No matter how good AI gets, it still needs humans to check: exec

To make AI work effectively, companies must establish a clear process and train the system using their data, and more importantly, no matter how advanced AI becomes, it still requires a human to verify its output, said Le Huu Nghia, general director of Le Thanh Construction Ltd.

Nghia made the remarks at a talk themed "Applying AI in SME business management - Where to start” and hosted by the Ho Chi Minh City Business Association (HUBA) last Saturday.

Sharing his company’s experience with AI adoption, Nghia recounted a humorous story that drew laughter from the audience - involving an incident where AI mistakenly “drove away” customers.

According to him, Le Thanh Construction began researching AI two years ago and spent more than a year preparing a budget and assembling a team for pilot testing.

The company eventually developed a system of five major AI platforms, each integrating around 10 smaller AI modules responsible for different processes - from customer consultations and order creation to delivery and post-sale follow-ups. These systems have been applied across a range of industries, including cosmetics, hospitality, and coffee shops.

“In the beginning, our revenue dropped by 50%. We even received complaints from customers due to miscommunications caused by AI misunderstanding local slang, abbreviations, or typos. One time, a customer asked about room prices but made a spelling error. The AI misread it and responded: Our hotel does not offer girl services,” said Nghia.

These misunderstandings caused the system to deliver incorrect responses, which led to issues like delivering wrong products. Eventually, the company had to shut down the entire AI system and start over to recalibrate and retrain it properly.

“If you want AI to be effective, you need a clear workflow and must train it with your company’s specific data. But most importantly, no matter how good AI gets, humans still need to review its work,” Nghia concluded.

Industry experts at the event emphasized that AI is not meant to replace humans but rather to enhance human capabilities. For businesses to adopt AI successfully, it’s essential not only to use robust software but also to have skilled personnel who understand both the technology and business processes well enough to train the AI effectively.

Ha Than, general director of Lac Viet Informatics JSC, advised SMEs to start small with accessible applications like chatbots for customer inquiries, quote generation, and information collection. Once they become familiar with the technology and see its benefits, they can gradually adopt more advanced AI solutions involving text, video, or big data processing.

Tran Manh Huy, CEO of Rainscales from the central city of Danang, suggested that companies adopt an “AI Plus” mindset - using AI to support, not replace, human labor. He also emphasized the need to develop a Computing Power Index (CPI) to evaluate their technological infrastructure and reduce reliance on external platforms.