Community Blog How Alipay Has Been Providing 24/7 Services since Spring Festival

How Alipay Has Been Providing 24/7 Services since Spring Festival

This article provides a detailed description of the work of Ant Financial's intelligent customer service technical team.

In order to win this inevitable battle and fight against COVID-19, we must work together and share our experiences around the world. Join us in the fight against the outbreak through the Global MediXchange for Combating COVID-19 (GMCC) program. Apply now at https://covid-19.alibabacloud.com/

By Alipay Technology Team.


"Hello, I am a customer service representative at the Alipay Anti-Fraud Center. Our system detected that the Alipay transaction you just made has a risk of fraud. Please check that this transaction is correct."

If you lodged a complaint for fraud transactions on Alipay recently, you may have come across the previous message, which was generated by Alipay's (Ant Financial) intelligent chatbot.

Over the past Chinese New Year holiday, AI-enabled intelligent voice robots have quickly stepped up to make up for the lack of manpower due to the recent coronavirus outbreak. Travel restrictions and office closures made it almost impossible for front-desk staff to conduct their day-to-day work, and many were forced to take extended breaks to ease the pressure on heavily affected cities. During this difficult time, these intelligent chatbots became the driving force in Alipay's communication with customers.

Each day, these robots were able to cover 10% to 60% of calls that were previously handled manually and quickly went on to perform many other tasks. What's more, they were providing personalized messages to customers, such as:

"Hello, I'm Xiao Bao, a customer service representative. The 1-year coverage period of the medical insurance you bought on Alipay has expired"

"Hello, I am a customer service robot at the Alipay customer service center. Due to the extension of the Chinese New Year holiday, the date for money market fund transactions has been postponed to February 3."

Based on follow-up survey data, most users did not even realize that they were talking to a robot. They reported that the communication was smooth and over 95% of them gave a positive review.

So, how did Alipay accomplish this? This article provides a detailed description of the work of Ant Financial's intelligent customer service technical team.

In fact, many enterprises and institutions have had to deal with a serious shortage of manpower because large numbers of employees were unable to return to work due to the novel coronavirus epidemic. While serving various businesses inside Ant Financial, the smart customer service technical team is actively providing its technical capabilities to external customers and working with professionals from various fields to train personalized intelligent robots for specific application scenarios.


Intelligent Speech Interaction Scenarios

Intelligent speech robots evolved from Alipay's intelligent customer service hotline technology. During previous Taobao and Alipay promotional activities, Alipay's customer service had to handle hundreds of thousands of hotline calls. Our customer service personnel was completely exhausted. During the activities, the customer service representative who handled the highest number of customer issues was given the "Golden Bladder" award.

Our personnel had to spend a lot of time clarifying problems and instructing customers with simple problems, while more customers were waiting for service in queue.

Obviously, as our business continues to grow, AI-enabled customer service has become an absolute necessity. Therefore, Alipay's technical team developed an AI speech robot, which was able to receive and clarify customer problems just like a human. Now, nearly 50% of simple problems are solved by robots. This saves the valuable time of our customer service personnel so that they can focus more on complex problems and deal with emotional customers.

Many other industries also urgently need the help of speech robots, including banking, retail, and security.

In banking, insurance, and other financial industries, industry regulations require that, after users purchase certain products, the company must conduct a dialog-based follow-up with the user to ensure that the user receives the necessary information about financial products.

To give another example, Alipay QR code-based payment has subverted the Chinese retail industry, and small businesses that use payment codes need to promptly be informed of relevant Alipay products and activities. This allows them to better cope with the new challenges of new retail.

Our security department blocks tens of thousands of fraudulent transactions every day to keep the funds of Alipay members safe. To do this, we need to understand these potentially fraudulent transactions and explain to users why their payments failed. Our "wake-up" customer service hotline has even become something of a viral sensation.

Our customer service department goes a step further and identifies possible difficulties in user operation procedures in a timely manner. This allows us to provide proactive customer service so that many users do not even need to call our service hotline, which gives our users a smoother experience.

Facing these requirements, we have turned to the intelligent speech robot technology developed in the intelligent customer service field. Alipay has cooperated with online merchants, insurance companies, businesses, security agencies, and other organizations to establish and improve the outbound calling platform. Our smart AI customer service representative now serves as insurance and financial account managers who patiently explain financial products to customers, marketing personnel who promptly notify users of the latest activities and help to ensure our activities have the desired results, and wallet managers and personal assistants who provide personalized services to users.

Innovating under Pressure


Chinese New Year in 2020 was anything but normal. As our customer service personnel could not work from the central office, our call robots had to fill in. During the holiday, Alipay had to quickly change how many of its departments used to operate, including the alerting department, security department, and wake-up hotline. Our daily volume of outbound calls increased by nearly 300%.

At the same time, in the face of the epidemic, Alipay actively cooperated with external partners, quickly developed robots for epidemic prevention and information services, and took a stand at the frontlines of epidemic prevention through big data. Leveraging the natural and smooth interaction and real-time data processing capabilities of telephone communication, together with big data analysis at the backend, we were able to quickly draw epidemic situation maps and promptly identify potential outbreak points. In this way, we provided more accurate support for epidemic prevention and control while ensuring that valuable human resources could be used where they were most needed. In addition to reducing personal contact, we saw our mission as providing quick and accurate information about epidemic prevention in order to give hope and confidence to the people.

Technical Solutions That Support Intelligent Speech

After more than half a year constructing an outbound calling platform, our team learned a great deal about AI-based robots for hotline customer service and online customer service. By focusing on addressing the special business needs of telephone interaction, we gradually developed a sophisticated intelligent interaction technical solution for outbound calls.


We have made full use of our expertise in algorithms to explore the natural language understanding and conversation status tracking technologies used by intelligent interaction robots in calling scenarios. Ultimately, we developed a set of core algorithm components:

  • Zero-process outbound call dialog: Based on the imitation learning and reinforcement learning techniques, robots can learn the appropriate dialog policies and select the appropriate outbound calling techniques based on historical dialog data. This allows them to naturally and smoothly interact with users to achieve the exact purpose of the call.
  • Intent recognition based on small sample learning: This reduces the complexity of user corpus configuration. In addition, it can be used with technologies such as corpus correction, corpus expansion, and new intent discovery to help the operational staff quickly and accurately complete intent and process configuration.
  • User simulation based on the retrieval and generation models: This component can generate user answers that meet certain user goals and user settings, simulate real users, and automatically evaluate intent configurations, processes, and conversation models.
  • Exclusive algorithm data tagging platform: Operational staff can use the platform to directly tag and train algorithm models and extract "uncovered tags" and "tag errors" from data that runs online. This helps the business operational staff quickly iterate the semantic understanding model.

This closed algorithm feedback loop makes AI interaction technology much easier to deploy. It not only allows the operational staff to use this technology directly, but also provides guidance and feedback on their application of the technology. In this way, they can quickly correct mistakes and errors in model configuration. With minimal investment in operations, customers can deploy comprehensive semantic understanding capabilities for outbound calls.

Cooperation Is the Key to Success


At the same time, we are working closely with multiple business partners, such as network businesses, insurance agents, merchants, and security personnel, to build a complete set of outbound calling business processes, intentions, and other general technical assets.

  • Comprehensive intent accumulation: We have accumulated many general user intent expressions based on the outbound calling requirements of financial businesses such as merchants, insurance companies, and banks. On this basis, some of the resulting capabilities include callback when a user is currently driving, user identity confirmation, and transfer to a human customer service representative for further questions. These capabilities can be reused quickly to build new outbound calling solutions.
  • Outbound calling sub-processes: You can build a knowledge base and use outbound calling sub-processes to resolve some issues that users may have with respect to the purpose of the outbound calls. If it is inconvenient for the user to talk, the user enters a sub-process to schedule a subsequent call. When the user performs sensitive operations, the user first enters the sub-process for ID card verification. This allows the business operation staff to focus on optimizing the interaction processes for main business goals while effectively ensuring a good user experience and improving operational efficiency.

The accumulation of technical assets based on refinement in business applications makes it even easier for new businesses to deploy this solution. When new businesses wish to adopt this solution, they can learn from or directly use existing processes and focus on refining main service interaction processes to quickly achieve their business goals.

At present, we abstract and integrate the main business formats to build a complete outbound calling operation platform and various outbound call solutions to further improve the efficiency of outbound calling service deployment. Currently, new businesses can deploy smart calling interaction capabilities within several hours.

The operation platform supports a drag-and-drop dialog process configuration framework, allowing you to quickly and flexibly deploy this solution for your business.

The scheduling system can support tens of millions of outbound calls per day and predicts the effective call time for users based on user features. It supports real-time outbound calling and scheduled outbound calls to effectively improve user contact efficiency.

The solution is deeply integrated with the communication platform and effectively integrates the notification, SMS, and outbound calling capabilities. It can reduce operating costs while ensuring effective contact. Through channel integration, the system can directly take users to activity details pages to increase user conversion.

This product-based outbound calling solution allows for in-depth cooperation with merchants and insurance companies in order to produce a wide range of solutions, such as activity notifications, activity participation guidance, and intelligent follow-up calls. You can now fill out a form to quickly build an outbound calling robot.

Now, we have combined the aforementioned technologies with the speech recognition and synthesis technologies of the DAMO Academy to produce a robot that is capable of natural and smooth interaction. This robot makes most users believe that they are interacting with a real customer service representative. This robot provides a main business conversion completion rate of over 80%. In a survey of a sample of users, over 95% of them rated the interaction as satisfactory. This solution ensures the completeness of interactions and delivers the desired business results.


The Future of Intelligent Speech Robots

We expect intelligent speech robots to usher in a revolution in telephone services once they can successfully imitate a natural human voice and hold multi-round dialogs with customers.

If this is the case, first, a large number of phone calls will be made by intelligent voice robots, saving a significant amount of time and improving the follow-up call performance. In addition, the robot interaction process will be more standardized than human interaction, incurring less risks and complaints. Robots can take their posts immediately without any training. In addition, AI-enabled speech robots allow for the continuous accumulation of business knowledge and the evolution of interaction patterns. Robots do not have to develop business skills over a long period of practice, and no knowledge can be lost once gained. With the ongoing development of AI technology, reinforcement learning technology can be used to proactively explore more effective communication patterns and quickly formulate interaction policies. Through algorithms and data analysis, this technology can also provide better data support for business understanding and greater insight into the real needs of customers and users.

Second, all daily calls can be automatically recorded. The work of the intelligent speech robots can be used for commonality analysis, tag-based customer relationship management (CRM), and to create multidimensional call recipient profiles. This can improve management efficiency and help generate personalized and comprehensive service solutions.

Third, intelligent speech robots can quickly learn from the experience of human customer service personnel and constantly summarize, analyze, and extract Q&A data from this experience. They can combine information about common issues and products and services in various industries into a powerful knowledge map to ensure a satisfactory customer experience.

Finally, this solution allows for more flexible scheduling and collaboration, including inter-machine collaboration and human-machine collaboration. If a business faces a lack of manpower, it can "hire" AI-empowered intelligent speech robots at any time to seamlessly take over from human employees.

We once rejected customer service robots because they were customer-unfriendly and could not solve customers' problems. In the future, however, intelligent customer service robots will feel as friendly as real people and even provide a better experience than a human can. It can be foreseen that they will be one more example of how technology can change the world.

While continuing to wage war against the worldwide outbreak, Alibaba Cloud will play its part and will do all it can to help others in their battles with the coronavirus. Learn how we can support your business continuity at https://www.alibabacloud.com/campaign/supports-your-business-anytime

0 0 0
Share on

Alibaba Clouder

2,624 posts | 723 followers

You may also like


Alibaba Clouder

2,624 posts | 723 followers

Related Products