AI Use Case: More Accurate FX Exposure Forecasting Using AtlasFX

# AI

# Foreign Exchange

AtlasFX’s Scott Bilter says one client’s FX hedging issues offer a great example of how AI can save time, boost accuracy.

In a NeuGroup video you can watch by hitting the play button below, AtlasFX co-founder and CFO Scott Bilter describes how an AI model created by the risk management company produced far more accurate exposure forecasts for a multinational client that needs to hedge currency risk for intercompany (IC) transactions between legal entities that combined have some 700 currency combinations and cross-currency pairs.

The company, he said, had “a lot of cross-currency exposures that were material, driven by these IC flows that nobody could forecast well. They were trying to, and they were spending tons of time doing it, and the forecast deviation was rather high. So that was just a really good use case for an AI model.”

The results were clear. “The AI model was just trouncing the manual one, just time and time again for many exposures,” Mr. Bilter said. This has allowed the treasury team to reduce forecast errors by $3.5 million per month—a 50% improvement—and save more than 300 hours of manual effort each quarter.

“It ended up being a great success story,” Mr. Bilter said. “That’s just one example; there can be a lot of different ways to apply AI to the split of FX at a company, which a lot of companies don’t do well.”

Much more useful. At a NeuGroup virtual session this spring, a treasury operations manager from the multinational joined Mr. Bilter and AtlasFX Chief Solutions Officer Katya Spolidoro to provide a case study detailing how machine learning and AI models that use clean historical data can take forecasting beyond what’s possible with manual methods.

It included this challenge facing the manager: “When I need to hedge my May balance sheet, I don’t even have April actuals finalized yet. So we’re flying a little bit blind in terms of what do I need to hedge for my May balance sheet because that rate in essence is going to be set before I even know my April balances.”

The AtlasFX AI model allows him to avoid basing a forecast on information provided by scores of business units with varying degrees of accuracy. He relies instead on historical data. “We use AI basically to generate, here’s what your natural positions generally are, and that’s our basis for hedging and hedging an incremental risk.”

Treasury then validates the forecast with business leaders and asks about anything “extraordinary” that would affect the forecast. “We find this approach of using the AI to just use history to set the balances and set the table for us much more useful than traditional forecasting methodologies,” he told NeuGroup members in attendance.

Ms. Spolidoro noted that while AI that uses advanced machine learning algorithms has similarities to basic regression models, AI does not “get stuck in past trends.” It offers treasury teams a “more powerful, layered and dynamic technique that, most importantly, always improves with new data.”

Sweating bullets? The video clip is from an upcoming episode of NeuGroup’s Strategic Finance Lab podcast featuring an interview with NeuGroup Insights editor Antony Michels . In it, Mr. Bilter, who worked in FX risk management and FP&A roles at Hewlett-Packard, says that recent volatility in currency markets—and a significant decline in the U.S. dollar—sparked by uncertainty about tariffs and other concerns may reveal issues with some FX hedging programs.

“It’s a tougher environment for sure,” he said. “Folks will be going through their Q2 results and may be getting some nasty surprises. A lot of FX risk managers are kind of sweating bullets right now.” Not everyone, of course. “Those that have a good hedging program should be doing just fine in this environment,” he added.

“Although for other reasons, forecasts could be going awry depending on the company’s sensitivity to some of the tariff announcements and things like that. Or if the economy is at an inflection point that can make forecasting your own exposures more difficult. So it’s a challenge to varying degrees probably for everybody.”