Ant International: numerical AI is the GPT of financial services

Singapore-based company says model accurately predicts trading volumes and reduces unnecessary hedging and risk-premium costs

Ant International predicts that its artificial intelligence (AI) model for foreign exchange (FX) could have as large an impact on financial services as OpenAI’s large language model (LLM) GPT has had in the broader business world.

The Time-Series Transformer (TST) AI FX Model developed by the Singapore-based fintech company focuses on numerical data prediction rather than content generation. Kelvin Li, general manager of the company’s platform tech business unit, called it “another track to AI”, alongside LLMs – the technology underpinning generative AI applications like ChatGPT.

In financial services, most companies are adopting LLMs to help minimise risk, “but these models have not gone to the core of financial service, for example, trading, pricing and transaction processing”, Li said in an interview.

“We believe the TST AI FX Model could be the foundational model for financial services, digital payments and even general economic activities.”

Ant International is a unit of Hangzhou-based Ant Group, the fintech affiliate of Alibaba Group Holding. Alibaba owns the Post.

The TST AI FX model currently forecasts Ant International’s cash flow and FX exposure on an hourly, daily and weekly basis, with more than 90 per cent accuracy, according to the company. This enables more accurate predictions of trading volumes and reduces hedging and risk-premium costs from banks.

The model serves customers including banks, airlines, online travel agents and e-commerce platforms. Ant International said it would soon announce a partnership with a low-cost airline in the region.

In early May, British bank Barclays said it integrated TST into its offerings to help businesses reduce FX-related costs and risks amid global volatility.

The technology could significantly slash banking costs, according to Li. In the company’s experience, it could cut costs for FX in the airline industry by up to 60 per cent, and in liquidity management, it could reduce costs by 30 per cent to 50 per cent, depending on the company’s treasury operation model, he said. “They can reduce the working capital needs by that much,” he added.

Compared with traditional currency-prediction models that track historical patterns, Ant International’s AI model could also account for external factors – such as weather – that affect economic activity, to deliver more accurate forecasts, Li said.

TST was based on nearly 2 billion parameters, and the company expected to increase the parameter count by multiple times this year, Li said.

“I think it’s still in the early stage of a large data model,” he said. “This technology is growing very rapidly, and it’s set to become a major force in the coming years.”

Parameters are the variables present in an AI system during training, which helps establish how data prompts yield the desired output. In general, more parameters allow for a more effective model. GPT-4 is estimated to have 1.8 trillion parameters, compared with 175 billion parameters in its predecessor GPT-3.

Ant International also offers a treasury-management platform, called Whale, which employs blockchain technology to make transactions less costly and time-consuming.

HSBC recently launched a tokenised deposit service after collaborating with Ant International on the Whale platform. The service allows clients to make real-time Hong Kong and US dollar payments and transfer funds between HSBC Hong Kong wallets 24 hours a day.