Why Most Thai Companies Are Failing at AI: And How to Fix It

 

Roughly two-thirds of corporate AI initiatives never make it beyond pilot projects, trapped in what industry insiders call "pilot hell".

 

For Thai businesses pouring resources into generative AI, this failure rate should be alarming. The problem, according to executives who work with thousands of companies globally, isn't the technology—it's how organisations approach deployment.

 

"If you don't put the right focus early on and you pick a use case but you don't have a plan to scale into production, you may be wasting a lot of time," said Sri Elaprolu, director of AWS's Generative AI Innovation Centre, in an exclusive interview with The Nation on November 4–5 during a visit to the Seattle headquarters.

 

His team claims a 65% production deployment rate—nearly double the industry norm—by requiring companies to follow a disciplined methodology before writing code. Whether this approach transfers to resource-constrained Thai SMEs remains to be seen.

 

 

 

Three Critical Errors

Through work with over a thousand customers, AWS has identified three mistakes that doom AI projects:

         

          Data Silos

"The number one mistake is that they only put some of their data into their data lake and leave other data in silos—data islands that sit outside," said Mai-Lan Tomsen Bukovec, AWS's Vice President of Technology for Storage, in a separate interview.

 

The consequence is that AI models trained on incomplete data deliver unreliable results, undermining confidence. For Thai companies, this often means critical business data remains locked in legacy systems whilst new AI initiatives operate in isolation.
 

 

          Wrong Use Cases

Many organisations begin with low-value experiments rather than high-impact business problems. Starting with use cases that deliver measurable outcomes and defining success criteria upfront matters more than the technology itself.

 

          No Scaling Plan

Perhaps most damaging is treating AI as an R&D exercise.

 

"We've seen companies do really interesting work in the lab, but they forgot how to scale into the real world," Elaprolu said. "Not enough resourcing, not enough planning."

 

 

 

The Productivity Multiplier—When It Works

When companies get implementation right, results can be dramatic. Deepak Singh, AWS's vice president of Developer Agents and Experiences, said 80% of developers inside Amazon now use AI tools to write software, with one senior engineer claiming 4.91 times higher productivity.

 

Real-world examples include:

• German AI startup nnamu: 46% of code is written by AI, 30% development time savings
• Deloitte: 70% reduction in testing time for a project, 30% increase in development speed generally
• Italian tyre maker Pirelli: 30% cost reduction through virtual testing
• Ireland Bank: Lawyers save one full day weekly on document review
• Rocket Close: Document processing reduced from four hours to two minutes

 

The pattern: companies that properly implement AI don't just improve incrementally—they multiply effectiveness by factors of three to five. But these are large enterprises with substantial resources, not typical mid-sized Thai firms.

 

 

Thailand's Patchy Progress

Thailand has seen encouraging early adoption. Chiang Mai University deployed AI serving 52,000 users. Bangkok Flight Services migrated to cloud infrastructure, enabling AI development previously constrained by storage limits.
 

 

Yet significant gaps remain. Foundation models offer basic Thai language support, but Elaprolu acknowledged "they're not the most perfect Thai translators." Companies requiring sophisticated Thai language AI must invest in model customisation—demanding resources many Thai SMEs lack.

 

The regional comparison is stark. Singapore's government worked with AWS to develop Sealion, customised for Singaporean English and local norms, plus a Legal GPT incorporating all Singapore laws. Thailand has no comparable national initiative.

 

"We realise local languages and cultural norms are critical, crucially important," Elaprolu said, noting teams across Asia work on model tailoring. "It is absolutely possible"—though he didn't explain why Thailand hasn't pursued this at scale.

 

 

 

The Coming Wave

Executives emphasised businesses shouldn't wait to perfect generative AI before considering agentic AI, where systems take autonomous actions rather than answering questions.

 

"According to Gartner, 15% of the work we do today will be automated through agents by 2028," Elaprolu said.

 

Models are doubling in performance every seven months, creating a dilemma: invest now in approaches that may soon be obsolete, or wait and fall behind competitors. Iberia Airlines already deploys over 10 AI agents across all operations.

 

AWS VP and Chief Evangelist Jeff Barr highlighted the efficiency gains from agentic AI, noting: "The agents don't take holidays and they don't take coffee breaks. You can start building several different research projects or solutions in parallel."

 

 

 

The Skills Paradox

The skills gap isn't what most expect. Barr countered the fear that automation would lead to developer job losses. 

 

"We actually see that historically, when there's a new technology change like this, it ends up needing more people rather than less," Barr explained. "In almost every other historical situation I've seen... every time we've said here's something that makes development more accessible, it ultimately results in more people coming into the field and more productivity."

 

He noted that customers are not seeking smaller teams, but teams with amplified capabilities. 

 

"We now have the power and the resources to do more things with the teams that we have," he quoted customers as saying.

 

As the need for coordination and diverse separate skills shrinks, teams can move into new business areas they previously lacked the resources to pursue.

 

Rather than mastering programming languages, workers need to learn to collaborate with AI systems—a fundamentally different skill from traditional coding.

 

"The better you are at instructing an agent, the better your agent will be," Singh explained. The most valuable skill isn't coding syntax but breaking down complex business problems into clear requirements.

 

This represents a profound shift from writing code to writing specifications.

 

"It's less important to become an expert in Java, but understanding how software works, learning how to write prompts that create good requirements—that's going to make you effective," Singh said.

 

"You have to give examples, good examples, to AI. You have to correct the AI operation," Bukovec explained, emphasizing the importance of learning to work alongside AI tools rather than simply letting them complete tasks.

 

For workers facing this technological shift, the advice was stark: "It's better to get on board and start upskilling so that we're prepared for tomorrow and not get caught without preparation when automation comes," Elaprolu said.

 

 

The Resource Question

For Thai SMEs without large technical teams, AWS suggests contacting account teams for architectural planning or co-development support. The company operates upskilling programmes in Thailand, though specific details weren't provided. The Bangkok region provides data sovereignty for Thai citizen data.

 

But this raises questions about accessibility. If successful AI deployment requires methodology consultants, solutions architects, and innovation centre partnerships, how realistic is adoption for mid-market Thai companies already stretched thin?

 

The underlying message was clear: technology exists, methodology is proven, support infrastructure is available. What matters is execution—and resources.

 

For Thai businesses, the question isn't whether AI will transform their industries, but whether they'll be amongst the 65% who successfully deploy it, or the remainder stuck in pilot projects that never deliver value.

 

The difference comes down to working backwards from business outcomes, consolidating data, testing in real conditions, planning for scale, and allocating proper resources.

 

These are simple principles that separate success from expensive failure—if companies can afford to implement them properly.