Major banks on both sides of the Atlantic are replacing real customer panels with AI-generated synthetic personas for product testing and model training. JPMorgan Chase uses synthetic financial data to simulate market behaviors for risk management and product design. US Bank models consumer segments, including high-net-worth households, to test messaging and refine campaigns before launch. In the UK, NatWest, Monzo, Santander, Barclays, Lloyds Banking Group, and UBS are participating in the FCA’s AI Live Testing initiative, which uses synthetic data ecosystems to train AI models for use cases including agentic payments, anti-money laundering detection, and know-your-customer checks.
The acceleration of synthetic data adoption in banking is driven by two converging pressures. Regulators require evidence that AI systems work accurately and fairly before deployment at scale, and real customer data creates consent, privacy, and bias exposure risks at every stage of that testing process. Synthetic personas sidestep those consent issues while compressing the testing timeline. JPMorgan’s approach to synthetic market data is designed specifically to speed up risk model development without touching live transaction records.
The original insight is that synthetic testing does not eliminate bias: it reconfigures where bias enters the system. As EY’s AI practice leader for the Americas noted, governance is what makes these systems deployable at scale. Synthetic data trained on historical records carries forward the patterns embedded in those records, potentially amplifying biases behind a layer of abstraction that makes them harder to detect and audit. Banks using AI personas for product testing are not just accelerating development; they are transferring the governance question from consent management to data provenance. That is a more tractable problem, but only if institutions treat it as one. Backbase’s acquisition of Kasisto is building toward the governed execution layer that makes these synthetic testing pipelines operationally meaningful.
Source: PYMNTS
