Derisking machine learning in banking

Artificial intelligence and machine learning are set to transform the banking industry. Applying these technologies to improve decision making, tailor services, and enhance risk management could generate more than $250 billion in value across the sector, according to the McKinsey Global Institute. In fact, risk management is one of the functions likely to reap the greatest benefits, especially in fraud prevention and credit underwriting.

But the power of these technologies comes with a downside. Algorithms can incorporate biases that result in unfair discrimination; recommendation engines can propose unsuitable products to customers; and AI models can deteriorate quickly over time. Poorly managed machine-learning models could expose a bank to reputational, ethical, regulatory, and financial risks.

To combat these risks, banks need to refine their model governance and model validation practices, as McKinsey partner Derek Waldron explains. That doesn’t mean throwing out old validation frameworks used for established statistical models, but rather adding a small number of new focus areas, such as interpretability and bias. Derek explains a few of these examples in this video; for a deeper dive on the topic, see our recent article “Derisking maching learning and artificial intelligence.”

Derisking machine-learning models will undoubtedly take time—but committing to building, testing, and refining thoughtfully designed new policies and practices will give banks a head start in harnessing the full power of these transformative technologies.