How Banks Are Using AI-Generated Synthetic Customers to Transform Product Development
For decades, launching a new banking product meant navigating a slow, expensive gauntlet: months of regulatory vetting, recruiting real customers for focus groups and pilots, and managing the significant legal exposure that comes with handling sensitive consumer data. Today, a growing number of the world's largest financial institutions are taking a dramatically different approach — they are building the customer themselves.
Artificial intelligence is enabling banks to create synthetic customer profiles, essentially AI-generated stand-ins that mimic real consumer behavior without using actual personal data. These digital clones are reshaping how banks design, test, and launch products, compressing timelines that once stretched across quarters into a matter of weeks.
What Are Synthetic Customer Profiles?
Synthetic customer profiles are AI-generated datasets that replicate the statistical and behavioral characteristics of real consumers. Rather than drawing on actual account holders' information — which triggers a cascade of compliance obligations under regulations like GDPR and the CCPA — banks generate fictional but statistically plausible consumer identities from scratch.
These profiles can represent virtually any demographic or financial segment: high-net-worth households, first-time borrowers, small business owners, or digitally-native millennials. Because no real personal data is used, the compliance exposure traditionally associated with customer research is dramatically reduced. And because synthetic profiles cost almost nothing to generate at scale, banks can run testing scenarios that would have been prohibitively expensive or logistically impossible with human participants.
The result is not just a faster development cycle. It is a fundamentally different model for how financial products reach the market.
Major Banks Already Deploying the Technology
Adoption is spreading rapidly across major financial institutions on both sides of the Atlantic, signaling that synthetic customer testing is moving from experimental curiosity to mainstream practice.
U.S. Bank is using synthetic audiences to model specific consumer segments, including high-net-worth households. The bank uses these profiles to test messaging strategies and refine marketing campaigns before any real customer ever sees them, allowing teams to iterate quickly and confidently ahead of launch.
JPMorgan Chase has gone a step further, generating synthetic financial data to simulate market behaviors for both risk management and product design purposes. By stress-testing products against a vast range of synthetic scenarios — economic downturns, sudden market shifts, changes in consumer spending patterns — the bank can anticipate failure points and design more resilient offerings from the outset.
In the UK, NatWest, Monzo, and Santander have each built or adopted synthetic data ecosystems specifically to train AI models. These ecosystems allow their machine learning systems to learn from rich, varied datasets without ever touching real customer records, keeping the training process clean from a regulatory standpoint while still producing capable, well-calibrated models.
The FCA Steps In: AI Live Testing and Regulatory Oversight
The scale and speed of synthetic data adoption has not gone unnoticed by regulators. In the UK, the Financial Conduct Authority (FCA) has moved proactively to bring the practice inside a formal regulatory framework, recognizing both its transformative potential and the governance questions it raises.
The FCA launched its AI Live Testing initiative with a first cohort that began in October, including NatWest, Monzo, and Santander. A second cohort launched in April, expanding the program to include Barclays, Lloyds Banking Group, and UBS. The FCA has described this initiative as the first of its kind in the financial services sector — a structured environment where institutions can test AI-driven applications under regulatory supervision before wider deployment.
The use cases being tested are consequential and wide-ranging. They include agentic payments — AI systems capable of initiating and completing transactions autonomously — as well as anti-money laundering detection and know-your-customer verification processes. All testing is scheduled to conclude by the end of 2026, with a full evaluation report expected in the first quarter of 2027.
The initiative reflects a broader regulatory recognition that AI in banking is not a future concern but a present reality. By engaging directly with institutions through a supervised testing environment, the FCA is attempting to stay ahead of the curve rather than react after deployment has already occurred at scale.
Why This Shift Matters for the Future of Banking
The move toward synthetic customer testing represents more than a technological upgrade. It signals a philosophical shift in how financial institutions relate to the product development process itself.
Traditional consumer research is inherently slow because it depends on human availability, consent, and behavior — all of which are unpredictable. Synthetic profiles remove that bottleneck entirely. A bank can generate thousands of behavioral scenarios overnight, test a product's performance across each one, and refine its design before a single human ever encounters it. That kind of iterative speed is transformative in an industry where being second to market can mean losing significant market share.
There are also meaningful implications for risk management. By simulating consumer behavior across extreme edge cases — scenarios that might occur only once in a thousand real-world interactions — banks can identify and address vulnerabilities that traditional pilot programs would almost certainly miss.
Governance Questions Still to Be Resolved
Despite the clear advantages, the widespread adoption of AI-generated synthetic customers is not without its challenges. Critics and regulators alike have raised questions about whether synthetic data can ever fully replicate the complexity and unpredictability of real human behavior. A model trained entirely on synthetic profiles may perform brilliantly in testing and stumble when it encounters the genuine messiness of real consumers.
There are also emerging questions about accountability. When a product designed and tested using synthetic data causes harm to real customers, who bears responsibility — the bank, the AI vendor, or the regulatory framework that permitted the approach?
These are questions the FCA's evaluation report, due in early 2027, will need to begin answering. In the meantime, the momentum is clear: synthetic AI customers are not replacing human consumers in banking boardrooms as an afterthought. They are becoming a central pillar of how the industry builds its future.