Laiye Technology Files IPO Prospectus: AI-Powered Conversational Agents Driving Digital Transformation in Healthcare with Multi-Fold Customer Service Efficiency and Precision User Profiling

Intelligent interaction is an industry that has long been overlooked. The diversification of consumer demand scenarios, coupled with the need for high-quality customer service access and precise service matching, is driving the customer service industry toward intelligent development.

Whether on Taobao, JD.com, or the WeChat platform, the massive traffic behind these platforms reflects users’ expectations for precise decision-making. How customer service representatives provide professional, prompt, and high-quality responses while maintaining user retention and high engagement is crucial to facilitating subsequent transactions and sustaining community vitality.

Nowadays, many companies have concealed channels for human customer service, replacing it with fully automated intelligent systems. However, most of these intelligent feedback mechanisms lack the support of knowledge graphs and merely respond to requests based on simple keyword matching. For issues that require human intervention, these products often fail to promptly determine when to transfer inquiries to human agents, resulting in a poor consumer experience.

Managers are equally frustrated. While everyone in the internet era recognizes the value of data, the sheer volume and lack of structure often deter them, making it difficult to find an appropriate approach for categorizing and processing information.

Intelligent interaction can address this issue, and in the future, “human-machine collaboration” will undoubtedly become the mainstream mode of production and service. Consequently, many technology companies are engaging in this field, striving to enhance human-computer interaction efficiency and endow computers with cognitive capabilities.

Among numerous intelligent interaction enterprises, the benchmark company Beijing Laiye Network Technology Co., Ltd. (hereinafter referred to as "Laiye") is attempting to leverage artificial intelligence technology to propose an intelligent conversational operations solution, aiming to address customer service issues in e-commerce through human-machine collaboration. Meanwhile, Laiye is also utilizing knowledge graphs to organize relevant data, thereby uncovering potential value for its clients.

How Can Customer Service Companies Build Competitive Moats? What Developments Lie Ahead for Human-Machine Collaboration in the Era of Intelligent Interaction? With These Questions in Mind, VCBeat Conducted an Exclusive Interview with Laiye.

Laiye was founded in 2015 by a distinguished group of co-founders, including PhD graduates returning from Ivy League universities and MBA holders from the Massachusetts Institute of Technology (MIT). The company is committed to becoming a globally influential intelligent robotics enterprise in the era of human-machine symbiosis. Today, Laiye possesses core technologies such as deep learning, reinforcement learning, natural language processing (NLP), personalized recommendation, and multi-turn multimodal interaction, striving to leverage cutting-edge technology to penetrate both commercial and consumer sectors.

Wang Guanchun, Co-founder and CEO of Laiye, previously served as the project lead for Baidu’s Intelligent Interaction Team, Xiaodu Robot, and Baidu Kuaishou, among other initiatives. These experiences have given Wang unique insights into intelligent interaction technologies. After leaving Baidu, he began exploring how to integrate interaction technologies with the increasingly digitalized service industry, aiming to build a human-machine collaborative robot platform.

During the R&D process, Wang Guanchun discovered that pure knowledge graphs could not fully address question-answering scenarios. In contrast, the combination of AI and HI (Human Intelligence) proved to be a more practical approach. Therefore, Laiye leveraged databases to create conversational data for specialized domains and establish knowledge points. When users pose questions to Laiye’s bots, the system decomposes the queries based on these knowledge points and performs preprocessing on the user inputs.

Wang Guanchun, CEO of Laiye, told VCBeat: “Upon receiving a query, Laiye’s bot will send feasible answer options to customer service representatives, who then confirm the final response. This approach has reduced the time customer service representatives spend thinking about and drafting responses from over a minute to just ten-odd seconds—transforming what was once a fill-in-the-blank task into a multiple-choice selection. This shift will drive an overall improvement in the quality of the entire customer service system.”

When it comes to the challenges that the intelligent interaction industry needs to address, they are nothing more than user acquisition, retention, engagement stimulation, and conversion. These tasks require robust data mining and processing capabilities. Laiye has established strong technical barriers in these areas, enabling a perfect integration of business and technology.

Today, Laiye’s Wulai conversational bot platform serves industries such as maternal and infant care, consumer retail, and telecommunications, with nearly 100 large enterprise clients including China Mobile and Meituan. In the healthcare sector, Laiye has built a top-tier customer service platform for Wyeth, addressing its challenges in managing WeChat-based customer relationships.

The maternal and infant sector was the first pan-health domain that Laiye entered, and the Wyeth project was its inaugural initiative targeting the B2B market. This case was honored as one of the “Top 30 Global AI Application Cases” by Synced in 2018. Over the past year, Laiye has accumulated extensive experience and data through its collaboration with Wyeth, and its knowledge graph has become increasingly mature through continuous practical application.

Maternal and infant consumers exhibit prolonged and highly cautious decision-making cycles in the healthcare sector. The maternal and infant context is relatively complex, with distinct issues arising across different scenarios. Unlike other industries that focus more on simple, judgment-based queries, conversations in the maternal and infant domain typically involve complex, multi-turn dialogues. This requires customer service representatives to guide consumers, progressively gathering relevant information before reaching a conclusion.

In contrast, Laiye also proposed a human-AI collaboration solution for Wyeth. Specifically, artificial intelligence first performs semantic analysis and classification of questions raised by mothers, then routes them to the appropriate customer service representatives. Meanwhile, based on a knowledge graph, the AI generates a set of suggested answers for the representatives. Instead of typing responses manually, customer service agents can simply select the approved answer with a click, enabling rapid replies to consumers.

With the assistance of artificial intelligence, customer service representatives' work has shifted from "fill-in-the-blank" tasks to "multiple-choice" decisions, resulting in a several-fold increase in the speed of handling consumer inquiries.

Consumers have also benefited. In the past, prolonged wait times resulted in poor consultation experiences. Today, faster customer service response rates enable more rapid communication between consumers and businesses, thereby enhancing customer satisfaction. Meanwhile, when consumers need to compare products across different categories, customer service representatives previously had to manually retrieve relevant product information. With AI assistance, information retrieval and product comparison can now be completed in an instant. Additionally, AI can help address questions that human customer service agents were previously unable to answer.

Meanwhile, Wyeth is also grappling with customer management challenges. Marketing and customer acquisition costs are rising, registration rates are low, user churn is frequent, and retention and sales conversion rates remain less than ideal. Attracting sticky maternal and infant users and delivering precise services have become Wyeth’s top priority.

In addressing this issue, Laiye has gradually found solutions through its collaboration with Wyeth.

During customer interactions, the chatbot continuously collects data on consumers’ purchasing habits, spending power, and various needs. As data accumulates, the consumer profiles for Wyeth become increasingly clear. Consequently, the bot tags different consumers and delivers personalized product recommendations aligned with their spending capacity, purchasing habits, and consumption needs.

In this way, Wyeth’s customer acquisition strategy has shifted from traditional, broad-based advertising to highly efficient targeted delivery. This transition enables Wyeth to acquire more loyal customers at a lower cost per conversion.

Compared with knowledge graphs based on electronic health records (EHRs) and disease-specific data aggregations, the knowledge graph developed by the maternal and infant customer service platform more precisely aligns with the lifestyle needs of users and consumers. Moreover, the entire personalized product can be delivered within just a few months. Specifically, the development process comprises the following steps.

I. Mining Historical Corpora to Analyze Key User Needs: During communications with Wyeth, Laiye obtained three sets of conversational data, from which nearly 600,000 dialogues were cleaned and extracted. The majority of these dialogues are typical multi-turn conversations, with an average session length of 11 turns; sessions with fewer than 6 turns account for approximately 17% of the total. This level of complexity far exceeds that of customer service interactions in other industries.

II. Summarize Requirements and Identify Pain Points for Human Customer Service Representatives: By analyzing conversational corpora, Laiye discovered that factual messages responded by customer service representatives are generally lengthy. In addition, customer service representatives frequently use template messages (such as greeting scripts at the start of a conversation, closing scripts at the end of a conversation, and educational scripts). These scripts feature fixed content; if customer service representatives manually input them each time, it creates a bottleneck in response efficiency. This is a typical pain point for customer service representatives using the Wyeth Multi-Customer Service System. Therefore, the specific goal of the Laiye Q&A System is to prioritize coverage of these high-frequency response scripts.

III. Establishing a Knowledge Base: Upon completion of the aforementioned analysis, Laiye commenced the targeted development of its knowledge base. By comprehensively employing data processing techniques such as hierarchical clustering, classification, and domain-specific keyword mining, combined with manual review by AI trainers and multiple iterations, the final knowledge base comprises over 1,500 knowledge points and more than 20,000 questions.

IV. Building a Q&A Bot: Laiye has developed a “retrieval + ranking”-based Q&A bot solution, leveraging cleaned dialogue corpora and a knowledge base reviewed by trainers. Specifically, the dialogue data and knowledge points are imported into the ElasticSearch retrieval system. Upon receiving a user message, the system searches for relevant knowledge points or historical conversation snippets from the retrieval engine. It then employs a reranking algorithm to refine the search results, ensuring that the most relevant knowledge points or historical conversation snippets are prioritized. Finally, the top 6 results are displayed on the multi-customer service interface for customer service representatives to select.

5. Provide input suggestion features based on business scenarios: Drawing on the experience of using Laiye’s internal systems and the usage scenarios of customer service representatives, Laiye believes that enabling customer service representatives to retrieve complete scripts via keywords, or having the system automatically retrieve complete scripts based on their current input, would significantly improve response efficiency and ensure a better user experience. Therefore, Laiye has developed an input suggestion feature for use by customer service representatives.

6. Develop a BI System: The BI system provided by Laiye supports custom keywords. The system automatically monitors messages containing these keywords, tracks the number of conversations in which they are mentioned, and analyzes recent trends. This enables the identification of meaningful patterns, such as the frequency of mentions of “cold” during winter or user interest levels in different infant formula product lines.

The construction of the entire knowledge graph varies in duration from one to six months, depending on the scale of the project. Rapid delivery enables enterprises to swiftly transition from traditional management models to AI-empowered digital models. For Laiye, this accelerated project implementation highlights its capability to productize artificial intelligence solutions.

The success in the maternal and infant sector has laid a solid foundation for Laiye’s expansion into the health industry. Meanwhile, companies like Wyeth are widely present in the medical consumer and healthcare sectors.

Due to the specialized nature of medical knowledge, consumer-facing (C-end) users’ engagement with medical products will involve more intricate, extensive, and professional AI-driven interactive Q&A sessions. Moreover, many projects in the healthcare sector require human-AI collaboration to enhance efficiency.

This presents both opportunities and challenges for Laiye. In its 2019 practices, Laiye established collaborations with major pharmaceutical companies such as AstraZeneca, helping AstraZeneca build compliance robots to reduce operational costs and unlock data value through AI-powered products.