The localisation gap: Why multilingual AI isn’t enough for APAC markets

The conversation surrounding AI voice technology has evolved significantly in recent months. Initially, businesses were primarily concerned with whether AI could engage in natural conversations. However, today's focus has shifted to understanding how effectively voice AI performs across various markets, languages, and customer segments. Companies are no longer asking if voice AI works; they want to know its performance in diverse market environments.

Industry data supports this trend. According to Gartner, 85% of customer service leaders plan to explore or pilot customer-facing conversational AI in 2025, with 44% specifically evaluating voice AI as part of their customer experience strategy. This growing adoption has revealed two key assumptions that businesses often make when deciding to deploy voice AI. Both assumptions require closer examination.

The first misconception is that strong English performance guarantees AI readiness for the APAC market. Many leading voice models demonstrate impressive English capabilities, often portraying smooth and natural conversations. However, real customer interactions in Asia Pacific are far more complex. People frequently switch between languages within a single conversation, depending on the context.

For example, a customer might discuss payment details in Bahasa Indonesia before switching to English to mention a product feature. In Singapore, callers may effortlessly transition between English and Mandarin. Across Thailand, Vietnam, Malaysia, and the Philippines, regional accents, local vocabulary, and conversational habits further increase this complexity.

These scenarios are not isolated cases but rather everyday interactions. Speech recognition technology, while significantly improved, still faces challenges in accurately identifying different languages, maintaining context during linguistic shifts, and interpreting customer intent amidst variations in pronunciation, vocabulary, and sentence structure.

Therefore, strong English performance should be seen as a starting point rather than definitive proof of readiness for every APAC market.

The second misconception is the belief that supporting multiple languages equates to localization. While supporting additional languages is essential, true localization encompasses much more than merely expanding language menus. Customers who speak the same language may still communicate differently. Regional expressions, industry-specific terminology, pronunciation variations, and code-switching can all impact conversation dynamics.

A system that performs well in one APAC market may require further adaptation to deliver the same level of experience in another. Research from Microsoft highlights this point, demonstrating that multilingual users naturally switch between languages during conversations and are more likely to respond positively to voice AI that adapts to these shifts rather than remaining strictly monolingual.

Customer expectations also underscore the importance of localization. A CSA Research study found that 76% of consumers prefer products with information presented in their language, and 88% of Indonesian consumers favor content in Bahasa Indonesia. Although this study focused on digital content, the principle extends to voice interactions as well.

Customers expect seamless communication that feels natural, not simply translated. Ultimately, businesses must prioritize how well voice agents can be adapted for new markets, their performance across multilingual contact centers, integration with existing workflows, and ability to serve customers with diverse communication styles effectively.

This shift reflects the market's maturity, with organizations moving beyond experimentation and focusing on achieving high-quality deployment outcomes.