Summary of Key Points
AI customer service systems were once criticized by users for their rudimentary technology, being described as “artificial idiots” (they would provide irrelevant answers or make it difficult to transfer the call to a human representative). However, companies rely on them to reduce costs. With the advent of large-scale language models (LLMs), the capabilities of AI customer service have significantly improved—AI can now understand context, complete tasks, and even show empathy. This has reinvigorated the industry, with both large corporations and specialized vendors gaining advantages. Nevertheless, AI still has its limitations; complex scenarios still require human intervention, and the business models for AI customer services are still being explored (moving from a subscription-based model to a pay-per-result model). Ultimately, AI is changing the way people work, not replacing them entirely.
The Dilemma Behind User Complaints: Companies Love It for Cost Savings, Users Hate It for Its Inconveniences
Many people have experienced frustration with intelligent customer services: when a package doesn’t arrive, the system simply states “signed for”; the conversation is interrupted before the issue can be fully explained; or when you finally get through to a human representative, you have to repeat your entire problem. In 2024, the number of complaints about intelligent customer services nationwide increased by 56%, with the main issues being irrelevant responses and difficulty in contacting a human representative.
But why do companies continue to use them? Because customer service tasks are highly repetitive—questions about returns, address changes, etc., are asked hundreds of times daily, and the cost of employing human representatives is high (a single representative can earn several thousand per month, while an AI system may only cost tens of thousands of dollars per year). Early AI systems were inefficient (they relied on keyword matching, which failed when users expressed their issues in slightly different ways). Companies are willing to sacrifice user experience to cut costs, creating a contradiction between user complaints and their reliance on AI.
Large-Scale Language Models Boost AI Customer Service: From “Answering Questions” to “Completing Tasks”
With the emergence of LLMs, AI customer services are no longer “artificial idiots.” Engineers no longer need to break down each question into keywords for the system to process; instead, they use a combination of knowledge bases and retrieval algorithms (RAG) that allow AI to search through corporate databases for answers, even handling unfamiliar issues.
What can AI customer services do now? For example, they can instantly respond to standardized questions about orders or shipping progress. They can also perform tasks: Sierra Corporation in the United States provides AI services for security company ADT, helping with payments, scheduling service changes, and purchasing accessories. China Telecom’s GOAT-SLM model can detect users’ anxiety from their tone of voice and respond with empathy (“I understand you’re anxious; I’ll check for you immediately”).
In simple terms, it used to be “you ask a question, I provide an answer”; now it’s “you state your need, I handle it for you.”
Industry Landscape: Large Corporations Expand, Specialized Vendors Fill Niche Markets
The AI customer service market is highly competitive, with two main types of players:
- Large corporations like Alibaba Cloud and Baidu Smart Cloud, which have the advantage of large-scale models, computing power, and corporate client networks. They integrate AI customer services into their cloud computing or LLM solutions without needing to promote separate products.
- Specialized vendors such as Ronglian Qimo and Zhichi Technology, as well as startups like Sierra. The customer service industry is highly fragmented by sector (e-commerce, finance, healthcare, etc.), so large corporations are reluctant to invest in customizing their services for each niche. Specialized vendors focus on specific industries; for instance, Sierra specializes in handling complex after-sales processes and is valued at over $15 billion.
No company has a monopoly; everyone has its own role in this market.
The Limits of AI Customer Services
Despite improvements, AI still cannot handle everything. Complex issues such as refund disputes, complaint resolution, and financial security require human judgment, fact-finding, and balancing corporate and user interests—AI cannot make final decisions.
Regulators have also specified that AI should handle routine tasks, while more complex issues should be referred to humans. Current AI systems primarily serve as assistants; for example, NetEase Cloud’s AI can automatically extract users’ names and questions, freeing human representatives from repetitive work so they can focus on more challenging tasks.
Business Models: Moving from “Function-Based Pricing” to “Result-Based Pricing”
How do AI customer services generate revenue? There are two main models:
- SaaS subscription model: Companies pay a monthly or annual fee to use the service, like Zhichi Technology’s products. This model provides stable cash flow but requires the service to be consistently effective.
- Project-based model: Customized solutions for large clients in finance, government, etc., with higher project costs but longer delivery times.
Interestingly, Sierra uses a “pay-per-result” model: companies pay based on the actual problems solved by AI—for example, if an after-sales issue is successfully resolved and prevents a customer from canceling an order. This model demands higher AI capabilities but is more appealing to companies since they only pay for results.
In terms of technology, most companies do not train their own LLMs (they are too expensive) but use existing models like Baidu Wenxin and Alibaba Tongyi, focusing on optimizing them for specific industry scenarios.
The Future of AI Customer Services: Human-AI Collaboration, Not Replacement
The story of AI customer services illustrates how technology is adopted. It requires balancing user experience and corporate needs. AI will not replace human representatives but will make their work more efficient by handling repetitive tasks, allowing humans to focus on complex issues. This path is still evolving, but it’s clear that AI is quietly changing the way we work, making human-machine collaboration the new norm.