In the third installment of our Director of ML Insights series, we turn to the finance sector, where machine learning directors grapple with legacy infrastructure, regulatory scrutiny, and the critical question of trust. We spoke with experts from U.S. Bank, Royal Bank of Canada, Moody's Analytics, and a former Bloomberg AI researcher to uncover how ML is reshaping finance.
Ioannis Bakagiannis – Royal Bank of Canada
Background: A machine learning expert with a passion for scalable, production-grade solutions, Ioannis also hosts the Bak Up Podcast. Fun fact: He was a junior Greek national tennis champion.
1. How has ML made a positive impact on finance?
ML has created new financial products like personalized insurance and targeted marketing, but its most profound effect is sparking a conversation about trust in financial decision-making. Previously, humans trusted other humans—experts—for loans, rates, and portfolio management. When ML entered the scene, the question "Why trust a model?" forced the industry to examine transparency and ethics. This dialogue, still evolving, is making finance more accountable.
2. What are the biggest ML challenges within finance?
Established financial institutions face the persistent challenge of legacy systems. They must evolve to compete with agile newcomers while maintaining the robustness that underpins global finance. The risk is compounded: financial risk scales linearly with solution size, but system failures also bring regulatory and reputational risks. Balancing innovation with stability is perhaps the greatest hurdle.
3. What’s a common mistake people make integrating ML into financial applications?
Many assume ML models can simply replace existing rules without understanding the business context. The key is to start with interpretable models and gradually introduce complexity, always aligning with regulatory requirements and customer impact.
4. What excites you most about the future of ML?
The potential for ML to democratize financial advice, making personalized investment strategies accessible to everyone, not just the wealthy. This could transform personal finance on a global scale.
Debanjan Mahata – Moody's Analytics
Background: With over 100 patents, Debanjan is a prolific innovator in applied ML for finance.
1. How has ML made a positive impact on finance?
ML has improved risk assessment and fraud detection dramatically. For instance, behavioral models can now flag anomalies in real-time, reducing false positives and saving billions.
2. What are the biggest ML challenges within finance?
Data quality and availability remain critical. Financial data is often noisy, incomplete, or biased. Plus, models must be explainable to satisfy regulators, which conflicts with the complexity of deep learning.
3. What’s a common mistake people make integrating ML into financial applications?
Over-reliance on historical data without accounting for regime changes. Financial markets evolve; a model trained on data from a bull market may fail in a downturn.
4. What excites you most about the future of ML?
Causal ML—moving beyond correlation to causation—could revolutionize how we understand financial risk and policy impacts.
Soumitri Kolavennu – U.S. Bank
Background: A cycle polo player who regularly played at the world's oldest polo club, Soumitri now applies ML to banking.
1. How has ML made a positive impact on finance?
ML has enhanced customer experience through personalized recommendations and faster loan processing. It's also improved operational efficiency by automating compliance checks.
2. What are the biggest ML challenges within finance?
Bias in training data is a major concern. If not handled carefully, ML can perpetuate discrimination in lending or insurance. Governance frameworks are essential.
3. What’s a common mistake people make integrating ML into financial applications?
Ignoring the human-in-the-loop. ML should augment, not replace, human judgment—especially in high-stakes decisions like credit approval.
4. What excites you most about the future of ML?
Federated learning, which allows models to train on decentralized data without compromising privacy. This could unlock insights from sensitive financial data while maintaining security.
Disclaimer: All views are from individuals and not from any past or current employers.
This edition highlights that while ML in finance is still navigating legacy systems and regulatory hurdles, its potential to build trust and democratize services is immense.