The labor cost savings from replacing agent duties are not significant.
Deep learning must evolve to learn patterns and identify errors on its own.

As enterprise chatbots, which were expected to reduce labor costs, have failed to produce significant results, it has been pointed out that revenue generation should be sought by applying deep learning technology for personalized recommendation functions.

Wisenut (CEO Yong-seong Kang), an AI solution development company, stated in a recent survey of its clients that chatbots have not been effective in actually reducing manpower.

Among the Voice of Customer (VoC) types, simple, repetitive inquiries such as delivery, general merchandise, and system inquiries are being replaced by chatbots, but claims, such as delivery delays and accident handling, are often handled directly by agents. In other words, while approximately 45% of VoC can be automated through chatbots, the effect of reducing manpower for customer service has not been significant. This is because VoC is handled on a standby basis rather than by surplus agents.

At an artificial intelligence seminar hosted by Tui Consulting, Director Jang Ju-yeon of Wisenut explained, “In particular, customer service that is not affected by language barriers or time, as well as repetitive inquiries, can be replaced by chatbots, but agents play a significant role in improving the quality of customer service.”

When will deep learning capabilities evolve?

Currently, chatbot technology is capable of handling simple responses. Although it is being introduced and utilized in various fields ranging from food ordering to education, finance, and quotation inquiries, there is little difference in chatbot APIs. When a customer asks a question, it processes natural language errors such as typos and spacing to find an answer from a predefined list of conversations. Chatbot providers review incorrect feedback at regular intervals and carry out 'corrective work'.

Chatbots have not yet evolved to the point of deep learning capabilities that allow them to independently study patterns and identify errors. As Elon Musk has stated that he will focus on reinforcement learning to build open AI technology, research into artificial intelligence technology is currently ongoing. Imcloud (CEO Lee Doo-sik), a chatbot service provider, is also researching reinforcement learning capabilities that allow chatbots to converse with each other to practice situational speech functions and intent recognition.

Senior Vice President Song Ho-seok, who gave a presentation at the SW Engineering Technical Seminar, stated flatly, “While it is possible to build a chatbot within a week, it is difficult to build one capable of conversation with current technology.” Song described natural language processing capabilities as “starting from scratch,” explaining that it is a challenging task.
He went on to say that since the customers of chatbot services are 'companies,' it is important to provide a management system, such as system errors, rather than feedback on the chatbot service. It is pointed out that the chatbot field is still in its infancy and that growth in related services is necessary.

For example, to accumulate data quickly, the entire service process must be done via chat, but considering that clients are not yet familiar with chatting, a hybrid chatbot that combines clicking and chatting must also be provided.

The conclusion is that data accumulation is important, and thorough prior planning is necessary from the start of implementation.

To expect additional revenue beyond the establishment of a chatbot, customers need to obtain marketing materials based on data collection. This is because, while there were limitations in proposing new products to customers over the phone, emotional marketing becomes possible by providing appropriate recommendations and advertisements based on data learned from the chatbot.

Director Jang of Wisenut also emphasized the importance of data accumulation. He stated that thorough planning is necessary to build a sustainable service, rather than a service that is forgotten once the chatbot trend passes.

“To expand the scope of utilization, one must be aware of knowledge about chatbots, management plans for data expansion, and legal procedures regarding the use of customer data in advance,” he said, advising that chatbot services can succeed only if approached from a planning perspective.