Home > Columns > CRM Columns

Building a True Multilingual Contact Center with GenAI

How advances in AI and real-time translation are eliminating language barriers, once and for all, in the customer contact center 

Contributed Article by Maik Hummel, Principal AI Evangelist for Parloa   

In the contact center world, resolving a customer service issue requires effective, efficient communication. Agents must know how and when to ask the right questions. They need to serve as caring, empathetic professionals and detectives who can quickly sleuth out and solve a problem.    

There’s nothing more frustrating for a customer than a communication breakdown. Imagine calling a contact center for help, only to realize that the person on the other end of the phone doesn’t understand what you’re trying to say. It’s equally frustrating for the call center agent, who may not have a translator on staff with the corresponding language skills to help bridge the communication gap.    

The good news is that this scenario is swiftly becoming a thing of the past, thanks to significant advancements in genAI technology. As more contact center solutions embed advanced AI technologies, enterprises can finally build a true, multilingual customer contact center.

The Growing Need for Multilingual Support

Most U.S. companies already understand the importance of multilingual support centers. The U.S. doesn’t have an official designated language. While English is the most widely used language, approximately 430 languages are spoken by a diverse population. According to the most recent U.S. Census, more than 25 million people in the United States have limited English proficiency (LEP), about 8% of the population. A LEP customer is someone who does not speak English as their primary language. They may have limited ability to read, speak, write, or understand English. Every LEP customer is different. They may speak English well, not well, or not at all. They may struggle to find or understand words in English. This makes it more essential for contact centers to seamlessly support multiple languages, 24/7.    

However, many contact centers still struggle with properly executing multilingual customer support. For years, the approach to multilingual customer support involved hiring contact center agents fluent in multiple languages and ensuring direct communication with customers in their preferred language. Some contact centers tried to organize dedicated teams of agents or departments for specific language groups to handle calls, chats, or emails. Others turn to third-party interpretation services to support agents while working with a customer.    

Given the sheer number of languages spoken across the U.S. and the volume of calls coming into larger contact centers, these approaches have proven to be unsustainable. There simply aren’t enough skilled contact center agents who speak enough languages to support every language at scale. Additionally, the costs of recruiting, hiring, training, and retaining multilingual agents are relatively high, even when outsourcing these agents from a third party. Moreover, third-party interpretation services can sometimes be ineffective in a call center setting due to several factors, including the lack of visual cues, potential delays in connecting with an interpreter, potential audio quality issues, difficulty managing complex technical jargon, and the need for context specific to the company and its products, which a remote interpreter might not fully grasp, leading to potential misunderstandings and customer frustration.     

Removing Barriers With Advanced AI Technologies

You may have contact center agents on staff who can speak a few languages fluently, which already puts you ahead of the game. And yet, if your contact center supports a large market with thousands of incoming calls, that won’t be much help. For enterprises with contact centers that must scale to operate across multiple regions, it may seem like forced off-shore customer service or hiring a larger team for broad language support are your only options. However, AI and advanced real-time translation can provide an engine to help your agents converse easily in non-native languages.

Real-time translation (RTT) technology has advanced significantly over the last few years due to improvements in artificial intelligence (AI), machine learning (ML), and natural language processing (NLP). Today’s RTT is faster, seamless to the customer, more accurate, and more accessible across different industries, such as e-commerce, healthcare, and financial services. AI-powered RTT systems empower businesses with multilingual contact centers to communicate with international customers in real time with customized vocabularies for industry-specific terms. This is a massive leap beyond outsourcing translation to human agents, which can cause lags in response time.

Moreover, traditional, phrase-based language translation models are evolving thanks to advances in Neural Machine Translation (NMT). This technology uses neural networks to provide more natural, context-aware language translations. To date, Google Translate, Microsoft Translator, DeepL, and other major players have all integrated NMT, leading to more accurate real-time translations.

While NMT excels in literal translation from one language to another, large language models (LLMs) enable the possibility of further refining the translation. LLMs are a type of AI that can generate human-like written responses to user queries. In the context of language translation, we call this “polishing” LLM interpolation.    

LLMs can be instructed to pay attention to industry- or domain-specific wordings, bridge cultural differences (like the different politeness levels in Japanese), and correct potential mistakes from NMT engines.    

In addition to NMT and LLMs, we have seen a significant advancement in the quality and accuracy of Text-to-Speech capabilities. Today’s TTS can give human agents an artificial speaking voice that can talk to the customer in any language. This was not possible in the past because the AI-powered voice sounded robotic and non-authentic. But now, the capability exists to provide an artificial, multilingual voice to human agents that sounds and feels like the “real thing” to the customer.     

Today’s RTT works by using the original transcript from the text-to-speech and the NMT translation and combining these with a dedicated prompt (context) that defines all the rules that we want to have considered by the LLM when proposing an interpolated translation. While this approach increases the latency slightly, it greatly impacts the end-customer experience due to the enhanced translation quality.    

With AI acting as a real-time translator across calls and chat, human agents don’t need to speak a language fluently—or at all. AI can immediately translate an audio or written exchange for the human agent and then back into the customer's preferred language. Thanks to AI, customers may never be aware they are not chatting with someone who speaks their language.     

This capability enhances customer satisfaction, eliminates errors and confusion due to language barriers, and allows for more effective, nuanced communication between customers and live agents. Advanced genAI models can even help agents be more empathetic with their customers to build stronger customer loyalty and deliver a more human experience, regardless of the languages spoken.     

In essence, generative AI has elevated real-time translation into a sophisticated, adaptive tool that allows for more natural, contextually aware, and personalized multilingual communication in the customer contact center. This leap makes RTT more valuable for businesses, travelers, customer service teams, and global collaborators alike. The days of language as a barrier to seamless communication between agents and customers are about to be over.    

About the author

Maik Hummel is the Principal AI Evangelist for Parloa, a leader in AI-powered automation and assistance for customer service.