Data Moats in the Age of Foundation Models

Everyone can access GPT-4. Nobody can access your customer data. This is the real moat in AI competition post-foundation models.

The mistake is thinking that foundation models make data irrelevant. The opposite is true: foundation models make proprietary data more valuable, not less. Why? Because the foundation model is a generic reasoner—good at language, good at patterns, good at zero-shot tasks. But it knows nothing about your customers, your business, your context.

Nubank's AI delivers 256% ROI not because they use a better foundation model—they train on 100M customer records no competitor can touch. That proprietary context encodes domain knowledge that no vendor LLM will ever have: Which customers are likely to repay? Which ones are credit risks? What financial patterns predict churn? A foundation model trained only on public internet data can't answer these questions for Nubank's specific market. But Nubank's model, fine-tuned on their actual customer data, can.

Ramp's merchant classification system improves continuously because it learns from millions of real transaction sequences. Every expense categorized creates training data that makes the next categorization more accurate. The more transactions Ramp processes, the better the model becomes—a flywheel that competitors without transaction volume can't replicate.

RBC's NOMI works because it understands their customer base's unique financial patterns. Traditional banks were skeptical of AI credit decisions. RBC proved that when you train on the specific economic patterns of your customers, the model becomes trusted. That's not generic AI—that's bespoke advantage.

The implication is stark: if your AI advantage depends on vendor models, you don't have an advantage. You have a subscription to someone else's advantage. You need a strategy for capturing and leveraging proprietary data. Build data collection into your product. Make every user interaction training data. Design workflows that create high-quality labeled examples. That's how you build moats that last.

About the Author

Durai Rajamanickam is a Business Transformation Leader and author of The AI Inflection Point: Volume 1 - Financial Services. With over two decades of experience, he specializes in AI-driven enterprise transformation, designing evidence-based ROI frameworks, and helping organizations modernize legacy systems with intelligent automation.

His work focuses on translating AI ambition into measurable business outcomes, with case studies spanning Ramp, Nubank, Coinbase, RBC, and Stripe—all showcasing AI ROI between 256% and 1,700%.

Connect on LinkedIn