined keywords and familiar problem categories. When a customer's issue falls outside this narrow scope, the bot responds by cycling through the same automated solutions repeatedly—typically links to FAQ pages that do not address the specific situation. President Siraj Jalil describes this as the "infinite loop" phenomenon, a trap that leaves consumers repeating themselves endlessly without any apparent path to resolution.
The root cause, according to technology experts working in the region, reflects a fundamental misalignment of priorities. NTT Data Malaysia's managing director Henrick Choo argues that many organisations have designed their chatbot systems not to solve problems but to minimise contact between customers and human agents. "The metric became 'how many customers did we keep away from agents?' instead of 'how many issues did we resolve?'" he observes. For Malaysian companies operating under tight cost constraints, this approach initially appears attractive—reduce headcount in contact centres, deploy AI, watch expenses fall. Yet the strategy produces the opposite effect, generating customer frustration that inevitably circles back as repeated contact attempts, social media complaints, and long-term reputational damage.
Research from Johns Hopkins University provides scientific backing for what Malaysian consumers report instinctively. The study identifies a phenomenon called "gatekeeper aversion," wherein customers develop immediate suspicion of automated systems designed to shield human workers from taking their calls. Associate Professor Evgeny Kagan's experiments found this resistance remarkably persistent. Users perceive a high risk of chatbot failure from the outset and actively resist engaging with automation, particularly when the system lacks a prominent, immediate option to escalate to a human representative.
The frustration intensifies exponentially at the moment of transfer from bot to human. Even when customers finally reach a live agent, many companies have failed to implement information handoff mechanisms, forcing people to retell their entire problem from scratch. When connection timeouts occur mid-conversation—a common technical failure—users discover their chat history has vanished, necessitating another queue wait and another complete explanation. Siraj notes that consumers consistently describe this experience as emotionally draining, disrespectful of their time, and characterised by a sense that the organisation views them as an inconvenience rather than a valued customer.
Choo emphasises that trust ruptures precisely at the handoff moment. Customers often prove willing to attempt self-service solutions, recognising the legitimate operational benefits of efficient automation. What they cannot tolerate is becoming trapped in what he terms an automated "doom loop"—cycling through repetitive prompts and failed resolution attempts without a visible escape route to human assistance. The solution, he argues, lies not in abandoning AI but in fundamentally redesigning how these systems connect to human agents. When a customer reaches a human representative, that person should immediately view the complete chat transcript, the customer's full profile, historical transaction data, sentiment analysis, and recommended next steps. Instead, most Australian and Malaysian companies have implemented systems where the human sees only a new incoming ticket with no context.
The underlying architecture of these customer service ecosystems reveals the true problem. Most organisations have connected their chatbots to knowledge bases but failed to integrate them with the actual systems where resolution occurs. A chatbot can retrieve information from an FAQ page easily, but resolving billing disputes, account issues, or payment problems requires access to customer relationship management systems, billing databases, identity verification protocols, approval workflows, audit trails, and compliance rule engines. Without this integration depth, the bot can answer questions but cannot take action, creating a fundamental disconnect between what customers are told and what the system can actually accomplish.
Mesolitica CEO and co-founder Khalil Nooh identifies another critical failure point: the knowledge bases themselves. Many organisations operate under the mistaken assumption that they can simply deposit all their documents into a large language model and expect reliable, accurate responses. In reality, most corporate knowledge bases suffer from what he calls "knowledge-base rot"—obsolete pricing information, conflicting policies, expired terms, and outdated procedures embedded throughout the system. When an AI model draws from such corrupted data sources, retrieval precision collapses and the system begins generating plausible-sounding but false information, a phenomenon known as hallucination. Customers receive answers that sound authoritative but contradict actual company policy.
Furthermore, some Malaysian and regional organisations operate under a misconception that AI chatbots should eventually replace customer support entirely. This view fails to account for the reality that even well-designed systems require proper escalation pathways for genuinely complex issues. When companies strip away experienced human frontline agents in pursuit of cost reduction, they eliminate the institutional knowledge and judgment necessary to handle edge cases, exceptions, and the uniquely human situations that defy categorisation. The result is a system that becomes progressively worse at its core function as it expands in scope.
Choo frames the challenge not as a limitation of artificial intelligence technology itself but as a failure of experience design and organisational thinking. Companies know how to build systems where context flows smoothly from human to human. The task involves extending that same contextual richness from AI systems to human agents and, critically, giving AI systems access to the same tools and data repositories that human problem-solvers use. This requires genuine integration work across multiple organisational functions—data architecture, access permissions, system interconnection, workflow design, and compliance—rather than simply deploying a conversational interface on top of existing infrastructure.
For Malaysian consumers and businesses, the implications are significant. As competition intensifies across e-commerce, fintech, logistics, and service sectors, customer experience increasingly determines market position. Companies that master the integration of AI and human support—ensuring seamless transitions, complete context transfer, and genuine problem resolution—will capture customer loyalty and reduce operational costs simultaneously. Those that persist with "doom loop" designs will face mounting complaints to consumer protection bodies, negative social media exposure, and gradual customer migration to competitors. The Malaysia Cyber Consumer Association's rising complaint volume signals that many organisations have not yet grasped this reality.
