Why Financial Institutions Are Rethinking AI Models

Financial institutions have been entrenched in AI for years, crafting models for fraud, credit, and risk, among others. Yet, these models often operate in isolation, like solo players in a team sport. That's a big miss.

Siloed systems mean missing out on the larger picture of consumer behavior. As data grows, the gap between what these institutions know and what their AI can actually process becomes glaring. It's a gold mine of opportunity for those ready to harness proprietary data.

The AI Transformation

According to NVIDIA's 2026 State of AI in Financial Services report, a hefty 65% of financial outfits are already using AI. Almost 90% are either deploying or assessing it, with most not skimping on the budget. But here's the catch: scaling up AI also amps up complexity. Fragmented models are becoming a bottleneck.

So, what's the industry doing? They're rethinking the architecture itself. Enter transformer-based transaction foundation models. These aren't just buzzwords, these models are game-changers, allowing firms to understand consumer behavior in a unified way, all trained on their own data.

Imagine a payment at midnight. If it's the fourth in ten minutes, on a new device, and in an unfamiliar city, it spells something different. Context is king here. This shift isn't just structural, it's transformational.

NVIDIA Leads the Pack

Take Revolut, for instance. They've collaborated with NVIDIA to roll out PRAGMA, a suite of transformer-based models trained on 24 billion financial events worldwide. The results? Their models outperform traditional ones across various domains, from credit scoring to fraud detection. And they do it without the cumbersome feature engineering that used to take weeks or months.

Why should we care? Because this isn't just tech for tech's sake. It's about making financial services smarter and more efficient. Fundraising isn't traction, but this tech shift could be just that.

The Big Picture

Mastercard and Adyen are also in on the action, embracing similar foundation models. Mastercard's large tabular foundation model aims to scale enormously. Early tests outperform standard methods, with applications in cybersecurity and fraud detection.

For Adyen, even a fractional improvement of 0.1% in authorization can mean a ton of value. It's about maximizing conversion and minimizing risk. So, why wouldn't a financial institution want to hop on this train?

In a world where transaction data is king, the ability to turn that data into actionable insights is invaluable. The infrastructure is ready, the architecture is proven, and the data is already there.

As financial firms adopt NVIDIA's Build Your Own Transaction Foundation Model, they're not just catching up, they're stepping into a future where AI isn't just a tool but a cornerstone. So, where do you see your institution in this AI revolution?