When flights get rescheduled, food orders arrive damaged, or packages go missing, customers expect swift resolution. Yet across Malaysia's digital landscape, these routine problems increasingly funnel into the same dead end: an artificial intelligence chatbot incapable of helping. As businesses nationwide deploy AI-powered support systems to manage customer inquiries at scale, a troubling pattern has emerged—one where the technology meant to enhance service instead creates frustration, erodes trust, and drives consumers toward competitors.
The Malaysia Cyber Consumer Association has documented a sharp rise in complaints about automated customer support systems in recent years. According to the association's president, Siraj Jalil, the core issue is what industry observers call the "infinite loop" phenomenon. These chatbots function by recognising only predetermined keywords and triggering scripted responses. When a customer's problem falls outside the narrow parameters the system was trained to handle—which is often the case with nuanced or complex issues—the chatbot responds by recycling the same FAQ links regardless of relevance. Consumers find themselves trapped in an exhausting cycle, pressing buttons and typing queries that generate identical unhelpful responses, with no clear pathway to human assistance.
The root cause of this dysfunction, according to Henrick Choo, managing director of IT services firm NTT Data Malaysia, lies in how many organisations conceptualise the purpose of customer service AI. Rather than designing systems to resolve problems efficiently, companies have optimised chatbots to deflect customers away from human agents. The success metric has inverted: instead of measuring how many issues were actually resolved, management now tracks how many customers were prevented from reaching a live representative. While cost reduction matters in today's competitive Malaysian market, this approach frequently backfires. When customers perceive the chatbot as a barrier rather than a helper, they grow frustrated, return with repeat complaints, and damage the company's reputation—ultimately generating far more work for human teams than prevention would have cost.
Research from John Hopkins University validates this consumer perception. Assoc Prof Evgeny Kagan and colleagues identified what they call "gatekeeper aversion," the natural resistance users feel when interacting with systems designed to block access rather than provide help. The experiments revealed that this aversion takes root immediately and persists throughout the interaction. Users quickly sense that chatbots function as gatekeepers protecting human employees' time rather than as genuine problem-solvers. This instinctive distrust intensifies when the system offers no visible option to escalate to a human, or when such escalation proves frustratingly difficult to locate.
The frustration compounds dramatically when the handoff finally occurs. Siraj explains that many Malaysian consumers report a phenomenon he terms "contextual blindness." When a customer is transferred to a human agent, that agent often has no access to the conversation history. The consumer is greeted with generic automated prompts like "How can I help you today?" and forced to recount their entire problem from scratch. If the connection drops, they rejoin the queue and repeat the ordeal. This is not merely inconvenient—it communicates profound disrespect for the customer's time and intelligence. Conversations are deleted if connections refresh, problems compound, and what began as a simple issue becomes a test of patience.
Choo identifies the handoff point as where most Malaysian companies lose customer trust. Customers, he observes, are generally willing to attempt self-service options. They accept that not every query requires human attention. Yet their tolerance evaporates when trapped in what he calls the automated "doom loop"—sequences of prompts and failed resolutions with no emergency exit. The solution, he emphasises, requires passing complete context to human agents: full chat transcripts, customer profiles, transaction histories, sentiment analysis, and recommended next steps. A human representative armed with this information can resume work immediately and actually help. Without it, both parties waste time and frustration multiplies.
The design failures, Choo stresses, extend beyond the chatbot itself into the underlying infrastructure. Many systems lack proper integration with operational platforms—the customer relationship management databases, billing systems, identity verification tools, and approval workflows that human agents access routinely. Chatbots can retrieve FAQ information readily, but resolving actual account issues requires deeper system access. A bot can provide an answer; acting on that answer demands permissions and tools most chatbots were never granted. The fundamental barrier is integration depth. Companies often connect chatbots only to knowledge bases—repositories of information—rather than to the systems where real work happens. This architectural shortcut guarantees failure for any problem requiring genuine action.
Khalil Nooh, CEO and co-founder of Malaysian language model company Mesolitica, highlights another widespread failure: companies assume they can simply upload all existing documents into a large language model and expect flawless operation. Knowledge bases, however, frequently suffer from what he calls "knowledge-base rot." Pricing information becomes outdated, policies contradict each other, terms expire, and archived documents accumulate. When AI systems attempt to retrieve information from corrupted databases, their retrieval precision collapses. Instead of admitting uncertainty, large language models "hallucinate"—generating plausible-sounding but factually incorrect responses with complete confidence. A customer receives false information, takes action based on inaccuracy, and when problems arise, blames the company.
Nooh warns that some Malaysian organisations labour under a fundamental misconception: that AI-powered chatbots should replace customer support entirely. This assumption ignores basic reality. Complex issues will always require human judgment, especially when proper escalation mechanisms are absent. Organisations lack trained human frontline staff familiar with underlying systems. Support channels deteriorate when humans are removed entirely. The solution requires integration: chatbots handling routine, clearly-defined queries while seamlessly routing genuine problems to equipped humans. Without this balance, companies sacrifice quality and incur reputation damage far exceeding any labour savings.
For Malaysian consumers and businesses alike, the implications extend beyond mere frustration. Trust in digital services erodes when interactions feel designed to obstruct rather than assist. Companies that implement AI thoughtfully—ensuring full context passage, deep system integration, proper escalation pathways, and accurate knowledge bases—gain competitive advantage. Those that deploy chatbots primarily to reduce costs without addressing fundamental experience design questions will find customers increasingly defecting to competitors offering genuine support. The technology itself is not the problem; the execution reflects whether organisations prioritise customer outcomes or cost metrics.
