The power of Transaction Foundation Models: Building the unified intelligence layer for payments

2026-06-02 · Source: Thoughtworks Insights · Field: Finance & Economics — FinTech & Digital Financial Services, Banking & Financial Services · Depth: Advanced, long

Summary

Transaction Foundation Models (TFMs) represent the next evolution in payments intelligence, offering domain-specific AI models trained on vast financial transaction and event data. These models create a reusable intelligence layer, addressing the fragmentation, task-specific nature, and scaling challenges of traditional analytics in banking. TFMs learn the "language of money movement," converting patterns into embeddings for diverse downstream applications like fraud detection, credit scoring, and customer engagement. Notable examples include Mastercard's LTM, Revolut's PRAGMA (showing 130.2% credit scoring improvement), Stripe Radar (32% fraud reduction on \$1.9 trillion volume), and Adyen Uplift. NVIDIA provides a developer blueprint, partnering with Thoughtworks to operationalize these models, which promise better performance, faster reuse, improved data leverage, stronger scalability, and strategic differentiation for financial institutions.

Key takeaway

For AI Engineers and ML Directors modernizing payments intelligence, Transaction Foundation Models offer a strategic shift from fragmented analytics to a unified, reusable intelligence layer. You should explore NVIDIA's TFM blueprint and consider a staged implementation, starting with high-value use cases like fraud detection. This approach proves the model's value and builds a proprietary transaction intelligence asset, enhancing performance and scalability across your organization.

Key insights

Transaction Foundation Models learn the "language of money movement" from large-scale financial data to create reusable intelligence across banking and payments.

Principles

• Shared representation layer enables reuse.
• Contextual patterns improve predictive accuracy.
• Hybrid architectures enhance regulated environments.

Method

Building TFMs involves seven technical layers: data inputs, preparation, tokenization, sequence modeling, embedding learning, downstream adaptation, and production deployment with monitoring and governance.

In practice

• Start with fraud detection for measurable lift.
• Adapt embeddings for credit, operations, marketing.
• Combine TFMs with graph models for network intelligence.

Topics

• Transaction Foundation Models
• Financial Services AI
• Fraud Detection
• Payments Modernization
• Machine Learning Embeddings
• NVIDIA AI
• Thoughtworks

Best for: Investor, CTO, VP of Engineering/Data, AI Engineer, Machine Learning Engineer, Director of AI/ML

Related on AIssential

• How infrastructure is changing agentic payments: ‘A quiet but fundamental shift’ — Agentic payments are shifting fintech focus from user experience to robust, programmable financial infrastructure.
• Unlocking the relational power of graph neural networks in fraud prevention — Graph Neural Networks enhance fraud detection by modeling relational data, uncovering networked fraud invisible to traditional tabular methods.
• AI Hub — Domain-specific AI models are crucial for production systems in specialized industries like banking.
• UK payments startup Primer raises €86.2 million Series C to expand AI capabilities and accelerate US growth - EU-Startups — Unified payment infrastructure is essential for effective AI deployment in finance, preventing fragmented data from causing incorrect decisions.
• Sapiom raises $15M to help AI agents buy their own tech tools — Sapiom's financial layer enables AI agents to autonomously purchase and integrate external tech services.
• Strange intersections: The state of 21st century financial crime — Modern financial crime leverages FinTech and AI to blend illicit funds with legal transactions, complicating detection.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Thoughtworks Insights.