What proprietary AI has U.S. Bank built in-house?
U.S. Bank builds its own AI applications and agent software and runs them on a mix of its own, open-source and vendor models. Chief AI Officer Prashant Mehrotra said in July 2026 that the bank uses 'some proprietary models and open source models', balanced with models from OpenAI, Anthropic and Google so it is not stuck with one; it has not named the proprietary models. CIO Dilip Venkatachari says 'the agentic software that we developed underpins all the early AI use cases, both within and outside technology' (June 2026), and the bank has built a common platform with shared guardrails and an internal AI marketplace for reuse (October 2026). Named in-house tools: Statement Analyzer (STAT), developed in-house and rolled out in 2022, which processed more than 30,000 merchant statements in a year; Design Assistant, built by the Experience Design team and the AI Center of Excellence and used by designers since December, with about 90% adoption and 30% fewer accessibility defects (March 2026); a GenAI Developer Assistant for outside developers, quietly launched in October 2025 on an unnamed large language model; a GenAI code-review assistant that cuts review time by 75%; and the voice-first Smart Assistant, launched in 2020. Customer-facing agents are being built on vendor infrastructure: the May 2026 AWS agreement names Amazon Bedrock, Nova Sonic and Connect Customer. Mehrotra's rule is to build where a use case relies on internal data, processes or intellectual property and to buy software where it does not.[1][2][3][4][5][6][7][8][9][10][11]
Data as of Oct 22, 2026 · U.S. Bank profile updated Oct 16, 2026 · 22 sources cited on this page · 140 banks profiled on this site · Markdown
Which AI systems has U.S. Bank built itself, and what do they run on?
'Built on' is what the bank has said about the models underneath; 'not disclosed' means it has not named them.
| Platform or model | What it does | Built on | Since | Status |
|---|---|---|---|---|
| In-house agentic software | 'Underpins all the early AI use cases, both within and outside technology'; a streamlined agentic infrastructure for launching business use cases, with mortgages, credit decisioning and KYC named as areas to explore | Developed by the bank's technology team; models not disclosed | 2025 | In production[2][12][13] |
| Common AI platform and internal AI marketplace | Shared services and guardrails; teams discover and reuse each other's AI tools, with citation-style credit for creators, to avoid '1,000 AI experiments becoming 1,000 technology stacks' | Bank-built | October 2026 (described) | In use[3] |
| Statement Analyzer (STAT) | Reads merchant statements and turns them into pricing proposals for Merchant Payment Services reps; rebuilt after rep feedback; now reads poor-quality statements visually; more than 3,500 statements in a month and 30,000-plus in a year | Developed in-house; visual AI replaced document scanning; Mehrotra says the merchant-quote work combines generative AI, language models, vision models and predictive analytics | 2022 | In production[4][1] |
| Design Assistant | Checks designs in Figma against brand, content, tone and accessibility standards, cites the standard and offers a one-click fix with a human in the loop; about 90% designer adoption; 30% fewer accessibility defects | Built in-house by the Experience Design team and the AI Center of Excellence over a couple of quarters; model trained on 'what good looked like'; vendor not disclosed | December (rollout reported March 2026) | In production for hundreds of designers[5][6][14] |
| Developer Assistant | GenAI chat in the Developer Portal for outside developers: finds APIs, explains errors, generates sample code; draws on all the bank's APIs, testing scripts and documentation | An existing large language model the bank declined to name; gated to the portal | October 2025 (quiet launch); announced January 2026 | Live for clients and partners[7][15] |
| GenAI code-review assistant | Reviews code for engineers; review time cut by 75% | Not disclosed | 2025 | In production[8] |
| Smart Assistant | Voice-first mobile assistant using natural language processing to recognize intent; Spanish version from 2022 | Built by the bank's digital team ('The customer experience drove how we built the Smart Assistant') | July 2020 | In production[9][16] |
| Agentic self-service | AI agents for voice, chat and SMS across mortgage, cards, wealth and commercial banking | Amazon Bedrock, Amazon Nova Sonic and Amazon Connect Customer, according to AWS | May 2026 | Being built[10][17] |
Who builds U.S. Bank's in-house AI?
| Person or team | Role | Since |
|---|---|---|
| Prashant Mehrotra, Chief AI Officer | Leads AI strategy and execution; describes the job as 'evangelizing, educating, enabling and executing'; previously head of AI at Allstate and director of data engineering at Capital One | Mid-2025 ('a little over a year' by July 2026)[1][11] |
| Dilip Venkatachari, CIO | Senior EVP and chief information and technology officer; led the agentic software and infrastructure; holds more than a dozen US patents | On the record 2025-2026[2][13] |
| AI Center of Excellence | Co-developed Design Assistant with the Experience Design team led by Caleb Schmidt | 2025[5] |
| Merchant Payment Services AI team | Built and rebuilt STAT; Raj Bharadwaj is principal engineer for AI | 2022[4] |
| Digital (Ankit Bhatt) | Chief digital officer for consumer, small business and platforms; Smart Assistant and the Developer Assistant | 2020; 2025[9][7] |
| Federated builders | Started centralized and is federating out; central teams own foundations, developer tools and governance while AI talent sits with business teams; 'thousands upon thousands of builders' | 2026[1][3] |
| Use-case governance body | Reviews every proposed use case through a common lens; buckets them into growth, cost savings and unmeasurable productivity | 2026[1] |
| Engineers | 70%-75% of all engineers trained and actively using new AI tools within three months | June 2026[2] |
When does U.S. Bank build AI and when does it buy?
Mehrotra, asked in July 2026 whether the bank favors frontier or in-house models: 'We've been using some proprietary models and open source models. We balance that out with models from OpenAI, Anthropic and Google.' He said the bank wants a balanced environment 'where we are not stuck with one' and to protect 'highly proprietary, confidential data that makes those models so much better'.[1]
His decision rule, given in May 2026: build for use cases that rely heavily on internal data, processes or intellectual property, and use SaaS products where they deliver the function cost-effectively; for smaller projects, ask whether a built capability 'will stand the test of time' or will soon be available from a third party.[11]
Venkatachari said in November 2025 that after multiple proofs of concept the bank learned it is better to pick a tool that is easier to integrate, even if it is slightly behind the cutting edge. In October 2026 he said the hard part is not the models but 'the plumbing' of data, identity, integration, observability and deployment, and that maturing vendors now let banks use external scaffolding as long as it connects to the bank's data and controls.[13][3]
Design Assistant shows the build case: the bank built it inside Figma because 'if we wanted our teams to use something that we built, it needed to be part of that broader ecosystem that they're working in every single day', head of experience design Caleb Schmidt said. Coding tools are bought: GitHub Copilot first, then Cognition and Claude Code.[6][1]
What has U.S. Bank not disclosed about its proprietary AI?
The bank has not named its proprietary models or said what they do, and declined to name the large language model behind the Developer Assistant. It declined to say how much it invested in Design Assistant; total technology spending runs at about $2.5 billion a year, with no AI line disclosed.[1][7][6][18]
U.S. Bank holds a granted patent on agentic on-device adaptation (August 2025) and filed one on LLM-assisted hierarchical code review (September 2025), but has not said whether they cover its agentic software or its code-review assistant. Headcount for the AI Center of Excellence is not published.[19][20][5]
What does the U.S. Bank profile record on this topic?
The rows below are the entries of the U.S. Bank AI strategy profile that bear on this page, as of Oct 16, 2026; the profile holds the full record of 34 dated moves, 18 use cases and 51 sources.
| Use case | What the record shows | Status | Source |
|---|---|---|---|
| Agentic software layer | Internally developed agentic software underpinning early use cases inside and outside technology. | In production | [2] |
| Design Assistant and developer-portal assistant | In-house generative tools for designers and API users. | In production | [5][7] |
| STAT merchant statement analysis | Vision AI reading inconsistent merchant statements. | In production | [4] |
| Smart Assistant | Virtual assistant in the mobile app that works by text and by voice. | In production | [21] |
| GenAI code-review assistant and coding agents | A GenAI code-review assistant cuts review time by 75%; engineers moved from GitHub Copilot to a wider set of tools including Cognition and Claude Code, and AI is also used to test more code. | In production | [8][1] |
| Multi-model, multi-vendor stack | Proprietary and open-source models balanced with models from OpenAI, Anthropic and Google, so the bank is not stuck with one provider; AWS and Kyriba are the named platform partners. | In production | [1][17][22] |
| Internal AI marketplace and shared platform | A common platform with shared services and guardrails, plus a marketplace where teams find and reuse each other's AI tools. | In production | [3] |
| Date | Move | Detail | Source |
|---|---|---|---|
| Oct 7, 2026 | Venkatachari: from proving AI to industrializing it | The CIO describes a federated model with shared guardrails and an internal AI marketplace, and says the first use case may need a village but the tenth should take an afternoon. | [3] |
| Sep 23, 2026 | Smart Assistant cited as a voice-capable virtual assistant | Keynova's scorecard, which ranks U.S. Bank's mobile banking first for a third year, notes only it and Bank of America offer voice in their assistants. | [21] |
| Mar 30, 2026 | Design Assistant launched | Built by the Experience Design team and the AI Center of Excellence to enforce accessibility, brand and content standards. | [5] |
| Feb 23, 2026 | 2025 annual report: AI delivers measurable results | Code-review assistant cuts review time by 75%; Smart Assistant used 2.5 million times a month; contact-centre agents get a GenAI assistant; the CEO letter says the bank built in-house AI capabilities in 2025. | [8] |
| Jan 1, 2022 | STAT statement analyser built in-house | Rolled out to merchant-services reps; later rebuilt on vision AI that reads poor-quality statements. | [4] |
Does U.S. Bank build its own AI models?
Partly. Chief AI Officer Prashant Mehrotra said in July 2026 that the bank uses some proprietary models and open-source models alongside models from OpenAI, Anthropic and Google, but it has not named the proprietary models. Most of what it describes building is software around models: in-house agentic software, an internal AI marketplace, STAT, Design Assistant and the Developer Assistant.
What AI tools has U.S. Bank built in-house?
Named examples are Statement Analyzer (STAT) for merchant pricing proposals, rolled out in 2022; Design Assistant for experience designers, rolled out in December and reported in March 2026; the GenAI Developer Assistant for outside developers, launched quietly in October 2025; a GenAI code-review assistant; the agentic software the CIO says underpins its early AI use cases; and the Smart Assistant from 2020.
What is U.S. Bank's Design Assistant?
An AI tool built in-house by the Experience Design team and the AI Center of Excellence. It runs inside Figma, checks designs against the bank's brand, content, tone and accessibility standards, cites the standard broken and offers a fix that a designer approves. About 90% of designers use it and accessibility defects fell 30%, the bank said in March 2026.
Who leads AI at U.S. Bank?
Prashant Mehrotra, chief AI officer since mid-2025, formerly head of AI at Allstate and director of data engineering at Capital One, leads AI strategy and execution. Dilip Venkatachari, senior EVP and chief information and technology officer, led the bank's in-house agentic software, and an AI Center of Excellence builds tools with business teams.
Which vendor AI does U.S. Bank use?
Models from OpenAI, Anthropic and Google, per Mehrotra; GitHub Copilot, Cognition and Claude Code for engineers; and, under a May 2026 agreement, Amazon Bedrock, Amazon Nova Sonic and Amazon Connect Customer as the base for agentic self-service in voice, chat and SMS.
- Beyond token-maxing: How U.S. Bank AI chief navigates costs
- Dilip Venkatachari's development of agentic AI and digital-asset infrastructures is driving use cases
- How U.S. Bank Is Industrializing AI
- AI-powered Statement Analyzer simplifies payment services for small businesses
- U.S. Bank launches Design Assistant, an AI solution to empower designers at scale
- U.S. Bank arms designers with AI tool
- U.S. Bank gives embedded banking clients a gen AI assistant
- U.S. Bancorp Annual Report 2025
- Introducing the U.S. Bank Smart Assistant
- U.S. Bank expands collaboration with AWS to accelerate progressive technology transformation and AI-driven customer experience innovation (according to AWS)
- U.S. Bank Chief AI Officer on Transformation: 'If Not Now, When?'
- Three U.S. Bank leaders named to American Banker Most Innovative list
- U.S. Bancorp CIO on Scaling AI: 'The Technology Is the Easy Part'
- Bank of America and U.S. Bank Move AI Into Core Internal Operations
- GenAI assistant helps companies embed U.S. Bank solutions into their platforms (undated; accessed 2026-10-15)
- U.S. Bank launches nation's first Spanish-language voice assistant for banking
- U.S. Bank shifts critical apps to AWS for AI push
- U.S. Bancorp at Barclays Conference: Strategic Focus on Growth and Stability
- US12401680B1 - Agentic on-device adaptation for increasing node efficiency by accommodating distribution shifts
- US12602308B1 - LLM-assisted hierarchical code review system
- U.S. Bank tops mobile banking ranks for third straight year
- U.S. Bank launches AI-driven cash forecasting tool
When U.S. Bank moves on AI, you'll read it here first.
when one of the banks we cover moves on AI, it is in the next morning's brief · six sourced stories · 7 am PT · free
plus every regulation, bank and agent page update, the morning after · leave any morning