AI is changing the way businesses serve their customers. But for AI to truly take part in a conversation, intelligence alone is not enough. AI needs to understand what customers are saying, understand what the business offers, and know how to respond appropriately in each situation.
This is also how Viettel Customer Service (VCX) approaches CXBot – a Generative AI-powered virtual call center assistant developed from VCX’s own experience in contact center operations.
Join Mr. Do Duy Son – CXBot Product Director – to explore how VCX is building AI Customer Service from real operational challenges, so that AI can not only answer questions but also support businesses in every customer conversation.
CXBot is built from real operational challenges
CXBot can interact with customers across both voice and chat channels. On voice, the system can handle both inbound and outbound calls. On chat, CXBot operates across websites, Zalo OA, and enterprise applications. The system supports Vietnamese with three regional accents – Northern, Central, and Southern – as well as English, Khmer, and Lao. Through API integration, CXBot can also retrieve or update data from business systems during conversations.
What makes CXBot distinctive, however, is not only its technology. The product was developed from VCX’s own operational experience. VCX has around 20 years of experience operating contact centers and has directly handled tens of millions of customer interactions. As a result, CXBot did not start as a purely technological solution, but from real needs arising in day-to-day operations. The product was first deployed and tested internally before being commercialized.
Today, CXBot handles more than 2 million calls and interactions every month. Around 70% of conversations are completed without being transferred to a human agent. Vietnamese recognition accuracy reaches up to 92% on voice and 99.9% on chat.
Why was CXBot developed as an independent AI?
The decision to develop CXBot as an independent AI capability comes from a simple reality: businesses already have contact center systems that have been invested in and operated for many years. When they want to add AI, they do not necessarily want to change their entire existing infrastructure.
CXBot is therefore designed to connect with existing systems, allowing businesses to add AI capabilities while continuing to use their current contact center infrastructure.
Another factor is the different pace of technological development. A contact center system may be used for many years, while language models can be updated and improved within just a few months. An independent architecture allows CXBot to upgrade its AI capabilities more flexibly without being fully dependent on the lifecycle of the contact center system.
At the same time, customer experience is increasingly spread across multiple channels. Customers may move from a phone call to Zalo, a website, or an application. AI therefore needs to be present across multiple touchpoints rather than being tied to a single contact center platform.
This approach allows CXBot to adapt to different infrastructure models and business needs, while creating a foundation for AI to become a scalable customer service capability.
How is CXBot different from traditional Chatbots and Callbots?
Traditional Chatbots and Callbots typically require businesses to build conversation trees in advance, including intents, questions, conversation branches, and responses. This approach provides strong control over content, but also makes expanding the conversation flow time-consuming. When customers ask questions outside predefined scenarios, the bot may not be able to handle them. When products or policies change, the conversation scripts also need to be updated.
CXBot takes a different approach through Generative AI. Instead of building every question and answer manually, businesses can provide business documents for the AI to read, understand, and use as a source of information during conversations.
As a result, customers can express the same issue in different ways while the system can still understand the context. When business information changes, business teams can update the documents instead of rebuilding the entire conversation flow. This can reduce the time required to bring a use case into operation from weeks to just days.
However, greater flexibility also creates a higher need for control. While traditional bots mainly face the limitations of predefined scripts, generative AI introduces a different challenge: ensuring that responses are both natural and accurate.
The unique challenges of AI Customer Service
AI can be developed globally, but when it is applied to customer service, each market has its own characteristics. In Vietnam, these differences are reflected directly in the way customers communicate.
Zalo is an important interaction channel for Vietnamese users. AI Customer Service therefore needs to operate on Zalo OA alongside websites, applications, and voice channels.
Language is another important challenge. Vietnamese has three regional accents – Northern, Central, and Southern – with differences in pronunciation and word usage. On chat channels, customers may also write without diacritics, use abbreviations, mix in English, or make typos. These characteristics require systems to be built and tested using real-world interaction data.
For voice calls, the first few seconds are particularly important. As users become increasingly cautious about unfamiliar phone numbers, the bot needs to clearly introduce who it is, explain the purpose of the call, and get straight to the point.
Another factor is the expectation of human support. CXBot takes a transparent approach by introducing itself as a virtual assistant from the beginning, while maintaining the ability to transfer the conversation to a human agent when necessary. According to Mr. Son, a good experience does not necessarily mean that AI must handle every conversation. Instead, AI needs to understand when it can continue and when human support is needed.
Controlling hallucinations: AI needs to know its limits
With Generative AI, hallucination is a risk that cannot be overlooked, especially in customer service, where an incorrect answer about pricing, fees, promotions, or terms and conditions can directly affect customers.
According to Mr. Son, no solution can reduce hallucinations to an absolute zero. What matters is controlling the system so that AI responds based on reliable sources of information and minimizes situations that could have significant impact.
CXBot applies multiple layers of control. First, the system prioritizes knowledge sources provided by the business. When there is not enough information, the bot does not make assumptions. For high-risk information such as pricing, fees, promotions, or contractual terms, the system can use controlled content or retrieve information directly from business systems through APIs.
In addition, CXBot is designed to identify situations where its confidence is insufficient and transfer the conversation to a human agent or record the request for further handling. Conversations are also incorporated into VCX’s quality control process, allowing issues identified in real operations to be used to continuously improve the knowledge base and the system.
From the outside, AI Customer Service can be measured by the number of calls handled, automation rates, or the ability to operate across multiple channels. But behind those numbers lies a bigger question: how can AI understand correctly, respond accurately, and know its own limits?
For CXBot, the answer lies in combining AI technology with real-world customer service expertise. A good AI needs not only the ability to hold natural conversations, but also to operate within boundaries that businesses can control and customers can trust.
Built from real contact center challenges, CXBot is gradually making AI a more natural part of the customer experience – where technology and people work together to create more effective customer conversations.
About CXperts
CXperts is a content series featuring insights from Viettel Customer Service's product leaders, technology experts, and development teams on Customer Experience (CX), omnichannel contact centers, AI, and the emerging trends shaping the future of customer service. Through conversations with the people behind the products, CXperts offers a closer look at the technologies powering modern customer engagement.
Viettel Customer Service is a subsidiary of the Viettel Group, operating in the technology and customer service sector. The company provides a comprehensive ecosystem including services (Contact Center Outsourcing, BPO, Upsale, Customer Experience, Loyalty, Document Processing and Digitization, Warranty & Repair Services) and technology solutions (OmniX, CXBot, KnowX Hub, AI-DMS, vCOC, InsightCI, WorkforceX).
Contact Viettel Customer Service today for solutions tailored to your business!
Contact Information:
Hotline: 1800 9397 / Email: dvkh@viettel.com.vn
Website Viettel Customer Service / Facebook Viettel Customer Service / Zalo Viettel Customer Service
Address: 23rd floor, Vinacomin Building, 3rd Duong Dinh Nghe Street, Yen Hoa Ward, Hanoi.
