A rescheduled flight on a long-awaited trip, a spilt drink in a food delivery order, and a missing item in an e-commerce shipment may seem worlds apart. But when something goes wrong, they all end the same way: with a call, message or chat with customer service.
As companies increasingly deploy artificial intelligence (AI)-powered chatbots to deal with the slew of user queries they receive each day, customers themselves have expressed frustration over both their inability to get their problems resolved and the difficulty of getting a hold of a real human for help.
A simple search on platforms like X (formerly Twitter) and Reddit reveals countless users discussing their individual problems, and more strikingly, how customer service chatbots have more often than not been unable to help resolve them.
The Malaysia Cyber Consumer Association (MCCA) has seen a noticeable increase in customer complaints relating to such customer support systems over the past few years, according to its president, Siraj Jalil.
He explains that consumer frustrations are centred on what is called the “infinite loop” phenomenon, in which chatbots are hard-coded to recognise only specific keywords.
In cases like these, the chatbot will continually provide links to existing FAQ pages when it is unable to handle the non-standard problem at hand due to its specific nuance and complexity.
“This leaves consumers trapped in a repetitive cycle without an exit strategy,” Siraj adds.
IT services company NTT Data Malaysia managing director Henrick Choo says this is due to many AI chatbots being designed to deflect issues away from human agents, rather than resolve and contain problems.
“The metric became ‘how many customers did we keep away from agents?’ instead of ‘how many issues did we resolve?’,” he claims. “This is especially relevant for Malaysian companies under cost pressure.
“Cost efficiency is important, but when AI is implemented mainly to reduce agent contact, it often creates the opposite result: more frustration, repeat contacts, complaints, and reputational damage.”
He adds that many companies use chatbots as a way to reduce call volume rather than to solve problems, saying, “Customers sense this immediately – they feel the bot is there to block them rather than help them.”
According to a study on AI chatbots in customer service from John Hopkins University in the United States, this sentiment is known as “gatekeeper aversion”.
The study explains that gatekeepers, who are the least expensive first-line responders to customer queries, protect higher-paid employees’ time.
In their chatbot experiments, Assoc Prof Evgeny Kagan and his colleagues found that gatekeeper aversion is very persistent and hard to overcome. “From the outset, users perceive the risk of chatbot failure to be high, and they don’t want to engage.”
This can be especially pronounced when the chatbot lacks an option to immediately direct customers to a human, a frustration that is made worse when users finally reach a human agent but are forced to repeat all their information if there was no mechanism for information to be passed on from the chatbot to a human agent.
“Consumers face intense frustration regarding contextual blindness, where the system completely deletes their input history if a connection refreshes or times out,” Siraj says.
He adds that many consumers describe going through this process as draining, disrespectful of their time, and repetitive.
“When a chatbot finally fails and routes the user to a live customer service representative, the consumer expects that representative to have reviewed the chat log.
“Instead, they are (often) met with standard automated greetings like, ‘How can I help you today?’ and are forced to explain their entire grievance from scratch,” he says, adding that if the live chat is disconnected, customers might be forced to rejoin the queue for assistance and repeat the process all over again.
Choo concurs: “The handoff is where many companies lose trust.”
He adds that customers are often willing to try self-service, but become frustrated when they cannot easily exit what he calls the automated “doom loop” – a cycle of repeated prompts and failed resolutions – to reach a human agent.
“Context is the difference between efficiency and frustration. If a customer has already explained their issue to the AI, the human agent should see the full transcript, customer profile, previous transactions, sentiment, and recommended next steps,” Choo explains.
Why does this happen?
The way that Choo sees it, the problem does not just come from the chatbot, but also the underlying systems behind it, including areas such as data and escalation rules. He stresses that these are not limitations of artificial intelligence, but failures in experience design.
Among the most common design mistakes he points out, aside from failure to pass conversation history to human agents, is the lack of permissions and tools for AI customer experience (CX) agents to actually take action.
“Answering is easy. Acting is hard. A bot can retrieve an FAQ, but resolving an account issue requires access to CRM (customer relationship management), billing, identity verification, approval workflows, audit trails, and compliance rules.
“The main barrier is integration depth – whether the AI has access to the same systems, data, and tools that human agents use to resolve customer issues. Many companies connect the chatbot to a knowledge base, but not to the systems of record where real work happens,” he says.
Khalil Nooh, the CEO and co-founder of local language model firm Mesolitica, highlights that incomplete or outdated databases and FAQ pages can also lead to further problems with consumers.
He believes a major misstep is the assumption that a company can dump all its documents into a large language model (LLM) optimised to retrieve and incorporate the information provided to it, and that it will work perfectly.
“Most knowledge bases are not AI-ready. Legacy stores carry ‘knowledge-base rot’ – obsolete pricing, conflicting policies, expired terms – so retrieval precision collapses, and the model hallucinates,” he says.
Khalil adds that some organisations may be under the misconception that AI-powered chatbots should take over customer support entirely, without considering proper escalation when issues remain unresolved, and lacking human frontline agents familiar with the systems.
According to Choo, another issue has to do with governance, where companies are wary of letting AI take more significant actions, such as approving refunds, changing account details, or making financing decisions without safeguards.
He stresses: “Agentic AI will help, but only where permissions, data, governance, and monitoring are ready. Without those foundations, agentic AI simply creates faster failure.”
Agentic AI is an advanced form of artificial intelligence focused on autonomous decision-making to achieve pre-determined goals that requires minimal human supervision.
Choo adds that companies concerned about preemptively allowing AI to perform sensitive tasks should consider a more phased approach when it comes to autonomy.
This could start with high-volume, low-risk tasks such as order status updates, appointment changes, store information and delivery tracking, before moving on to matters like claims and refunds, and account updates.
Meanwhile for cases involving high-value clients, disputes, and regulated decisions, humans should still be kept in the loop, Choo adds.
Coping measures
While some customer service chatbots have a clearly visible button for customers who find themselves at a dead end and want to speak to a human agent, others do not, leaving consumers stuck in a loop.
When caught in such a situation, Siraj says that there may be some workarounds available due to the nature of LLMs.
He advises consumers to “try using explicit trigger keywords such as ‘Speak to an agent’, ‘Human’, ‘Live operator’, or ‘Complain’, which can sometimes bypass the bot’s scripting and force the system to route the queue to a person”.
However, should that fail, he says consumers can try other official escalation channels, such as contacting the company through its verified social media accounts or customer-service pages, where public-facing support teams may respond more quickly. Consumers should keep their complaints factual, include relevant details and avoid making unverified allegations.
“Ultimately, if the company remains unreachable and the issue involves monetary loss or contractual breach, consumers must document everything with screenshots and immediately escalate their case to a formal regulatory body like the Domestic Trade and Cost of Living Ministry, or the National Consumer Complaints Centre,” he says.
From Siraj’s perspective, the responsibility for dealing with these issues falls on the corporate sector, with AI technology serving to improve human customer service rather than replace it.
He says that while such automated systems benefit companies by saving on labour cost, this often comes at the expense of the consumer, who he says can end up forced “to act as an unpaid troubleshooter”.
Siraj adds that while tech-savvy users consider these systems annoying, it can become burdensome for the elderly and those with low digital literacy, who could end up unwittingly excluded from accessing services.
Steps forward
On the other hand, Khalil says the situation is likely to improve, viewing current challenges as early teething issues rather than the norm, which he expects will ease as technology develops and companies adopt more fully agentic customer service chatbot systems.
He believes that agentic AI will be the key to achieving this, with systems that reason, plan, retain context, take action, and keep humans in the loop.
Choo similarly believes that agentic AI will be a turning point. However, the technology will not magically resolve all the issues. He says the real turning point isn’t “AI replacing humans”, but rather “AI moving from answering to resolving”.
He further says that “a good implementation is not just a chatbot on a website”. It would come in the form of an integrated service model, with a clear purpose for the AI, what it should and should not handle, a clean, updated knowledge base shared between bots and humans, along with all the necessary integrations for it to take action and actively resolve issues.
“The right model is hybrid. AI handles volume; humans handle value. AI is best for FAQs, delivery status, appointment scheduling, password reset, account balance checks, simple routing, and summarising cases for agents.
“Humans remain essential for complaints and angry customers; bereavement, health, financial stress, or other emotional situations; disputes and exceptions; policy flexibility; fraud, risk, and regulated decisions; and high-value or complex business-to-business customers.
“AI should make human agents better, not make them disappear. AI agents, IVR (interactive voice response), and agent-assist tools should take care of routine queries, shortening wait times, improving first-call resolution, and freeing human agents for complex, high-value conversations,” Choo stresses.
Meanwhile, Siraj believes that there needs to be a mandate from regulatory bodies to ensure that consumers always have the right to a human customer service agent, with a prominently displayed option to transfer to a live agent in a reasonable timeframe.
“Furthermore, companies that choose to isolate their customer base entirely behind non-functioning automated walls during critical service failures must face heavy statutory penalties for unfair trade practices, and they should be legally required to continuously audit their AI models to ensure they do not discriminate against users based on digital literacy or language variations,” he says.
