At U.S. Bank, the Real AI Bottleneck Is Decision Design
The systems behind turning AI pilots into production tools
In the rush to adopt artificial intelligence, many organizations have focused first on tools, frameworks, and pilots. At U.S. Bank, a different problem sits at the center of its AI strategy: how decisions are designed and improved, not just how technology is deployed.
In a conversation on CXOTalk, Prashant Mehrotra, the bank’s Chief AI Officer, described that focus. Rather than launching technologies without a clear view of value, the bank starts by questioning whether an existing process is worth transforming at all, and if so, how to do it responsibly and at scale.
Across the banking industry, adoption has accelerated quickly. Surveys show that a growing share of U.S. banks report deploying or actively using generative AI in some capacity. Mehrotra’s emphasis, however, is not on adoption alone. It is on measurable outcomes, repeatable processes, and decisions that can move from pilot to production without stalling.
Build, Measure, and Decide Faster
Mehrotra said ideas for AI initiatives emerge from across the bank, both top-down and bottom-up. “We evaluate these ideas not just for the value they bring, but also for things like technical feasibility, managing the risk, aligning with the risk posture of the bank, and ensuring that we are doing these things in a transparent and communicative manner,” he said.
That screening determines which ideas move forward, and which do not. Once pilots begin, the goal is convergence rather than proliferation. “We ran three or four different pilots,” Mehrotra said, describing work on AI-assisted code reviews. “We brought the group together… decided on one final product, and then we quickly scaled it.”
The principle is that pilots are not an end state. They are a means to identify what should be scaled (and funded) across the enterprise.
A recent example is the bank’s developer platform. In late 2025, U.S. Bank introduced a generative AI–based Developer Assistant within its external Developer Portal. The assistant helps business customers and partners identify relevant APIs, generate code snippets, and troubleshoot integrations. According to the bank, the tool is designed to shorten integration timelines and reduce manual effort.
Industry reporting described the assistant as a way to speed partner onboarding and reduce friction in embedded banking implementations.
Measurement plays a central role in deciding what scales. “You need to also have a baseline of how you are doing those things today,” Mehrotra said. Improvements are evaluated against prior performance, not expectations.
He pointed to contact-center use cases where generative AI supports agents with real-time information. Mehrotra said responses that once took multiple minutes could be delivered in tens of seconds, changing the pace of customer interactions without removing human involvement.
U.S. Bank has also productized AI for corporate clients. Its AI-driven cash-forecasting tool, developed with Kyriba, applies predictive analytics to treasury data to support liquidity planning and scenario analysis.
Embedding Governance and Culture into Decisions
For a regulated institution, scaling AI depends as much on governance as on technology. Mehrotra said risk, compliance, and governance partners are involved from the start. “We engage my risk partners very early on in the process,” he said. “Whatever we can actually learn during one new capability… we build it in the platform and we roll it out repeatedly.”
According to Mehrotra, this approach has reduced approval timelines in some cases by making governance repeatable rather than bespoke. The goal is not to lower standards, but to avoid relearning the same lessons with every new deployment.
Human oversight remains a core principle. As AI-driven outreach and automation expand, the bank maintains “human in the loop” guardrails and reviews how automated interactions behave in practice. Mehrotra said this is essential to meeting regulatory requirements and customer expectations.
Customers, he emphasized, are the reference point. “Our clients are the center of everything we do,” he said. That focus shapes how the bank uses customer data, aiming for relevance and timeliness without crossing into discomfort or intrusion.
Internally, U.S. Bank has paired its AI strategy with workforce education. Mehrotra described role-based learning programs designed for leaders, builders, and frontline employees, updated continuously as capabilities evolve.
Mehrotra described AI not as a standalone feature, but as something more structural. “I would call it an intelligence layer, not just a capability or a technology,” he said. The value comes from how that layer reshapes decisions: how quickly they are made, who makes them, and how consistently they improve.
As banks prepare for more advanced forms of automation, including agentic systems, Mehrotra said the groundwork must be laid now: data foundations, governance, and an educated workforce.