There Is More to AI Than AI Models

We had the opportunity to interview Kaweewut Temphuwapat on his thoughts and perspectives on artificial intelligence (AI), its limitations, and its transformative potentials in the economic industry. 

AI technologies, including chatbots like ChatGPT and Gemini, are changing consumers’ lives, especially in the realms of finance and business. Today, SCBX is prioritizing AI as the main organizational driver and incorporating it in multiple areas of business. Here are some of the real-world applications of AI that are often overlooked.

Key pain points in AI adoption

Temphuwapat mentioned in our interview that many financial institutions, primarily the banking institutions, are bringing AI in to drive growth. I More human labor leads to increased financial expenses and higher tendencies for human errors. 

For example, banks that want to offer more loans want to decrease costs and maximize accessibility for customers. In the past, “In the past, if a customer wanted a loan they would have to go to a bank branch to fill out and complete  paperwork,” explains Temphuwapat.. “Banking officials would have to review all that paperwork and decide  whether or not to approve the loan, and for how much.” 

This places a limitation on banks. If more loans were to be approved by a bank, larger numbers of staff needed to be present, costing more valuable time and resources. AI  can resolve these issues.

AI is not new to SCBX

The average person might see AI as something fresh and exciting. However, this technology is hardly new to Temphuwapat, who has been integrating AI into digital lending processes at SCBX through automation. 

Customers enter their data through the app instead of in person, and this data stored and fed into SCBX’s algorithms to calculate the appropriate loan amount to grant. “This significantly reduces the number of human resources,” Temphuwapat adds.

AI also has a role to play in digital marketing. You need to be sure that you are targeting the right people for the products you’re marketing. If you offer the wrong products to the wrong customer groups, it  takes a toll on customer satisfaction and your overheads.

SCBX also developed their own large language model (LLM), Typhoon, about a year ago. 

“We developed Typhoon because the LLMS on the market were costly and ineffective,”  says Temphuwapat. Today, Typhoon is considered the largest Thai LLM with its estimated efficiency equating to GPT3.5 and GPT-4 in Thai. 

Typhoon can be used to interpret and answer customer queries submitted via social media on behalf of some of SCBX’s subsidiaries. The LLM has helped lower costs by 30% and has helped SCBX’s call centers use their time on more complex customer issues, instead of wasting time repeating FAQs. 

Additionally, Temphuwapat hinted at future upgrades from the chat model to newer models with more advanced features like Automatic Speech Recognition (ASR), that can analyze call center recordings to see if the customer support team is relaying the messaging that management wants them to. 

A smaller, faster, and cheaper AI model

In Temphuwapat’s view, more ‘basic’ uses of AI, including processing of large amounts of data, should be prioritized first. There are already several solutions that are market-ready, including automation. 

Meanwhile, generative AI should be developed and implemented as something capable of responding to user requirements.  

“There’s no need for LLMs that might slow you down,” says Temphuwapat. “Some tasks don’t require an omniscient AI. A call center only needs a model that answers some questions.”

Meta’s latest LLM, Llama 3.1, is another prime example of a great (but very large) model with no clear purpose. “I think we’ll see AI models start to shrink in size, be more user-centric, and operate at an even greater speed when compared to larger models,” says Temphuwapat.

In the bigger picture, however, Temphuwapat predicts that businesses will have a greater choice of AI models that best suit their industries, including investment, or for customer services.

However, Temphuwapat emphasizes that any use of AI technologies must comply with the rules and regulations of governing bodies. 

For data-sensitive industries like the financial sector, any minor issue can be devastating. That’s why the Bank of Thailand prioritizes requirements for data security, PDPA compliance, and other data protection issues. If stakeholders in the banking and finance sector want to introduce AI into operations, the process will have to go through a regulatory sandbox first.

The most important thing is data

In Temphuwapat’s opinion, banks already have a vision for the future of lending, including integrating automated processes, and better credit modeling. However, the most important area for improvement beyond AI models  is data.  

“What I’m worried about is that people are so fixated on what it takes for AI to work,” says Temphuwapat. 

For an AI model to create anything new, it requires  high-quality data from multiple sources. This means you’ll have to find partners to provide that data, which is another problem that isn’t rooted in AI tech. 

AI models cannot be the main focus, but it's about finding new data sources, which Temphuwapat believes Thailand should start being more open to. A business can use its own data to train AI, and the more data that a business provides, the smarter a model can be. 

Temphuwapat leaves us with something to consider. “The quality and use of the data must meet business needs,” he says. “But so many companies have incredibly low data utilization, or haven’t even digitized their data. It’s a shame, because there’s nothing more important than high quality data.”  

Catch Mr Kaweewut at the upcoming AI Panel Event, held by Seven Peaks Software, on Wednesday, September 21st, 2024.