When something goes wrong with a flight booking, food delivery, or online purchase, customers expect quick resolution through support channels. Yet an expanding reliance on artificial intelligence-powered chatbots has created a frustrating new problem: rather than solving issues, these systems often trap users in what industry experts now call 'doom loops' — endless cycles of irrelevant responses and dead ends.
The Malaysia Cyber Consumer Association has documented a sharp rise in complaints about AI-driven customer support systems, with president Siraj Jalil pointing to a fundamental design flaw at the heart of the problem. Many chatbots are programmed to recognise only specific keywords and cannot adapt when faced with non-standard complaints that require nuance and context. When a query falls outside these narrow parameters, the bot offers only links to generic FAQ pages. Customers then find themselves repeating the same search, hitting the same dead ends, and never reaching resolution — all without an obvious way to escape and speak with a human representative.
What makes this pattern particularly troubling is the underlying business logic driving it. Henrick Choo, managing director of NTT Data Malaysia, reveals that many companies have implemented chatbots primarily as cost-reduction tools designed to keep customers away from human agents rather than to genuinely resolve problems. The success metric becomes 'how many customers did we deflect?' rather than 'how many issues did we actually fix?' For Malaysian companies facing financial pressure, this approach seems attractive — until it backfires. By prioritising cost efficiency over customer satisfaction, these organisations often trigger the opposite outcome: escalating frustration, repeat contact attempts, mounting complaints, and reputational harm that ultimately costs far more than the savings generated.
Research from Johns Hopkins University in the United States provides scientific backing for what many Malaysian consumers intuitively understand. A study by Associate Professor Evgeny Kagan found that users develop what researchers call 'gatekeeper aversion' — a deep-seated distrust of automated systems that clearly prioritise blocking contact over helping. From the first interaction, customers perceive a high risk that the chatbot will fail them, making them reluctant to engage at all. This resistance intensifies when the system offers no clear option to bypass automation and reach a human immediately.
The experience deteriorates further when chatbots finally do connect customers to live agents. Siraj explains that many consumers describe the process as draining and disrespectful of their time, particularly when forced to restart entirely from scratch. The underlying problem: no mechanism exists to pass conversation history from the bot to the human. A customer who spent fifteen minutes explaining their issue to an automated system must then hear 'How can I help you today?' and recount everything to the agent. If the connection drops and the customer returns to the queue, the entire frustrating cycle repeats. This fragmentation of context transforms what should be a simple handoff into another loop of futility.
Choo identifies the handoff between automation and human service as where most companies lose customer trust irretrievably. Customers are often willing to attempt self-service resolution, but patience evaporates when they realise they are trapped in an automated cycle with no visible exit. What distinguishes efficiency from frustration, he argues, is whether the human agent receives complete context. This means the full chat transcript, customer profile, transaction history, sentiment analysis, and recommended next steps. Without these, the human agent operates in darkness, unable to build on previous attempts or understand what the customer has already tried.
Choo emphasises that the limitations are not inherent to artificial intelligence itself but rather stem from poor experience design and inadequate system architecture. Beyond chatbot configuration, the underlying infrastructure matters enormously. Among the most consequential design failures is the absence of real action capabilities. A chatbot can easily retrieve and recite FAQ answers, but actually resolving account issues demands access to customer relationship management systems, billing platforms, identity verification tools, approval workflows, audit trails, and compliance frameworks. Many organisations connect their chatbots only to knowledge repositories while leaving them disconnected from the systems where actual work occurs — the databases and tools human agents use daily. This integration gap essentially guarantees failure on anything more complex than a routine query.
Khalil Nooh, CEO of local language model firm Mesolitica, identifies a separate failure point that compounds the problem. Many organisations assume they can simply upload all their documentation into a large language model and expect flawless performance. This underestimates the degradation of knowledge bases over time. Legacy systems accumulate what Nooh calls 'knowledge-base rot' — outdated pricing information, conflicting policies, expired terms, and contradictory guidelines. When an AI system retrieves from corrupted data, accuracy collapses and the model generates plausible-sounding but entirely false information, a phenomenon researchers call 'hallucination'. Malaysian companies relying on such systems unknowingly feed customers false information while the chatbot remains confidently unhelpful.
The misconception that AI chatbots should entirely replace human customer support further compounds these difficulties. Nooh points out that some organisations remove human frontline agents while deploying insufficient escalation mechanisms for unresolved issues. This creates an impossible situation: complex problems have nowhere to go, human expertise becomes unavailable, and customers face dead-end systems with no pathway to actual assistance. The solution requires abandoning the premise that automation should eliminate human contact, and instead designing systems where AI and humans work in complementary roles, with seamless escalation when needed.
For Malaysian consumers and businesses alike, the lesson is clear: chatbots deployed without thoughtful design, proper system integration, well-maintained knowledge bases, and genuine escalation pathways create nothing but frustration. The cost savings imagined when implementing such systems rarely materialise once accounting for repeat contacts, complaints, regulatory issues, and damaged customer relationships. Companies that succeed with AI customer support treat automation as an assistant to human service, not a replacement for it, ensuring customers trapped in problems can always find a way out.
