Financial institutions are increasingly replacing real customer data with synthetic profiles—AI-generated datasets that mimic human behavior—to accelerate product development and bypass regulatory hurdles. By using these artificial stand-ins, banks can test credit cards, loan products, and fraud detection systems without the privacy risks or compliance burdens associated with handling genuine consumer information.
Why Banks Are Turning to Synthetic Data
Traditional product testing often requires months of regulatory vetting and recruitment of live participants. Financial institutions now use synthetic data to compress these timelines. According to Global Finance, major firms like U.S. Bank deploy synthetic audiences to model high-net-worth households and refine marketing campaigns before public launch.
JPMorgan Chase utilizes synthetic financial data to simulate complex market behaviors, which aids in risk management and internal product design. Meanwhile, European institutions including NatWest, Monzo, and Santander are building synthetic data ecosystems to train their proprietary AI models, moving away from a total reliance on historical data that often becomes stale.
The Role of Regulatory Sandboxes
Regulators are actively working to bring these practices under formal oversight. In the United Kingdom, the Financial Conduct Authority (FCA) launched an AI Live Testing initiative to monitor how firms implement these technologies. The first cohort, which began in October 2023, included NatWest, Monzo, and Santander. A second group joined in April 2024, adding Barclays, Lloyds Banking Group, and UBS.
The FCA aims to evaluate use cases such as agentic payments, anti-money laundering (AML) detection, and know-your-customer (KYC) processes. This testing phase is scheduled to conclude in late 2026, with a comprehensive evaluation report expected in the first quarter of 2027. Firms participating in the sandbox have identified the program as a vital tool to overcome "proof of concept paralysis," a state where innovation stalls due to regulatory uncertainty.
Risks and Governance Challenges
While synthetic data is often viewed as a "safe" alternative to real consumer information, industry experts warn that it carries significant hidden risks. Mudit Gupta, an AI practice leader at EY, noted that synthetic data is not inherently risk-free.
- Inference Risks: Sensitive information can leak through patterns within the synthetic datasets.
- Bias Scaling: AI can replicate and amplify historical biases, embedding them into systems where they become difficult for human auditors to detect or challenge.
- Systemic Reliance: Because these systems are used for real-time judgments on identity and authorization, errors in the synthetic training data could lead to failures in fraud detection.
The urgency of these governance questions is heightened by the nature of banking threats. According to PYMNTS, unauthorized-party fraud accounts for 71% of financial institution losses, driven largely by account takeovers and credential theft. As banks rely more on AI to verify intent and identity, the FCA plans to publish a "good and poor practice" report on AI in financial services by the end of 2026 to guide industry standards.
Key Takeaways
- Efficiency: Synthetic data allows banks to simulate consumer behavior without the privacy exposure of real-world datasets.
- Regulatory Oversight: The FCA is actively testing AI use cases in banking, with a final report on best practices expected in early 2027.
- Hidden Dangers: Experts warn that synthetic data can mask historical biases and create new security vulnerabilities if not properly governed.
- Operational Shift: Adoption has moved beyond simple marketing simulations into critical areas like treasury operations and real-time fraud detection.