Banks Swap Real Customers for AI Clones in Product Tests

Testing a new credit card or banking product once required months of regulatory vetting and customer recruitment. Now banks build the customer instead.

Financial institutions are replacing live customers with artificial intelligence-generated stand-ins – synthetic profiles that cost almost nothing and carry none of the compliance exposure tied to real customer data, Global Finance reported. The synthetic consumer doesn't just compress timelines. It changes how banks bring products to market.

The adoption is spreading across major institutions on both sides of the Atlantic. U.S. Bank deploys synthetic audiences to model consumer segments such as high-net-worth households, testing messaging and refining campaigns before launch, Global Finance reported.

JPMorgan Chase generates synthetic financial data to simulate market behaviors for risk management and product design. NatWest, Monzo and Santander are using synthetic data ecosystems to train AI models.

In the U.K., the Financial Conduct Authority (FCA) has moved to bring the practice inside a regulatory framework. The FCA's AI Live Testing initiative launched its first cohort in October, including NatWest, Monzo and Santander. A second cohort began in April, adding Barclays, Lloyds Banking Group and UBS, the FCA announced. Use cases include agentic payments, anti-money laundering detection and know-your-customer checks. Testing concludes by the end of 2026, with an evaluation report due in the first quarter of 2027.

The FCA described the initiative as the first of its kind in the financial sector, the agency reported. Firms cited AI Live Testing as a way to overcome what they called "proof of concept paralysis," where AI initiatives stall because of regulatory uncertainty.

The sandbox has limits. Most banking leaders believe agentic AI can move faster if governance were not perceived as a constraint, Mudit Gupta, EY's AI practice leader for Americas financial services consulting, told Global Finance. "In practice, governance is what makes these systems deployable at scale," Gupta said.

Gupta added that synthetic data is often treated as inherently safe. It is not. It can leak sensitive signals through inference and linkage risks. It can also replicate and scale historical biases, embedding them behind a layer of abstraction that makes them harder to detect, audit and challenge, he said.

The scale of adoption makes the governance question urgent. As PYMNTS reported, the technology is already moving into treasury and finance operations, where forecasting models have historically relied on data that quickly becomes stale.

Regulators are unlikely to treat AI outputs as abstract or low stakes. Unauthorized-party fraud accounts for 71% of incidents and losses at financial institutions, driven by credential theft and account takeovers, PYMNTS reported. These are the areas where AI is being deployed to make real-time judgments about identity, authorization and intent. The FCA has said it will publish a good and poor practice report on AI in financial services later in 2026.

Among the banks in the cohort, JPMorgan Chase carries an Alpha Score of 54 out of 100 (Mixed), while Lloyds scores 65 (Moderate) and Barclays scores 59 (Moderate). The scores reflect the market's mixed view on how these institutions are positioned for the regulatory and operational shifts ahead.