Why can’t the “explodingly popular” ChatGPT become an intelligent customer service?

Any emerging technology is like a singularity, full of infinite possibilities and unlimited imagination. You can never imagine how it will appear in front of people.

Before the 21st century, the idea of ​​an “AI Big Explosion” seemed to be just the unfounded and worrying idea of ​​science fiction writers.

Nowadays, more and more people are beginning to seriously think about a question: When the technological singularity arrives, are we ready?

1. The "awakening moment" of customer service

Since its launch on November 30, 2022, the various performances of the new generation of generative artificial intelligence chat robot ChatGPT, It's quite amazing.

From answering questions continuously, generating summaries, translating documents, to classifying information, writing code, writing scripts, doing homework and writing papers, ChatGPT can handle almost everything with ease.

Moreover, ChatGPT will also question the premise of the question and even reject inappropriate requests.

As an important milestone in the history of AI development, ChatGPT will have an influence that cannot be underestimated in certain industries, such as customer service.

Gartner estimates that there are approximately 17 million customer service centers in the world today. "Gartner Vice President Analyst Daniel O'Donnell believes, "Many companies face challenges such as customer staff shortages and rising labor costs. These costs account for approximately 95% of customer service center costs. Conversational artificial intelligence can make customer service more efficient, and at the same time It can also improve customer experience. ”

According to Gartner predictions, conversational artificial intelligence will help customer service centers reduce labor costs by US$80 billion by 2026, and 10% of agent interactions will be automated, which is more than the current level of automation achieved by AI. The estimated 1.6% has increased.

In fact, intelligent customer service is not a new product.

After more than thirty years of customer service system With the development of the industry, today's intelligent customer service system has moved from a single modular product to an integrated model of "service marketing collaboration management", all scenarios and all channels.

In the 1990s, the Internet had not yet become popular, customer service mainly relied on telephone communication, and call centers mainly used hard switching technology.

Entering the year 2000 , the Internet began to spread, traditional customer service software entered large enterprises, and call centers based on softswitch technology appeared.

Around 2010, mobile Internet, cloud computing, big data and other technologies began to be applied, SaaS-based cloud call center and cloud customer service software emerged.

Empowered by the new generation of AI technology, the customer service system is integrated with the Internet, and innovative intelligent service models are introduced into the customer service center.

In recent years, the new generation of customers’ demand for service experience has escalated, allowing customers to contact all link nodes from single after-sales service to brand building and even the customer life cycle.

For enterprises, they need to understand the minds and behaviors of customers in various links and scenarios, and use precise strategies to Each touch point enhances interaction with customers.

With pre-service in the marketing stage and superimposed marketing in the service stage, the boundaries between service and marketing behaviors are increasingly blurred.

As a result, as the boundaries of customer service continue to broaden and deepen, new growth space appears.

Intelligent customer service begins to enter more business scenarios based on the customer service provided by service companies.

In other words, all aspects of communication between enterprises and customers, that is, the entire process of pre-sales, sales, and after-sales, have become the layout scope of intelligent customer service manufacturers.

For example, if a CRM system is integrated into the intelligent customer service system, companies can carry out customer analysis and data mining based on the CRM system to greatly increase the probability of closing a deal and shorten the marketing development cycle.

For customers who have communicated and are interested in making a transaction, you can provide their personal details (industry, occupation, education level, income level, etc.), consumption information (consumption standards, consumption habits, Brand tendencies, etc.), friend circle characteristics (likes, active time, etc.), and needs are carefully prepared and managed, and tags can be set freely, so that customers’ personalized needs are clear at a glance.

AI is not a “master key”

However, the experience of intelligent customer service is not all good memories. At present, many intelligent customer service still have many problems, the most obvious one is "answering questions that are not correct".

The "2021 China Intelligent Customer Service Satisfaction Survey Report" shows that only 9.6% of users believe that intelligent customer service has higher problem-solving capabilities than manual customer service.

The same answers (59.1%), repeated cyclic operations (50.6%), and answers that do not answer the question (47.3%) are the main problems encountered by users when using intelligent customer service.

Compared with previous production-based conversation tools, ChatGPT has been greatly improved.

ChatGPT introduces "artificially labeled data reinforcement learning" based on the GPT3.5 large-scale language model.

That is, continuous fine-tuning through manual feedback allows the language model to achieve better understanding.

For example, learn to judge what kind of answer is high-quality (information-rich, content-rich, helpful to the user, harmless, does not contain discriminatory information, etc.).

From a technical point of view, ChatGPT has established a very large corpus through the vast amount of information on the Internet, and learns knowledge from these corpus through deep learning.

The reason why it can give a perfect answer is because the corpus it learns has exactly such context, and it searches it out and displays it to you. In terms of this search capability alone, ChatGPT is not as good as Google, because Google's corpus is much larger and the search accuracy will be better.

In the final analysis, the core issue that determines the experience of intelligent customer service is the constraints of the underlying AI technology.

The first is the restriction of understanding ability. At present, AI's ability to judge the intention of user input is still very limited.

The reason why we are able to understand users is that we have prepared a huge corpus, which contains a variety of questions that customers have asked. This collection is always limited, and The questions raised by users and the ways of asking questions are almost unlimited, which will inevitably lead to deviations in understanding.

The second is the restriction of responsiveness. This involves the technology of knowledge graph, which is a relationship network obtained by connecting all different types of information together.

If you want to give customers satisfactory answers, you need to build a very detailed and in-depth knowledge map for the questions that customers may ask, far beyond what search engines can provide. .

Building such a specialized knowledge graph is still a huge challenge.

Customer service work is not a small talk. This kind of dialogue has a very clear purpose, and this purpose cannot be satisfied by a simple answer.

may involve many domain-specific details. Pre-sales orders, products, selling prices, logistics, etc.

Without mastering this kind of domain-specific knowledge and building a network in these knowledge details through the knowledge graph, it is impossible to provide high-quality responses.

At the same time, the content of customer service replies is time-sensitive, such as iterations of product information, updates of logistics information, etc.

The data of the corpus used for ChatGPT training is as of 2021, and it does not have the ability to quickly update customer service knowledge.

From a cost perspective, ChatGPT training costs millions of dollars, and fine-tuning costs hundreds of thousands. This money is enough to hire many human customer service personnel, which goes against the enterprise The original intention of using intelligent customer service to reduce costs and increase efficiency.

In addition, there is another layer of emotional factors that is always a gap that AI cannot currently cross. Compared with cold smart customer service, being able to communicate directly with customer service staff makes people feel more at ease.

The reason why many customers are disgusted with smart customer service is not that they resist technological progress or deny its rationality, but that smart customer service sometimes cannot solve people's emotional problems.

Communicating with people itself is a release of emotions, but in the face of rational and even meticulous intelligent customer service, customers' emotions are obviously suppressed.

The relevant person in charge of Yunzhisheng believes that if ChatGPT wants to be fully popularized, there are still three problems that need to be solved.

First of all, ChatGPT needs to improve its knowledge reliability and avoid key applications to solve practical problems; secondly, it needs to enhance its ability to integrate real-time information and improve its timeliness; third Third, it is necessary to further reduce service and training costs and solve practical problems within an acceptable cost range.

In addition, industry standards and related regulations also need to be established and improved simultaneously to promote the healthy development of ChatGPT technology.

At present, Yunzhisheng has established a complete technical closed loop of "perception-cognition-generation" in the field of speech recognition technology and natural language understanding.

The cognitive part is mainly composed of the "BERT GPT2 Industry Knowledge Graph", which has carried out in-depth application practice in the fields of smart IoT interaction and smart medical decision-making, and has been awarded the Beijing Science and Technology First prize for progress.

So, even if ChatGPT achieves a good artificial dialogue effect, from a technical perspective, a business perspective and an experience perspective, it is not currently feasible to use it to upgrade intelligent customer service , can only add some unexpected fun in the process of providing intelligent customer service.

How to correctly view intelligent customer service?

While technology brings convenience to people, we must admit the limitations of technology. Before there is a major breakthrough in the underlying technology, we cannot expect a qualitative change in the experience of intelligent customer service.

Secondly, intelligent customer service has become a general trend driven by the application of manufacturers. Although it is impossible to adapt to changes like manual customer service, it can also provide some basic-level precise services.

Finally, the key is to establish a good collaboration mechanism between intelligent customer service and manual customer service so that users can have a smoother experience.

Just like third-level diagnosis and treatment, community hospitals will see small problems and automatically transfer them to higher-level hospitals if they encounter larger problems.

For user experience, what is important is not the experience of intelligent customer service, but the experience of a manufacturer's overall customer service system.

Just like the increasing number of online banking services, it cannot replace offline bank branches. In the future, the relationship between intelligent customer service and manual customer service will not be a simple matter of who replaces whom. The boundaries between the two parties will become increasingly blurred.

Users’ common problems can be dealt with by intelligent customer service. When users need face-to-face communication and guidance, manual customer service should also appear in time.

The ultimate principle is to be user-centered and choose service methods based on the actual needs of users.

In short, facing the sudden new species ChatGPT, we should not easily fall into the illusion of its "brave new world", but should use the unique wisdom of mankind to cooperate with it The rich functions provide maximum convenience for people’s lives.

The above is the detailed content of Why can’t the “explodingly popular” ChatGPT become an intelligent customer service?. For more information, please follow other related articles on the PHP Chinese website!