Cover Due to the lack of training data, AI usually underperforms in languages spoken in developing regions and dialects (Photo: Getty Images)
Due to the lack of training data, AI usually underperforms in languages spoken in developing regions and dialects (Photo: Getty Images)
1 / 1
Hong Kong start-up Fano is tackling the language gap that leaves non-English speakers behind in the AI race
If you prompt AI tools to write an English caption for social media, chances are they will do a fairly good job. But when it comes to writing in Vietnamese, the likelihood is you will need to spend time polishing or even rewriting the output entirely. Why can’t AI thrive equally across all languages?
Large language models (LLMs) are trained on vast amounts of data, most of which is in English, the lingua franca of the internet. Other languages, especially those spoken in developing regions, and dialects, lack sufficient high-quality data for training. As a result, AI tools tend to underperform in these languages, widening the digital gap. Regions that do not commonly use English may adopt AI more slowly and with less sophisticated capabilities.
Read more: Inside the one-time treatment that could cure an ‘incurable’ eye disease
Above Claude is a family of large language models created by the US-based company Anthropic (Photo: Getty Images)
Claude is a family of large language models created by the US-based company Anthropic (Photo: Getty Images)
1 / 1
Fano, the first Hong Kong start-up to be backed by tycoon Li Ka-shing’s private investment vehicle Horizons Ventures, is taking a different approach. Specialising in AI solutions to improve customer service for enterprises, it has built an engine that recognises a wide range of Asian languages, including Mandarin, Cantonese, Hokkien, Thai, Vietnamese, Bahasa Malaysia and Bahasa Indonesia. The company reports that its accuracy rate exceeds 90 per cent, even when speakers mix languages within the same sentence.
Grinding the data
Co-founder and CEO Miles Wen, who was recognised as a Gen.T Leader of Tomorrow in 2020, says supporting these languages is a natural business decision because the company is based in Asia and its clients serve communities that use them daily. Covering such Asian languages also aligns with his belief that technology should fill real gaps and deliver societal value, rather than be a means of competing with other developers to maximise profit.
Above Miles Wen, co-founder and CEO of Fano, was recognised as an outstanding emerging future entrepreneur by KPMG China in 2025 (Photo: KPMG China)
Miles Wen, co-founder and CEO of Fano, was recognised as an outstanding emerging future entrepreneur by KPMG China in 2025 (Photo: KPMG China)
1 / 1
Fano requires an enormous amount of data to build its technology, given that pronunciation can vary widely depending on factors such as age, accent and proficiency. Complexity increases further when different social groups create their own slang. It worked with enterprise clients willing to share data for research and also collected firsthand data by interviewing and recording people in various settings, tasks that required time and resources. “It’s just hard work. It’s just something you have to spend time on, and you have to grind it,” he says.
Today, the start-up serves enterprises in high-stakes industries such as banking and financial services, telecommunications and insurance, and is especially popular with companies operating call centres. Its technology supports interaction analysis, extracting insights such as customer sentiment, identifying compliance breaches and detecting fraud within conversations. The company also provides real-time guidance tools to help human agents respond to customers more effectively. Each year, Fano processes more than 100 million customer interactions.
Read more: Fashioning futures: where creativity meets technology and sustainability at Taylor’s University
Humans in loop
Although the company offers fully automated chatbots and voicebots, Wen says most organisations still prefer to use AI as an assistant rather than a replacement for humans. Many “unresolved problems” remain, he explains, which could lead to reputational damage or financial loss, including instances where bots provide inaccurate information due to hallucinations.
Full automation can also result in job cuts, rather than redeployment. “I’m pretty sure we’re going to make some big mistakes down the road when it comes to the push for full automation,” he says. “There might be a kind of pull-back after we learn those lessons, and then gradually, society and technology providers like us will figure out the best balance between AI automation, AI assistance and purely manual work.”
Above Telemarketers and customer service agents are among the 40 jobs Microsoft identified as being at higher risk of being affected by AI (Photo: Getty Images)
Telemarketers and customer service agents are among the 40 jobs Microsoft identified as being at higher risk of being affected by AI (Photo: Getty Images)
1 / 1
Critics have also highlighted the broader downsides of AI adoption, ranging from technological dependency to ethical and privacy concerns. Wen, who has spent more than a decade in the field, agrees that the current trajectory of AI development is worrying. “People are going in for a gold rush,” he says, pointing to the influx of speculative capital chasing quick returns. “We’re almost creating this new intelligence, but we’re not spending enough time and energy figuring out the safety part.”
A 2025 report by the global non-profit organisation Future of Life Institute found that leading AI companies’ safety practices are falling short, citing a lack of concrete safeguards, independent oversight and long-term risk management strategies.
This is why the Fano team prefers to work with large corporations, which tend to be more stringent when it comes to evaluating risk and implementation. Even when certain processes could be automated, Wen notes that they often keep humans in the loop to ensure their employees continue to think critically and retain institutional knowledge.
What matters is whether we’re doing something really good for society. Is this generating legitimate, solid, concrete value?
“There was a huge wave of AI from 2017 to 2019, and it died down from 2020 to 2022 for quite a number of them. Then ChatGPT came out, and suddenly it became hot again,” Wen says. The biggest takeaway of his entrepreneurial journey is that “perseverance pays off”: not only did his team stay on course even when market anticipation was low, but they also continued to improve the technology rather than chasing hype. Ultimately, the CEO believes that the focus should not be on whether or not his product is powered by AI. “What matters is whether we’re doing something really good for society. Is this generating legitimate, solid, concrete value?”
Now, meet more Gen.T Leaders of Tomorrow 2025 from the Technology sector.
NOW READ
Is remote work really a path to freedom, or just leaving workers vulnerable?
This speech app is transforming the way children with autism learn to communicate
While the world watches humanoid robots dance, this founder thinks we’re missing the point
Based in Hong Kong, editor Yoyo Chow covers the people and ideas redefining Asia’s future—from cutting-edge innovation and AI to bold moves in sustainability and diversity. She also drives content for Tatler Gen.T in Hong Kong, a platform and community spotlighting the region’s next generation of startup founders, creatives and changemakers.
Before that, she was a video journalist producing content for international TV and digital platforms, including Reuters and South China Morning Post. If you have a powerful story to share, she’s all ears. Send press materials, event invites and any inquiries to yoyo.chow@tatlerasia.com.
