Author | Sang Mingqiang

The marketing of smart customer service products and services has always been a headache. As a functional product, traditional smart customer service is mainly based on preset rules and knowledge bases to answer questions. Although this method is efficient in dealing with common or standard questions, if the "keywords" fail to match, it inevitably leads to jokes about "smart customer service not being smart."

Especially for consumer-facing brands, customer service plays a crucial role in pre-sales, sales, and after-sales processes. A good customer service product can not only enhance brand impression but also improve repurchase rates. However, traditional smart service systems are mainly based on general answers and cannot provide personalized services tailored to users' specific needs.

That was until the emergence of large models.

According to the latest data released by IDC, a well-known consulting firm, the overall market size of smart customer service solutions reached 3.08 billion yuan last year, an increase of nearly 36.9% from 2022. In other words, with the iterative upgrades of large model technology, many companies have begun to use customer service as an important scenario to explore the implementation of large models, ushering in a transformational moment for the smart customer service market.

'By processing large-scale data and requests through cloud technology, AI large models can achieve rapid response and multi-turn dialogue while improving the overall intelligence level of customer service teams to benefit other business segments,' said Zhang Shuangying, the Product Director of Smart Customer Service at Lingyang. In his view, while the smart customer service scenario may seem insignificant, it is not easy to do well.

Taking a leading home appliance brand as an example, its self-developed office software needs to cater to 100,000 employees across R&D, production, supply chain, sales, and service departments. These employees generate over 5,000 questions daily, involving more than 20 departmental areas of knowledge, requiring over 70 customer service operators. However, these questions have a 40% repetition rate, and the operators' answers are inconsistent, with cross-departmental support for professional knowledge being challenging.

In other words, a truly market-ready smart customer service product not only requires extreme accuracy but must also evolve with business development while ensuring data security. This means that vendors must not only have deep experience in smart customer service scenarios but also fully understand and analyze customer needs.

Taking Lingyang's newly launched Quick Service 2.0 smart customer service as an example, it relies on advanced large models, small business models, and Lingyang's decade-long experience in Alibaba's customer service scenarios. Building upon its 1.0 version, Quick Service 2.0 comprehensively enhances three core functions: AI Q&A, AI assistance, and AI knowledge base. Compared to similar products, it is the first AI Agent-based smart customer service product in the industry that covers all customer service scenarios and the first to receive certification from the China Academy of Information and Communications Technology's (CAICT) "Digital Native Applications Based on Large Model Smart Customer Service" standard.

This brings us to the choice of technical approach. Currently, there are two mainstream ideas for using large models to build smart customer service systems. The first is the RAG approach, which involves building domain knowledge into a vector database. When users interact with the question system, relevant domain knowledge is retrieved and provided to the large model. The second is the Fine-Tuning approach, which involves updating the large model itself with new knowledge, requiring model tuning, redeployment, and even retraining with GPUs.

Both approaches have their advantages, as Zhang Shuangying told Xinmou, "In practical applications, companies can choose either path. However, regardless of the chosen path, the first problem smart customer service must solve is the illusion problem." Taking Quick Service as an example, its accuracy rate in practical applications is as high as 93%, which is almost comparable to an experienced human customer service representative who is well-versed in all aspects of the company's business and products.

Crucially, this accuracy rate is based on a comprehensive analysis of many discrete and unstructured data sources, which is a core competency of Lingyang that other players lack.

As is well known, Lingyang, which originated from Alibaba's data mid-office team, has deep experience in data services. Its product matrix, built around data elements, has a good reputation among medium and large enterprises across various industries. Following this logic, the emergence of Quick Service within Lingyang, from the underlying data platform to upper-level application implementation, is natural and logical.

However, just as a coin has two sides, while large models are transforming various industries, new pitfalls have emerged. The most typical example is that as the hundred-model competition comes to an end, more and more people are focusing on application implementation, often leading to a situation where companies are looking for problems to solve with their large models, rather than the other way around. This approach can waste resources and lead to unrealistic solutions.

Smart customer service in the era of large models is no exception.

At present, smart customer service is nothing new. To gain customer recognition, it ultimately comes down to value. Especially for smart customer service with high scenario complexity, collaborating with customers is often the best option.

Zhang Shuangying admitted, 'Although we have a deep moat in data services, the refinement of Quick Service has also required a significant investment of time and manpower, which is closely related to the complexity of specific customer scenarios.' For some larger customers, it typically takes 1-3 months from project initiation to product launch.

Interestingly, during the actual implementation of Quick Service, Zhang Shuangying observed an intriguing transformation.

Taking SAIC as an example, as sales and dealership numbers surged, higher demands were placed on the intelligence, response speed, and customer acquisition capabilities of the customer service center. Lingyang's solution was the 'three comprehensives': comprehensive scenarios, covering everything from pre-sales lead generation and test drives to after-sales consultations and trade-ins; comprehensive touchpoints, integrating scattered customer acquisition channels through unified access to apps, websites, and mini-programs, as well as IM upgrades; and comprehensive intelligence, upgrading pre-sales/after-sales robots and human customer service capabilities to improve efficiency.

The results were immediate. According to Zhang Shuangying, Quick Service, based on a full-channel, full-scenario access intelligent service system, effectively improved knowledge production and service efficiency. On the one hand, it provided more confident answers to high-frequency questions such as store marketing and in-car device usage. On the other hand, pre-sales robots could assist brands in tasks such as follow-up calls for maintenance and lead information collection, contributing to a shift in value towards profit centers.

In other words, Quick Service, empowered by large models, has transcended the traditional customer service scope, significantly enhancing its understanding and interaction capabilities and even driving business growth for enterprises. To some extent, this transformation from passive to active service and from cost center to profit center was unimaginable for a customer service robot in the past but has now become a reality.

Returning to the initial discussion, if aligning with overseas practices to build domestic large models is a thing of the past, the key challenge facing everyone amid the wave of industrial digitization is how to truly bring new technologies out of the lab and into various industries. In this regard, the birth and iteration of Quick Service and the Lingyang behind it bring new ideas and directions to the industry.