How Chinese Fintech Giant Ant Uses AI to Help Banks Hedge Currency Risks
Will AI make investment bankers and traders irrelevant? That's one of the most talked-about questions in finance these days.
In reality, banks are already feeling the impact of AI on their behind-the-scenes plumbing, where payments, cash management, treasury and foreign exchange operations are embracing AI models to cut costs in these notoriously low-margin businesses.
Several big banks, including Citigroup, Barclays and Standard Chartered, have tapped data and analysis from Chinese fintech giant Ant Group to hedge foreign exchange risks for their corporate clients.
A billion Chinese people, both at home and abroad, use Ant’s Alipay, providing a huge amount of spending data. On top of that is data from Alipay+, a digital mobile payment network that Ant subsidiary Ant International has been building. Alipay+ connects more than 40 mobile payment apps from dozens of countries across Asia, Europe and the Middle East. More than 800 million people outside China use the network, which includes 150 million merchants in over 100 countries, mostly for spending while traveling.
That gives Ant International unique insights into swaths of cross-border consumer spending data, from a Chinese tourist’s purchases at a shopping mall in the U.S. to an Indonesian consumer’s e-commerce orders in Singapore. Ant International estimates that its platforms, including Alipay+, processed hundreds of billions of dollars’ worth of foreign exchange transactions last year. (The company said it only has access to anonymized data that removes identity and key demographic details.)
Rather than rely on the large language models used by ChatGPT and other AI chatbots, Ant International uses big data models, Falcon TST, that use Ant’s data and are built on the same transformer architecture that has been the bedrock of large language models. The Falcon models use the company’s historical data, and predict future values rather than next words. Tracking consumer spending can help companies predict trends and hedge exposures.
“It predicts the next number, not the next word, and can be used in serious applications such as financial and economic activities, which large language models can’t handle,” said Kelvin Li, general manager of platform tech at Ant International.
The foreign exchange market is so big—$7.5 trillion in transactions every day—and so hard to predict and hedge that banks are eager to gain any advantage.
Ant International’s models can predict the volume of FX transactions, not the movements in FX rates, which remain hard to predict. Nonetheless, they are useful in hedging foreign exchange risks in e-commerce and travel industries because banks lack data in these sectors. The first external customer for Ant International’s AI forecast model was AirAsia, Malaysia’s leading low-cost air carrier. “I think soon the AI community will see the value of big data models,” Li said.
The banks that have been using Ant International’s models do not run them on their own servers. Rather, they incorporate the prediction data feed generated by the models into their overall FX management platforms.
That’s likely because these models are still in the early stages of development. Li estimates that big data models are currently at a level equivalent to where large language models were in 2019, before the release of ChatGPT in late 2022.
Li said that to entice customers to try these early models, Ant International charges customers based on how much they save on FX hedging when they use output from the models. If customers don’t come out ahead, they don’t pay. Most AI developers charge customers to use their models regardless of the results.
Software’s Grim Future
You know things are bad in an industry when analysts compare it to the decline of newspapers and tobacco companies. That’s what Goldman Sachs did last week amid the nastiest days of a monthslong sell-off in software stocks.
Their point was that investors were selling on the assumption that earnings estimates for software stocks would decline as AI eats into their businesses. The selling would continue until earnings estimates bottomed out.
How much AI will hurt software makers is a great unknown, of course. What’s indisputable is that software stocks were expensive, meaning they were priced for strong growth. Any dent in that outlook was a signal to sell.
At the end of last year, the forward price-earnings multiple of the industry was 35, well above the overall market. Except for a surge in valuations when interest rates fell to zero in 2021–22, that valuation was the highest in 25 years, according to Goldman. Analysts estimated revenue would grow 15% annually for the next two years.
There are two takeaways from the software sell-off. First is that sustained high growth is hard to achieve for any industry. That’s a lesson for investors hoping to buy into one of the big AI initial public offerings expected this year.
The second is that until analysts adopt a more realistic growth rate for software companies, their shares will remain volatile. When it’s hard to figure out what something is worth, dealmakers struggle to make deals, delaying the wave of consolidation that’s likely to start in the coming years.—Ken Brown
New From Our Reporters
