Articles

An In-House AI Solution to Improve FX Exposure Forecasting Accuracy

  • By AFP Staff
  • Published: 2/3/2025
In-House AI FX Solution

Faced with a forecasting process that was time-consuming, labor-intensive and prone to error, the treasury team at ASML collaborated with its data science colleagues to create a fully automated AI-powered model.

The AFP Treasury Case Study series is designed to help you build up key treasury capabilities and skills by sharing examples of how leading practitioners have tackled challenges in their work and the lessons learned.

This case study is based on ASML’s Pinnacle Award entry. ASML was a finalist in the AFP 2024 Pinnacle Awards.

Insight: AI solutions can be developed using free open-source resources to increase the accuracy and efficiency of treasury processes.

Company Size:Large
Industry:Semiconductor
Geography:U.S., Europe, Asia
Topic:Emerging Technologies

Background

ASML is one of the world’s leading manufacturers of chip-making equipment. Founded in 1984 and headquartered in Veldhoven, the Netherlands, ASML operates in more than 60 locations across the U.S., Europe and Asia and sources components for its chip machines from a large worldwide network of suppliers.

Challenge

While ASML sources the vast majority of components in Europe, it sources some from the United States and pays in U.S. dollars. ASML sells its systems in Euro and is therefore exposed to EUR/USD exchange rate fluctuations.

To mitigate the effect on its profit and loss statement, ASML has a purchase FX hedging program in place. The foundation of this hedging program is the forecast of expected U.S. dollar-denominated material intake. The forecast had a low accuracy of only 70% and required a manual process that was labor-intensive, Excel-heavy, time-consuming and prone to error.

Approach

To generate forecasts more accurately and efficiently, the treasury and data science teams at ASML collaborated closely to create a fully automated AI model.

To start, they gathered five years of historic USD intake actuals and found twenty free open-source Python algorithms that looked promising for their use case. They taught the algorithms using the first three years of actuals and then tested their accuracy on the last two years of actuals.

Next, they chose the four best-performing algorithms out of the twenty and further developed them. After re-testing the four, they chose the best algorithm and further fine-tuned it. Before going live with the model, they met with internal stakeholders to ensure alignment.

The AI model recognizes patterns and trends in historic actuals and uses them to predict future intake. With each month, the model becomes “smarter” and improves the forecast accuracy even further.

Outcome

With the AI model, forecast accuracy went up from 70% to 96% and USD exposures were reduced by $25-50 million monthly, making the hedging program significantly more effective. The model also improves the treasury team’s understanding of the data while decreasing time spent on manual data gathering and processing, thereby freeing up time for more value-adding activities.

Because the approach is based on a free open-source algorithm, the AI solution is easily accessible and scalable, so it can be applied to more use cases within ASML and provide significant overall efficiencies.

Copyright © 2025 Association for Financial Professionals, Inc.
All rights reserved.