U.S. Bank Uses Synthetic Personas to Accelerate Marketing Research

U.S. Bank is using AI-generated synthetic personas to help its marketing teams reach some audience and positioning insights more quickly, while keeping the tools separate from decisions about individual customers. The program began in 2024 during development of a brand campaign and has since expanded into product marketing and other business questions.

Michael Lacorazza, the bank’s chief marketing officer, said the personas helped cut the time needed to answer certain research questions by about half. The gains were most notable in competitive analysis and in synthesizing information about the marketplace. The approach is intended to inform employee judgment, not replace research with real customers or make decisions such as determining credit offers.

From brand work to product campaigns

U.S. Bank initially developed personas around its core target audiences to support brand strategy. Teams used them to explore how the bank might present itself in the market, assess competitors, and evaluate advertising concepts. The work also helped identify potential competitive white space and consider which messages or product proof points might appeal to different audiences.

The personas draw on a mix of third-party information, social listening and the bank’s existing research. That material includes audience profiles and customer segments, as well as findings from questions previously put to real people. Lacorazza described sources such as YouGov and social listening as inputs that can indicate attitudes, alongside the bank’s library of proprietary research.

U.S. Bank has grown the program to dozens of personas, added data sources and developed more audience-specific uses. Beyond brand marketing, the bank applies them to product campaigns related to deposits, small businesses and merchant acquiring. The personas also help teams refine value propositions for new products.

That expansion places synthetic personas within a wider personalization and insight effort, but their role remains bounded: they provide a research aid for marketers and other employees. They are not a representation of a particular customer whose account activity is being modeled for an individual decision.

Data boundaries and traceability

For a regulated financial institution, the origin and use of data are central considerations. U.S. Bank began with publicly available third-party information and research findings rather than individual transaction data from its customers. Lacorazza distinguished using research to guide marketing from placing identifiable customer activity into a model that could make decisions on a person’s behalf.

He said that models involving first-party client data, especially systems that make decisions about credit or the movement of money, would require a different level of scrutiny. The personas described in the program instead inform employee decisions. That boundary does not remove the need to review the system’s answers: a plausible-sounding response can still rest on weak or incomplete evidence.

One safeguard is traceability. The system can identify the sources that informed an answer, allowing employees to judge whether its evidence is substantial enough to support a conclusion. Lacorazza’s point is that confidence should depend in part on the quality and depth of the underlying sources, not simply on how definitive a generated response sounds.

A faster input, not a substitute for research

The bank continues to use customer panels and custom customer-experience research. Synthetic personas have reduced the need for some traditional research, but U.S. Bank still turns to people when the question warrants it. Lacorazza framed the choice in terms of importance: teams may rely on the existing evidence for a lower-priority question, while investing in direct research when the stakes justify it.

He also cautioned against treating human-sourced research as automatically accurate. People may be hesitant to discuss personal subjects, and poorly framed questions can fail to capture the intended insight. Both conventional research and model-generated findings need to be designed, checked and interpreted carefully.

U.S. Bank reported improvements after the brand campaign in creative performance, audience engagement, brand recall, consideration and awareness. Lacorazza credited the results primarily to the creative work, while saying persona insights helped sharpen the team’s recommendations. He did not present the models themselves as the cause of the campaign’s performance.

Practical guidance for marketers

For organizations considering synthetic personas, Lacorazza’s advice is to begin with the available evidence and a limited, clearly defined objective. Teams should inventory their existing research, identify gaps that public or third-party sources could address, and choose an initial use case rather than trying to answer every business question at once.

He also recommends making source transparency part of the system, then testing it with questions for which the organization already knows the answers. Those checks can reveal where the model needs refinement. U.S. Bank works with multiple major AI models rather than tying the program to a single provider, reflecting Lacorazza’s view that vendors and capabilities will continue to change.

At U.S. Bank, marketing applications for AI extend beyond research to process automation, personalization and content creation. Examples include optimizing email headlines, body copy and image selection, developing web pages more quickly, and supporting broadcast production. For a commercial featuring Christian McCaffrey, the bank used AI to fill a stadium with approximately 50,000 people rather than hire extras.

Lacorazza does not describe the objective as building a fully automated marketing department or simply reducing headcount. He sees efficiency as a way to redirect resources toward work that depends on human expertise and judgment. For marketers, the case illustrates both the potential speed of synthetic personas and the importance of keeping their conclusions grounded in inspectable evidence.