HSBC Call Center Secret | Introducing AI to monitor sales conversations, customers’ emotions can be analyzed

Special interview

Author: Hu Xueneng

2020-12-30 07:00

Last update date: 2020-12-30 07:00

When you use bank telephone banking services and the bank operator answers your questions, have you ever thought about artificial intelligence (AI) also participating?

In recent months, HSBC has partnered with Google to introduce the Automated Quality Management (AQM) system to the phone customer service center for the first time. It uses AI and automation technology to monitor the quality of phone banking services and saves 2,400 hours a month. manpower.

The bank’s Hong Kong Wealth Management and Personal Banking (WPB) Phone Banking Center Supervisor Peng Shuzhen, and WPB senior data analysis designer Zhu Xiuhui accepted an interview with "Hong Kong 01", saying that the phone customer service center is now implementing the system, and the customer service director has Whether to clearly inform customers of the relevant terms and rules of the loan products they purchase, and hope to conduct sentiment analysis with AQM next year to understand customer satisfaction with the bank’s services and products

Financial technology (FinTech) has swept the banking industry, and it has become more and more "embedded."

In addition to virtual banks, even we call to the bank "Call Centre" every day to start using AI!

Peng Shuzhen said that the bank has cooperated with Google this time to apply the AQM system to its telephone customer service center for the first time, using artificial intelligence and automation technology to monitor the quality of telephone customer service. AQM is HSBC’s first project to apply artificial intelligence to voice processing. Hong Kong is the first region where the group has launched the system.

Peng pointed out that the system can automatically monitor all the conversations between customer service officers and customers, and analyze whether the staff clearly and accurately explain the terms and conditions. Through artificial intelligence, the system can analyze the tone of the conversation and the language of the customer based on the accumulated data. Satisfaction, thereby improving service quality.

The HSBC Telephone Customer Service Center uses the AQM system.

(Provided by HSBC)

Are Chinese and English artificial intelligence "intelligible"?

It sounds simple, but Hong Kong people like Cantonese and English. For example, "When is my credit card payment deadline?", "I want to see that this credit card has a waive annual fee", "How much is the credit card min pay?" "Wait, how does the system respond to this context?

Zhu Xiuhui pointed out that there is currently a lack of artificial intelligence technology that specializes in Cantonese, and many similar technologies can only be used for single-language dialogue analysis. It is difficult to find suitable options in the market. Therefore, the bank’s dedicated team used Google to use artificial intelligence. Language analysis technology (Natural Language Processing, NLP) constructs a new system.

The process is mainly to convert the mixed speech of Chinese and English into text, so that the artificial intelligence can "understand" the dialogue and analyze the language after understanding the language.

Peng Shuzhen (right) said that HSBC and Google cooperated to apply the AQM system to the telephone customer service center for the first time.

Next to Zhu Xiuhui.

(Photo by Gao Zhongming)

Implement AI in three stages, first ``listen'' to the loan sales dialogue

The bank's telephone customer service center applied the AQM system in three stages. Peng Shuzhen pointed out that the current first-stage system mainly covers conversations involving loan sales. "I hope that from the simplest to the deeper, the machine will have enough time to learn and master more complex conversations."

The system now focuses on monitoring whether the customer service officer clearly informs customers of the relevant terms and conditions of the financial services or products they purchase during the dialogue, such as whether they indicate the identity of bank staff before selling financial products; when the customer agrees to purchase financial services After the service or product, whether the relevant terms and conditions are clearly read out; or whether the transaction is carried out after obtaining clear instructions from the customer.

Peng Shuzhen said that at this stage, the system's criterion is more than 90%, and more products will be incorporated into the system in the sales process.

Zhu Xiuhui revealed that the second phase will be implemented next year, hoping to use AQM for sentiment analysis and to cooperate with the bank's evaluation criteria and standards to understand customer satisfaction with the bank's services and products.

The system will analyze the tone and tone of the customer's dialogue with the bank to know the customer's emotions during the dialogue, such as happy, angry, etc.

As for the third phase, which may be implemented in the second half of next year, it will mainly involve classification work. The system will extract examples of dialogues with high satisfaction scores to strengthen training in the future; dialogues with lower satisfaction scores will also be selected to check whether there is room for improvement.

It is hoped that through the implementation of the three stages, the customer experience and confidence in the bank will be improved.

Zhu Xiuhui said that when the AQM application enters the second phase next year, the system will process and analyze the tone and tone of the conversation between the customer and the bank.

(Photo by Gao Zhongming)

Save 2,400 man-hours to follow up non-compliant conversations faster

Before the system was implemented in phases, the bank could only sample a few telephone recording files for analysis for service improvement or training. Colleagues with lower qualifications or unsatisfactory performance had a higher rate of sampled conversations, and work performance reviews were also delayed "Because we listen to each conversation by hand, and then extract the essence of the conversation, and then someone will do the analysis. You can imagine that each process is manual operation, so it takes a long time. Now we introduce AQM, and all calls will be recorded in the future. All can be monitored, and the data and analysis capabilities are much greater," said Peng.

Peng Xu pointed out that in the first stage of the implementation of the system, the scope of application was to replace human listening with machines. When the system analyzes that there is something that can be done more satisfactorily, it immediately prompts the frontline manager to analyze whether it is necessary to communicate with customers again. Make up the areas that need to be clearly explained, so work efficiency is accelerated.

"When we see improvements, we can immediately follow up with customers. For example, during loan sales, front-line colleagues need to ask customers the reason or purpose of borrowing. Many times customers call to express their desire to borrow money, but the dialogue process is very Soon, the colleague may not clearly mention the relevant question in the conversation. Since the questioning process must be completed, the system can help detect that the frontline colleague has not raised these questions, and the frontline colleague can immediately contact the relevant customer to confirm the other party Purpose of borrowing and perfecting the entire sales process."

With the application of the AQM system, human resources can be saved for flexible deployment.

Peng pointed out that in theory, 100% of the sales calls that are manually monitored require an average of 2,400 hours of work per month. With this system, this time can be saved for training and optimization.

She described the application of the system as an investment, rather than from the perspective of streamlining manpower. "The starting point of the system is to assist wealth management and personal banking departments to provide customers with high-quality experience. For example, the data obtained by the system may be reported to the product department after analysis. Reflect that customers often ask questions in a certain category and take this opportunity to improve the process."

Peng Shuzhen said that 100% of the sales calls that are manually monitored would theoretically use an average of 2,400 working hours per month. With the AQM system, this time can be spared for training.

(Provided by HSBC)

HSBC Call Center Revealed | A response to provocative behavior in handling 40,000 calls per day

Monetary Authority Ruan Guoheng: Banks have applied compliant technology to approve credit but face budget constraints

HSBC Fintech Artificial Intelligence