"I'm sorry, I didn't quite catch that."

"Could you please rephrase your question for me?"

"Our human agent is currently busy. Please try again later."

Intelligent customer service is meant to be a cutting-edge tool for enhancing service efficiency and quality. However, in reality, it often leaves consumers grappling with a "dialogue dilemma": either lost in an endless loop of "press 1", "press 2" voice menus or receiving robotic responses that fail to address their concerns. Even when attempting to transfer to a human agent, they are frequently met with the message "the agent is busy."

Intelligent customer service seems to have hit a wall, struggling to achieve true "intelligence" in understanding and engaging with users.

Nevertheless, the advent of generative AI technology has ushered in an unprecedented era for intelligent customer service. Leveraging the unique capabilities of generative AI and the intelligent emergence of large models, intelligent customer service is inching closer to our wildest imaginations.

The question now is, how can generative AI technology be seamlessly integrated into intelligent customer service? And how does it function in specific scenarios once implemented? More importantly, does this shift offer tangible value for businesses and users alike?

A Wide Chasm Between Ideal and Reality

Many companies introduce intelligent customer service with the aim of reducing costs and boosting efficiency. However, in practice, intelligent customer service often only responds mechanically to straightforward queries. When complex situations arise, it begins to "respond randomly" or repeatedly prompts users to "please re-enter" their information.

After numerous attempts, consumers are often forced to wait for a human customer service representative, but the entry point for human intervention is deeply buried, and wait times can stretch to several minutes or longer.

Worse still, some intelligent customer services struggle to accurately comprehend user queries, often responding with irrelevant answers.

For instance, when a user inquires about a "refund process," the intelligent customer service might push a "product usage guide." When a user reports "order anomalies," the system repeatedly requests "identity verification," yet the actual solution is slow in coming.

This "inefficient communication" not only wastes users' time but may also escalate conflicts, leading to a decline in consumer trust in the brand.

Traditional intelligent customer service has consistently faced four major pain points, which not only mar the user experience but also constrain its future development.

Firstly, the "mental retardation" lamented by users stems from the machine's inadequate semantic understanding. When users pose a question, the machine frequently responds with irrelevant answers.

This is because previous generations of intelligent customer service relied on technologies such as keyword matching and BERT models. This mechanism necessitates extensive data annotation, with more annotation leading to better understanding capabilities, but annotation heavily depends on manual labor.

Insufficient manual annotation and training result in inadequate machine understanding, leading to irrelevant responses.

Secondly, the user experience is poor, and there is a notable lack of emotional expression. Users are emotional beings, yet previous generations of intelligent customer service were pre-fabricated. Regardless of the emotion with which a user approaches a query, the machine invariably responds with standardized answers, which can feel quite mechanical. Whether the machine can move the user and resolve their issues entirely depends on the person who designed the Q&A.

Thirdly, complex task processing is extremely rigid. For example, when booking a meeting room, previous generations of intelligent customer service typically used a process canvas. It first solicits information like the meeting room's time and participants, then retrieves the interface for booking the meeting room. It must adhere to the set process step-by-step. When the user's topic veers beyond the set parameters, the intelligent customer service will bluntly inform the user that it cannot assist, and the proportion of eventual transfers to human customer service is actually quite high.

Fourthly, the training cost is exorbitant. Previous generations of intelligent customer service required the establishment of specialized robot trainers because it needed to exhaustively list business questions and standard answers.

If it involves intricate business knowledge and processes, it is also necessary to organize the knowledge graph. The entire training process is quite complex and usually takes 3 months to half a year to achieve an 80% resolution rate. Configuring such a trainer specifically is quite costly for enterprises.

However, the emergence of large models provides a practical path for the new generation of intelligent customer service to address these pain points.

From a developmental perspective, intelligent customer service has traversed three primary stages. The first stage is the rule system and expert system, which responds by matching user intentions with keywords and is primarily used for online robots. This stage relies on a plethora of manually set rules to achieve basic functionalities.

Later, in the era of deep learning, with the maturation of technologies such as ASR (Automatic Speech Recognition), TTS (Text To Speech), and NLP (Natural Language Processing), intelligent customer service could not only serve the online environment but also expand to answering robots in telephone scenarios, transitioning from rule-driven to data-driven.

Currently, we are in the era of large models. With the introduction of large-scale pre-trained models boasting stronger generalization capabilities, understanding, and anthropomorphism, intelligent customer service has become more intelligent and can provide more precise and humanized responses when confronted with complex expressions.

Large models can be cold-started without labeled data. They only need to maintain the customer's business knowledge to achieve commendable results, offering more humanized responses, the ability to connect with users' topics, and an understanding of users' emotions, among other benefits.

For instance, when a user initiates a conversation and sounds agitated, the new generation of large model intelligent customer service will first soothe them and then enumerate the corresponding solutions. Since it can generate corresponding responses based on the user's emotional shifts, the user's conversation will flow smoothly and emotionally satisfy.

For process-related tasks, like booking a meeting room, large models do not necessitate fixed process settings or exhaustive topics.

Due to its robust reasoning ability, when the user provides it with corresponding prompt words, indicating what information is needed to book a meeting room, the large model can contemplate what information it has already acquired and what is still lacking, thereby guiding the user to furnish the complete required information.

Even if the user changes topics midway, the large model can seamlessly continue the conversation.

In terms of training costs, thanks to large models, there is no longer a need for specialized machine trainers, which reduces enterprise costs while further enhancing the accuracy of responses.

A New Wave of Transformation in AI Customer Service

After years of evolution, the AI customer service market has attracted various players, and industry competition has intensified. However, fierce competition often signals the dawn of the next wave of transformation, which also poses new challenges for leading AI customer service vendors.

AI capability is the linchpin of intelligent customer service advancement. How to make machine responses more accurate, akin to real humans, and deeply embedded in enterprise operations remains a universal issue confronting the industry.

In essence, AI capability transcends the technical proficiency of AI algorithms; it encompasses how AI technology is implemented in the industry and performs optimally in commercial scenarios, which is a more grueling test of vendors' capabilities.

For intelligent customer service in diverse industry fields, the questions to be addressed differ, necessitating intelligent customer service products to possess industry-specific knowledge bases to comprehend and respond to user queries across different industries.

With the advancement of various technologies, intelligent customer service has morphed into an application system that amalgamates multiple technologies, and increasingly complex needs will continue to propel the entire intelligent customer service ecosystem forward.

This sets higher standards for the comprehensive strength of intelligent customer service vendors, which not only demands robust PaaS technical capabilities but also extensive experience in customized development and delivery.

This necessitates AI customer service enterprises to not only forge full-link marketing and service solutions tailored to industry-specific needs, encompassing the entire business lifecycle of enterprises from pre-sales to after-sales, but also to craft bespoke intelligent customer service solutions for enterprises.

In 2025, generative AI technology is accelerating its global penetration, reshaping the customer service landscape.

Enterprises generally aspire to leverage AI to enhance service efficiency and experience. However, the "last mile" of technology implementation—transforming potent model capabilities into stable, efficient, and easily deployable business solutions—has emerged as a universal challenge.

Issues such as response latency, multi-language support, complex scenario comprehension, customization costs, and compliance in global deployment urgently require robust engineering capabilities to address.

Against this backdrop, Cloud Momentum Data has unveiled the ConnectNow omni-channel intelligent contact center system. This solution offers a suite of "out-of-the-box" intelligent service solutions to tackle the aforementioned industry pain points, aiding enterprises, particularly Chinese enterprises actively expanding into overseas markets, to swiftly establish an efficient and compliant global customer service system.

Since its inception, Cloud Momentum Data has chosen to align with Amazon Web Services (AWS), becoming one of the earliest service providers in China to invest in the AWS ecosystem.

In 2023, Cloud Momentum Data focused on the well-trodden contact center business and officially embarked on the development of ConnectNow. Built upon the cloud contact center service Amazon Connect and the fully managed generative AI platform Amazon Bedrock, Cloud Momentum Data has crafted the omni-channel intelligent contact center product, ConnectNow.

With capabilities such as omni-channel access, agent intelligent assistance, Agentic AI intelligent customer service/intelligent sales, and intelligent quality inspection, ConnectNow enables seamless customer support 24/7, accurately divines customers' deep-seated needs, dynamically generates personalized service strategies, and helps enterprises markedly improve global after-sales service quality and customer satisfaction.

Currently, when enterprises deploy artificial intelligence, particularly large language models (like the technology underpinning ChatGPT), to upgrade customer service, they often confront two major hurdles: integrating technology seamlessly into existing business processes (implementation challenges) and the sometimes sluggish response speed of AI systems (latency issues).

Instead of blindly pursuing the most cutting-edge or largest parameter models, Cloud Momentum Data has opted for a more pragmatic and engineered technical approach to solve these dilemmas. Zhou Lifeng introduced that Cloud Momentum Data deeply integrates the Amazon Bedrock service provided by AWS, enabling the ConnectNow system to automatically select the most suitable "AI brain" to handle diverse customer service tasks with remarkable "intelligence".

For tasks demanding lightning-fast speed, such as real-time translation of user language, it invokes the lightweight and efficient Claude Haiku model.

For tasks requiring profound thought and intricate reasoning, like deciphering the user's convoluted query intent or solving a multi-step problem, it calls upon the more powerful and profound Claude Sonnet model.

This precise matching strategy yields substantial improvements: the system can accurately discern the intent of over 95% of user queries, and in voice conversations, the delay from when the user finishes speaking to when the AI starts responding is compressed to within 2 seconds, approaching or even surpassing the response speed of human customer service.

Furthermore, in the e-commerce realm, voice scenarios are relatively scarce, with more emphasis on image and text displays, necessitating richer presentation formats.

Cloud Momentum Data's AI engine, akin to dismantling Lego blocks, offers over 30 independent functional modules (components). These modules cover a myriad of common scenarios. When users inquire about product details, the AI can not only respond but also directly display related images and text descriptions.

The system can proactively present other question options that users might be concerned about based on the current conversation or frequently asked questions.

Enterprise customers no longer need to construct intricate systems from scratch or be compelled to accept rigid solutions. They can flexibly combine these prefabricated components like picking and splicing blocks according to their unique business needs, swiftly customizing the most suitable customer service process for themselves, significantly lowering the usage threshold and customization cost.

With the assistance of AWS, Cloud Momentum Data has also achieved rapid global business expansion. AWS boasts a globally encompassing infrastructure with 117 Availability Zones across 37 geographic regions, aiding Cloud Momentum Data in swiftly completing overseas service node deployment, rapidly expanding its overseas footprint, and enhancing brand competitiveness.

Currently, the Cloud Momentum Data ConnectNow solution has been listed on the AWS Marketplace, effectively broadening its reach to an international clientele.

Through its deep collaboration with AWS, Cloud Momentum Data has extended its services to leading enterprises across various industries, including manufacturing, new energy, automotive, finance, e-commerce retail, and gaming. This comprehensive approach has significantly enhanced the quality of global after-sales service and customer satisfaction.

In key areas of overseas expansion, such as manufacturing, new energy, and automotive, ConnectNow has demonstrated its efficiency, garnering industry attention. For instance, Deye, whose equipment is sold in 110 countries, faced challenges with traditional customer service due to time differences and language barriers.

After integrating ConnectNow, the system enabled users and dealers to communicate through multiple channels like APP, official website, independent stations, email, WhatsApp, and Facebook, providing 7x24-hour support in over 30 languages. This resulted in a more than 30% improvement in customer service efficiency.

When collaborating with a charging pile enterprise, Cloud Momentum Data leveraged AWS to build a telephone robot customer service system. With ConnectNow, the enterprise achieved automatic recognition of 10 languages, including English, French, and German, and successfully acquired local numbers in 10 countries. The system's speech recognition and fault information entry accuracy surpassed 90%, leading to an over 50% increase in overall efficiency and annual cost savings of 5.5 million yuan. Additionally, the ConnectNow system was integrated with the customer's Salesforce system, enhancing after-sales and customer management integration, further boosting customer operation efficiency.

"AI customer service is not meant to replace humans but to revamp the service chain," said Zhou Lifeng, revealing the core value of intelligent customer service. Technology should not be a mere accessory but should genuinely address efficiency issues and create tangible value in business scenarios, infusing a human touch.

From being derided as "mentally retarded" to embodying a true "human touch," the evolution of intelligent customer service epitomizes the profound integration of technology and business.

With the continuous advancement of generative AI, large model technologies, and engineering capabilities, future customer service scenarios will understand users better, be more efficient and personable, reducing costs and boosting efficiency for enterprises. Each consultation will serve as a bridge connecting users and brands.

This is perhaps the ultimate purpose of AI serving humans: making technology invisible and allowing the experience to flow naturally.