The problems that plague customer service—a flight needing rescheduling, a delivery containing damaged goods, or missing items from online orders—might seem unrelated on the surface. Yet they converge on a single point of pain: customers attempting to resolve their issues through increasingly automated support channels. As businesses across Malaysia and the region embrace artificial intelligence-powered chatbots to manage the overwhelming volume of customer inquiries, a troubling pattern has emerged—these systems are frustrating rather than helping the very customers they were meant to serve.
The Malaysia Cyber Consumer Association has documented a sharp rise in complaints about customer support systems over recent years, with president Siraj Jalil identifying a specific culprit: the "infinite loop" phenomenon. When chatbots are programmed to recognize only predetermined keywords, they struggle with problems that fall outside their narrow parameters. Instead of acknowledging their limitations, these systems repetitively direct customers to frequently asked question pages, creating what Siraj describes as a "repetitive cycle without an exit strategy." Social media platforms from X to Reddit overflow with user accounts of this exact frustration, suggesting the problem extends far beyond isolated incidents.
The root cause, according to IT services firm NTT Data Malaysia managing director Henrick Choo, lies in how companies measure success. Rather than optimizing for problem resolution, many organizations have built their chatbot deployments around a different metric: keeping customers away from human agents. This distinction proves crucial. "The metric became 'how many customers did we keep away from agents?' instead of 'how many issues did we resolve?'" Choo explains, noting this approach carries particular appeal for Malaysian companies operating under cost constraints. Yet the economics backfire—frustrated customers who cannot get help escalate their complaints, return repeatedly with the same issue, and damage the company's reputation, ultimately generating more work rather than less.
Customers perceive these barriers intuitively. When a chatbot serves primarily as a gatekeeper rather than a problem-solver, users sense it immediately and recoil. Research from Johns Hopkins University in the United States has formalized this reaction as "gatekeeper aversion," with Associate Professor Evgeny Kagan explaining that users enter the interaction expecting the chatbot to fail them. "From the outset, users perceive the risk of chatbot failure to be high, and they don't want to engage," Kagan found. This psychological resistance becomes even more pronounced when customers cannot easily escape to speak with a human representative, transforming the experience from frustrating to demoralizing.
The handoff between machine and human represents another critical failure point in many Malaysian companies' customer service ecosystems. When chatbots finally surrender a case to a human agent, customers reasonably expect that representative to have reviewed the entire conversation history. Instead, they frequently encounter a fresh greeting—"How can I help you today?"—forcing them to explain their entire situation from scratch. Siraj characterizes this experience as intensely disrespectful of the customer's time and emotional energy. Should the live chat connection drop, the customer may find themselves back in the queue, required to repeat the entire process. Choo identifies the handoff as "where many companies lose trust," and the point at which customer goodwill definitively evaporates.
Context proves to be the fundamental differentiator between an efficient support experience and a frustrating one. When a customer has already articulated their problem to an AI system, Choo argues, a human agent should have immediate access to the complete transcript, customer profile, transaction history, emotional sentiment analysis, and recommended next steps. This information transfer rarely happens. The problem transcends the chatbot itself, extending into the infrastructure supporting it. Many companies connect their chatbots to knowledge bases alone, without granting them access to the customer relationship management systems, billing platforms, identity verification tools, and approval workflows that human agents use to actually resolve issues. Integration depth—whether the AI possesses the same system access as its human counterparts—determines whether a chatbot can merely talk about solutions or genuinely implement them.
Design failures compound these structural limitations. Aside from the failure to transmit conversation history, many chatbots lack the permissions and tools to take meaningful action. Henrick Choo observes that retrieving information from a database differs fundamentally from resolving an actual problem. A chatbot can pull up FAQ entries easily, but solving an account issue demands access to multiple interconnected systems alongside comprehensive audit trails and compliance frameworks. When a company deploys a chatbot without providing it these capabilities, the system becomes ornamental—capable of explaining policies but incapable of changing them.
Data quality issues lurk beneath the surface of many chatbot failures. Khalil Nooh, CEO and co-founder of local language model firm Mesolitica, identifies "knowledge-base rot" as a pervasive problem: outdated pricing information, conflicting policies, expired terms, and obsolete procedures lodged within company databases. When large language models attempt to retrieve and synthesize this corrupted information, precision collapses. Rather than admitting uncertainty, the models "hallucinate"—generating plausible-sounding but fabricated information that customers then rely upon. Many organizations labor under the misapprehension that simply uploading all company documents into an LLM will produce a functional customer service tool, overlooking the prerequisite work of cleaning, organizing, and validating that information.
The assumption that AI chatbots can entirely replace human customer service agents represents perhaps the most consequential strategic error. Some Malaysian companies have adopted this vision without establishing proper escalation protocols or maintaining experienced human staff who understand the underlying systems. Khalil emphasizes that this wholesale replacement approach fails to account for the complex, context-dependent nature of genuine problem-solving. When companies cut human staffing without ensuring the AI can genuinely resolve issues, they eliminate their safety net. The chatbot becomes not a helpful first line of defense but an impenetrable wall separating frustrated customers from the human judgment and authority required to help them.
For Malaysian consumers and businesses alike, the implications extend beyond individual frustrations. Each failed chatbot interaction accumulates into broader patterns: eroded consumer trust in online services, reduced willingness to engage with digital-first companies, and paradoxical increases in customer service costs despite supposed automation savings. The region's rapidly expanding e-commerce, fintech, and digital services sectors depend fundamentally on customer confidence in their support channels. As competition intensifies and customer expectations rise, companies that deploy poorly designed AI systems risk ceding market share to competitors offering genuine human support or properly integrated automated systems.
The path forward requires Malaysian companies to reconceive customer service automation not as a cost-reduction tool but as a capability enhancement. Effective AI systems demand robust integrations with operational databases, carefully curated and maintained knowledge bases, clear escalation pathways that preserve conversation context, and human agents equipped to solve problems that machines cannot. The technology itself does not fail; the implementation does. Companies willing to invest in proper integration, data governance, and human-AI collaboration can achieve genuine efficiency gains. Those pursuing automation as a shortcut to lower costs will continue trapping their customers in the very loops they sought to automate.
