In the digital age, intelligent customer service has become an important bridge between enterprises and consumers, but there are many problems hidden behind it. This article delves into the dilemmas faced by intelligent customer service in practical applications through a number of real cases: from the inability to understand the real needs of users and answer questions, to the obstacles of transferring to manual customer service, to the unfriendliness of special groups such as the elderly, intelligent customer service seems to be gradually becoming a kind of “technical shackles”.
“The typhoon is coming, can I still go to Hainan?”
When Ms. Li sent an inquiry to the intelligent customer service, there were only three days left before her departure. She booked a package itinerary on a travel platform, and once she encounters bad weather on the day of departure, she may not only be stranded at the airport, but may also lose a full amount.
But the first reply she got was: “Dear, our platform provides a lot of guides for popular tourist destinations, you can take a look first!” ”
She asked again, and the customer service jumped to the order page; She asked again, and the intelligent customer service enthusiastically introduced the attractions.
“I don’t want you to recommend any strategies, I’m asking if there is a typhoon response plan.” Ms. Li repeatedly typed and tried to communicate, but it always seemed like she was talking to a machine that couldn’t understand. She finally typed the words “Transfer to Labor” in the dialog box, and a new window popped up: “Please verify the order number, personal identity verification, and describe your problem in detail.” ”
After she did so, she was asked to select the type of problem and wait for the system to “recognize” again. In this way, in the gap of the “closed loop of the process”, Ms. Li was trapped for more than 20 minutes, and the page finally jumped out as “transferring to manual customer service”. She thought it was soon. After a few seconds, a prompt pops up: “The current manual customer service is busy, and the expected waiting time is more than 30 minutes.” ”
“I collapsed.” Ms. Li recalled, “At that moment, I really felt that the customer service was not helping me, but dragging me down. ”
B-end product manager’s ability model and learning improvement
The first challenge faced by B-end product managers is how to correctly analyze and diagnose business problems. This is also the most difficult part, product design knowledge is basically not helpful for this part of the work, if you want to do a good job in business analysis and diagnosis, you must have a solid …
View details >
Ms. Li had no choice but to continue waiting, during which she asked about the progress many times, but the customer service replied “please be patient” every time, and there was no substantive solution. In the end, she missed the time limit to adjust her itinerary or negotiate a refund, and had a very unpleasant travel experience.
Ms. Li’s story is not special. Relevant data shows that in 2024, complaints about intelligent customer service nationwide will increase by more than 50% year-on-year. The most common keywords are “difficult to transfer labor”, “answering questions that are not asked” and “ineffective communication”.
Smart customer service looks smart: it’s responsive, tireless, polite, and even recommends related products. But it is like a student who always answers wrong questions, only repeating the sentence patterns he has learned, and “has no perception” of the real emotions and semantics of human beings. In addition to standardization, those problems that are not preset by the system – such as typhoons, change of bookings, sudden illnesses, emotional loss of control…… Everything else is helpless.
And when you try to bypass it and find a “real person”, you find that the road has long been blocked by “technology”.
01 “Robot customer service, more patient than you”
Huang Yiran is a young man who calls himself an “online shopping expert”. For him, customer service chat records are like a copy of the game, whether it is applying for a refund, urging delivery, or asking for discounts, he knows where to order, what to say, and even which platform’s customer service is “good-natured”.
But most recently, he was completely defeated by “an angry customer service”.
“That day I wanted to ask a logistics question, obviously the content I filled in was written clearly enough, but it kept recommending me ‘guess the product you like’, and I was almost driven crazy.” Huang Yiran said.
He tried to type “I want to find a human customer service”, but the other party replied: “I can solve most problems for you, please describe your needs first.” He insisted on entering “manual”, and the system repeatedly popped up problem classification options, allowing him to choose step by step: order, after-sales, logistics, evaluation…… After one pass, the prompt “Please wait for manual customer service to access” finally appeared.
“I thought it was coming soon, but it started playing me background music.” It was a brisk background piano song, and the system prompted that “manual customer service is queuing, and the expected waiting time is more than 15 minutes.”
After waiting for more than ten minutes, the page suddenly popped up: “Hello, what questions do you need to consult?” He breathed a sigh of relief, thinking that the real person was online. But after entering the question, the other party’s reply changed back to “guess what you like……” Only then did I find that the so-called “manual customer service” is still an AI with human skin.
“It’s like a fake punch.” Huang Yiran said, “You never know who is talking to you in the next second – customer service is like a person, but not a person.” ”
In 2024, iiMedia Research data shows that 58.6% of users agree with the “efficiency” of intelligent customer service, and 66% of users agree with its “sense of experience”, but more than 30% of users still express dissatisfaction with the overall user experience. Among them, the most frequently complained about are “unable to transfer to labor” and “answering questions”.
It’s not difficult to understand behind this.
An e-commerce practitioner said: “For e-commerce, a big feature is that it will be divided into peak period and big promotion period within a year. During the promotion, when a large number of inquiries flocked in, it was impossible to deal with such a large number of inquiries in a timely manner by manual customer service alone. Now the platform has assessment requirements for merchants, and generally if the customer does not reply within 1 minute, the merchant will be deducted points. Therefore, many merchants choose to reply with intelligent customer service first, and only choose to transfer to manual customer service when intelligence cannot solve the problem. ”
“To put it bluntly, intelligent customer service is not for service, but to block the first wave of artillery fire.”
Another practitioner who has been engaged in customer service said bluntly: “At the beginning of training, the goal is to ‘stabilize the other party’. Later, AI came, and companies preferred to use it because it was never annoying, tired, or quarreled with customers. ”
Cost is also an unavoidable reason.
Zheng Juncheng, CEO of Beijing Guoran Zhihui Technology Co., Ltd., said: “Taking e-commerce platforms as an example, merchants have to face a large number of repetitive and process-oriented problems every day, and there is a ’28 law’ here – that is, about 80% of the users, and the consulted questions are concentrated on 20% of the problems.” A human customer service agent may reply to the same question dozens or even hundreds of times a day, which is quite boring. The top gold medal customer service can only receive more than 200 customers a day at most. If the customer service system has a strong ‘concurrency’ ability, it can reply to multiple customers at the same time, and there is no upper limit on the number of replies in a day, and it can be online 7*24 hours a day, which is not possible for manual customer service.
According to Zheng Juncheng, ordinary small and medium-sized businesses can build an AI customer service system for 20,000 to 30,000 yuan a year, which is far lower than the cost of hiring three customer service employees.
02 “The old man said it was a little girl, and her voice was very nice”
“My mother thought she was chatting with a real person.”
When Mr. Liu talked about this, his tone was a little angry, and he didn’t know how to explain it.
That afternoon, my mother, who was in her 70s, received a customer service call, and the other party had a gentle voice and standard Mandarin, and took the initiative to ask her: “Hello, we have sent you a coupon for the goods you ordered before, so that I can send you a text message to confirm?” ”
The old man agreed and said “thank you, little girl” three times in a row.
“But I knew as soon as I heard that it wasn’t human.” Mr. Liu said. He answered the phone and asked a few questions, but the other party could only repeat the words “coupon” and “SMS confirmation” until he said “you are a robot” and hung up there.
“My mother still doesn’t believe that she is chatting with AI, how natural that tone sounds. But she didn’t realize at all that if it was a scam call, she had already walked in step by step. ”
This is not the most troublesome thing.
What really makes Mr. Liu anxious is that the elderly cannot use intelligent customer service to handle any after-sales affairs – even the most basic refund application.
“Her eyes are not good, there are many small pictures on her mobile phone, she can’t find the customer service entrance at first, and she doesn’t know what to enter when she finds it.” Mr. Liu once taught his mother how to contact customer service, but every time the old man lost a question, he received an “automatic recommendation from the system”: check the order details, read the return rules, and check the product review.
“She couldn’t understand what these were at all, and she thought she was doing it wrong, so she became more and more panicked.”
Mr. Liu thought about helping his mother do it, but the platform account is bound to a mobile phone number, and face recognition or verification code confirmation is required every time, and sometimes customer service also requires voice verification. “She can’t figure it out herself, I’m still in a hurry.”
The old man just wanted to ask clearly: Can I return? As a result, the system asked her ten options, and she couldn’t even understand the questions. ”
To complicate matters, some platforms don’t even have a manual customer service entrance, or hide the “transfer to manual” button behind multiple pages. For the elderly, this is not “difficulty in operation”, but “physical isolation”.
“For her, finding a customer service is like walking a road where she doesn’t know if there is an end.”
According to the “Survey Data on the Development of China’s Intelligent Customer Service Market and Consumer Behavior” released by iiMedia Consulting in 2024, the inability to solve personalized problems, answer blunt machinery, and accurately understand questions are the three most unacceptable shortcomings of intelligent customer service; and 30.98% of users reported that the current intelligent customer service cannot take into account the elderly, the disabled and other groups.
Mr. Liu still remembers the sentence his mother repeatedly asked him in front of the customer service page: “Did I order the wrong thing?” The tone was not self-blame, but dazed.
03 Technology is not the original sin, shirk is
In the eyes of most enterprises, intelligent customer service is a “cost-effective business”:
It does not ask for leave, does not get tired, does not talk back, does not resign; It requires no training, no overtime pay, and no emotional breakdown.
In the high-concurrency traffic field, intelligent customer service has become the first line of defense, or even the only one.
From the point of view of business logic, it is truly blameless.
But the problem is precisely – it is too obedient and scary to hear.
Miao Fang, deputy director of the Department of Artificial Intelligence of the School of Information and Communication Engineering of Communication University of China, once said: In recent years, intelligent customer service systems based on large models rely on natural language processing technology, with deep learning capabilities such as semantic understanding, intent recognition, and contextual understanding. “In the early stage of the establishment of the intelligent customer service system, it is necessary to ‘feed’ the ‘corpus’ containing user questions and corresponding answers to the model, so that the intelligent customer service can continuously ‘learn’ the user’s question preferences in different scenarios, so as to form a ‘knowledge base’. In the specific use process, after the user asks a question, the intelligent customer service immediately retrieves the corresponding answer in the ‘knowledge base’ in the background, shortening the path between the question and the answer, which greatly improves the efficiency of customer service reply. ”
In real life, the user’s problems are never standardized.
Typhoons, missed flights, family members falling ill, wrong addresses, unexpected worries – these “accidents” are the norm in human life. But for AI customer service, they are “abnormal data”.
“Human semantics themselves also have a certain ambiguity, in the actual dialogue scene, the user’s questions are myriad, the demands behind the questions may be different, and even some users do not say what their demands are. This makes it difficult for intelligent customer service to judge the user’s true intention and understand the user’s emotions. The human customer service will effectively ask questions step by step according to your expression until it locates your accurate demands. From this point of view, intelligent customer service cannot replace manual customer service at present. Miao Fang said.
As Peng Deyu said: “AI customer service can be efficient, but it should not be indifferent; can be intelligent, but it cannot replace human understanding. ”
The existence of intelligent customer service should not be a threshold or a wall. It should not be a “shield” for enterprises to evade responsibility and delay feedback.
In fact, the solution is not without a solution: at the technical level, the “agent” system can be introduced to enhance the external information call ability of customer service, so that it can truly “move”; At the product level, a clearer one-click “manual transfer” function can be designed, especially for the elderly to set up barrier-free service paths; At the enterprise level, a more reasonable customer service ratio can be formulated and manual response mechanisms can be retained in key links.
It’s not a technical problem, it’s a matter of attitude. Consumers are not afraid of you using AI, but they are afraid that you will use AI as a shield.
People’s basic expectations for customer service have never changed – not to have perfect words, but to have a response that “understands people”.
AI may solve 90% of our standard problems, but it is the remaining 10% that often determines whether a service is “human”.
After all, we are not dealing with a system, but with an emotion, a trust, and a request that needs to be answered.
(Interviewees are pseudonyms upon request)
