When travellers need to reschedule a flight, food delivery customers discover missing items, or e-commerce shoppers receive incomplete orders, they typically turn to the same solution: contacting customer service. Yet as Malaysian and regional companies deploy artificial intelligence powered chatbots to manage the overwhelming volume of daily inquiries, a troubling pattern has emerged. Customers increasingly find themselves unable to resolve problems and unable to reach a human representative willing to help.
Social media platforms including X and Reddit overflow with accounts from frustrated users describing their encounters with unhelpful automated systems. The Malaysia Cyber Consumer Association has documented a sharp rise in complaints about customer support platforms in recent years, with president Siraj Jalil identifying a particularly vexing problem he terms the "infinite loop" phenomenon. These systems are programmed to respond only to specific keywords, and when faced with issues that fall outside their narrow parameters, they simply redirect users back to the same FAQ pages repeatedly, leaving consumers circling endlessly without finding resolution.
The root cause lies in how companies approach AI implementation. Henrick Choo, managing director of NTT Data Malaysia, explains that many organisations designed their chatbots to shield human agents rather than genuinely solve problems. The metrics that drive these implementations measure how many customers were deflected from speaking to real people, not how many issues were actually resolved. For Malaysian firms operating under tight cost constraints, this approach initially appeared attractive, but the strategy frequently backfires, generating more frustration, repeat contact attempts, complaints, and ultimately damaging the company's reputation.
When customers interact with these systems, they immediately sense that the chatbot functions as a barrier rather than a helpful tool. Research from Johns Hopkins University in the United States describes this psychological response as "gatekeeper aversion". The study, led by Associate Professor Evgeny Kagan, found that users recognise chatbots as inexpensive first-line responders designed to protect the time of more expensive human staff, and this perception proves remarkably resilient. Users assume from the start that chatbots will fail them, and they resist engaging with these systems when they have legitimate problems needing resolution.
The frustration intensifies when chatbots lack a straightforward mechanism allowing customers to immediately escalate to a human agent. The situation deteriorates further when customers finally reach a live representative, only to discover that no information transferred from the chatbot interaction. Siraj describes how consumers experience "contextual blindness", where their entire interaction history disappears if a connection refreshes or times out. They are then forced to repeat their entire explanation to the human agent, who greets them with the generic question "How can I help you today?" after the customer has already spent time explaining their problem to the automated system.
This handoff failure represents perhaps the most critical juncture where companies lose consumer trust. Choo emphasises that while many customers willingly attempt self-service options, they quickly become frustrated when they discover no clear exit from what he calls the automated "doom loop". The difference between an efficient experience and an infuriating one hinges on context. When a customer has already explained their situation to the AI, the human agent should have access to the complete chat transcript, customer profile, previous transactions, sentiment analysis, and recommended next steps. Instead, most customers encounter agents starting from zero information.
Choo identifies the underlying problem not as a limitation of artificial intelligence technology itself, but as a failure in user experience design. Companies frequently neglect to equip their AI systems with the permissions and tools necessary to take actual action. Retrieving information from a FAQ page proves simple, but resolving an account issue requires the chatbot to access customer relationship management systems, billing databases, identity verification tools, approval workflows, audit trails, and compliance frameworks. Integration depth determines whether an AI system can access the same tools and data that human agents use to actually resolve problems. Many organisations connect chatbots only to knowledge bases while leaving them disconnected from the systems where real work occurs.
Another overlooked dimension involves the quality of the information feeding these systems. Khalil Nooh, CEO and co-founder of local language model company Mesolitica, points out that outdated or incomplete databases and FAQ pages create additional problems for consumers. Companies often assume they can simply upload all their documents into a large language model optimised for information retrieval, expecting flawless performance without additional consideration. In reality, most knowledge bases suffer from what Nooh terms "knowledge-base rot", containing obsolete pricing, conflicting policies, and expired terms. When retrieval accuracy collapses due to poor data quality, the language model begins producing hallucinations—confident but inaccurate information that further frustrates customers.
Some organisations labour under a fundamental misconception that AI powered chatbots should supplant human customer support entirely. This approach ignores the critical need for proper escalation pathways when issues remain unresolved, and it removes human frontline agents who understand the systems and can provide contextualised assistance. The integration between automated systems and human support requires careful orchestration rather than simple replacement of one with the other. Without this balance, customers inevitably encounter situations that exceed the chatbot's capabilities with no effective route to resolution.
For Malaysian businesses facing competitive pressure to reduce operational costs, the temptation to implement cost-cutting chatbots remains powerful. However, Choo's analysis suggests that implementing AI primarily to reduce agent contact frequently produces the opposite result: increased frustration, more repeat contacts, escalated complaints, and damaged brand reputation. The apparent savings from fewer human agents evaporate when customers abandon the company or spread negative experiences through social media. Building effective AI customer service requires genuine commitment to solving problems rather than deflecting them, ensuring seamless handoffs between automated and human systems, maintaining accurate knowledge bases, and granting AI systems the tools necessary to take meaningful action.
The path forward demands that Malaysian companies reconsider their priorities when deploying AI in customer service. Rather than optimising for the number of contacts deflected from human agents, firms should measure success by the percentage of issues resolved on first contact, whether through AI or human assistance. This shift in perspective would naturally lead to better system design, more comprehensive information sharing between automated and human agents, and investments in AI systems that can actually perform transactions rather than merely retrieve information. Without these changes, consumers will continue experiencing the frustrating phenomenon of being trapped in endless automated loops, unable to reach anyone capable of genuinely helping them resolve their problems.
