What proprietary AI has Citizens built in-house?
Citizens builds AI applications and buys the models. CIO Michael Ruttledge said in September 2024 that 'our plans right now are not to build our own LLM models' and that the bank uses 'best-in-class models from our technology providers'. What it has built on top: a gen-AI portal in which employees choose OpenAI, AWS Bedrock, Anthropic or another provider (December 2025); a gen-AI model for contact-centre product questions that replaced a legacy app (July 2024); CiZi, the AI assistant in the mobile app since November 2024, credited with a 44% cut in mobile-banking calls; LLM voice agents that order debit cards through back-end systems, with 50%-plus of calls to be answered by a non-human over time; an AI lending onboarding platform built by Ruttledge's team; a cloud fraud platform with new detection models; and Digital Butler, a commercial assistant patented in 2024. The 3Q26 deck (October 16, 2026) adds AI phone support built for credit-card calls, AI meeting prep used by about 50% of commercial bankers and analysts, and AI coding tools used by about 95% of engineers. Chief data and analytics officer Krish Swamy said in July 2026 that some models run inside the bank's own VPC. Citizens has not disclosed which model runs CiZi or the voice agents, the size of its AI team or AI spend within its $300 million Reimagine the Bank budget.[1][2][3][4][5][6][7][8][9][10]
Data as of Oct 22, 2026 · Citizens profile updated Oct 13, 2026 · 33 sources cited on this page · 140 banks profiled on this site · Markdown
Which AI systems has Citizens built itself?
Systems Citizens describes as its own, oldest first. 'Built on' records what the bank or its vendors have said about the models and platforms underneath.
| Platform or model | What it does | Built on | Since | Status |
|---|---|---|---|---|
| Digital Butler | Commercial-banking virtual assistant giving clients advice and a customised experience; 30,000 clients using its gateway each week; patent granted for its 'high degree of personalization' | Not disclosed | Web first; commercial app Q4 2023; patent March 2024 | In production[8] |
| Contact-centre product model | Gen-AI model that replaced a legacy app: answers representatives' product questions and suggests the best product; a related on-screen product-suggestion tool was used by more than half of customer service representatives by December 2025 | Shared internal gen-AI platform; models from 'our technology providers' | July 2024 | In production[3][2] |
| CiZi | AI assistant in the mobile app: direct deposit, digital wallet, cards, balances and statements; mobile-banking-related calls down 44% | Model not named in bank materials | November 19, 2024 | In production[4][5][2] |
| Cloud fraud-management platform | Rebuilt detection, monitoring, disputes and claims; about 80% of claims volume since 2023; new detection models that the bank says detect fraud above the industry standard of 10% to 11%; machine learning for suspicious-activity reporting, agentic AI in pilot | MongoDB Atlas on AWS | 2023 | In production; agentic pilot[7][2] |
| Gen-AI portal | Employees pick a model per task: 'This gives us the freedom to adopt different ones where we think someone's got a leap on the market', Ruttledge said | OpenAI, AWS Bedrock, Anthropic or another LLM provider | December 2025 | In production[2] |
| AI lending onboarding platform | Built by the CIO's team; live for student lending, components reused for home equity and planned for credit cards | Not disclosed | December 2025 | Rolling out[2] |
| Agentic complaint handling and call QA | Agents classify incoming complaints and respond to customers; a gen-AI pilot reviews 100% of call transcripts for tone and accuracy against 5% reviewed by people | Not disclosed | November-December 2025 | In production; QA in pilot[11][2] |
| LLM voice agents | Understands callers, routes them and completes tasks such as reordering a debit card or reading recent transactions; voice agents inside the mobile app took adoption from 0% to about 60%, per AWS; target 50%-plus of calls answered by a non-human, halfway by end-2026 | An LLM (provider not named); Amazon Connect; deterministic workflows blended with agentic AI | March 2026 (pilot) | Rolling out[6][12] |
| Enterprise AI-chat agent | First enterprise-wide AI-chat agent for colleagues, starting with HR content; back-office re-architecture with AI agents orchestrating end-to-end workflows begun | Not disclosed | July 2026 (2Q26 deck) | Launched[13] |
| AI-Powered Banker | Packages internal and external data into a client 'health report' for bankers; AI research and meeting prep used by about 50% of bankers and analysts in Commercial Banking | Not disclosed | March 2026 | In production in Commercial Banking[6][9] |
| AI phone support for credit cards | Built for the most common reasons credit-card customers call; AI-driven complaint logging for Credit Card | Not disclosed | October 16, 2026 (3Q26 deck) | In production[9] |
Who builds Citizens' in-house AI?
Michael Ruttledge, Chief Information Officer and Head of Enterprise Technology & Security, runs the build: his team built the AI lending onboarding platform, and in July 2024 he described a hackathon that produced more than 90 ideas, narrowed to about half a dozen, with the first two production uses sharing one platform 'rather than having everybody develop' their own.[2][3]
Chief data and analytics officer Krish Swamy leads the data side; in July 2026 he described using large language models 'as a translator' that maps what callers say to a small set of things the bank can answer.[10]
Outside capacity comes from Infosys. In April 2026 Ruttledge said 'our AI-first Innovation Hub reflects Citizens' long-term commitment to building modern, secure, and intelligent banking capabilities', built with Infosys and its Topaz Fabric.[14]
The 3Q26 deck counts more than 500 colleagues dedicated to executing Reimagine the Bank, with 35-plus of 50-plus planned initiatives launched and about 95% of software engineers using AI coding tools. Citizens Bank's Deepak Nair presented the voice-agent build with AWS at Customer Contact Week 2026.[9][12]
What does Citizens build, and what does it buy?
Models are bought. Ruttledge said in September 2024 that the bank was working with its cloud providers, Amazon Web Services and Microsoft, to deploy large language models, and that 'our plans right now are not to build our own LLM models'. By December 2025 the gen-AI portal offered OpenAI, AWS Bedrock and Anthropic models side by side.[1][2]
Some hosting is in-house. Swamy said in July 2026 that, with the bank 100% in the cloud, certain models are hosted and run 'within our VPC' so that 'nothing goes out of the VPC', for some use cases but not all, because 'setting up a model instance and running it is also its own kind of engineering task'.[10]
Applications and workflows are built. President Brendan Coughlin said in March 2026 that the bank is 'building all these work streams' for the voice agent, which connects to back-end operations to complete requests rather than only routing calls.[6]
Vendors are being cut, not added. The 3Q26 deck reports a streamlined process for approving new technology, AI tools and vendors, 35 applications rationalised in 2026 and about 90% of the 2026 vendor savings target reached; on the October 16, 2026 call management said it has 'significant resources dedicated to leveraging AI'.[9][15]
What has Citizens not disclosed about its in-house AI?
Citizens' own materials do not name the model behind CiZi or the LLM voice agents, the number of employees using the gen-AI portal, or which models are hosted inside its VPC.[16][10]
No AI engineering headcount or AI-only budget is published; the AI work is funded inside Reimagine the Bank, a three-year, $300 million program targeting at least $450 million of pre-tax run-rate benefit by year-end 2028. The Digital Butler patent was reported by Bank Automation News without a patent number.[6][9][8]
What does the Citizens profile record on this topic?
The rows below are the entries of the Citizens AI strategy profile that bear on this page, as of Oct 13, 2026; the profile holds the full record of 35 dated moves, 20 use cases and 50 sources.
| Use case | What the record shows | Status | Source |
|---|---|---|---|
| CiZi | Mobile-app AI assistant; call volume down 44%. | In production | [5][17] |
| Coding agents | GitHub Copilot's agent for ~800 developers; +55% output for advanced-agent users. | In production | [11][5] |
| Payment-fraud defence | Head of fraud on AI deepfakes, BEC and bank impersonation. | In production | [18] |
| LLM call-centre automation | Targeting ~25% of calls handled without human interaction by year-end 2026 and ~50% over time; AI also in the credit-card contact centres. | Rolling out | [19][13][20] |
| Back-office AI agents | AI agents orchestrating end-to-end workflows to cut manual tasks; AI case classification for client inquiries. | Pilot | [13] |
| Enterprise AI-chat agent | The first enterprise-wide AI-chat agent, starting with HR content for colleagues. | In production | [13] |
| AI in the 10-K risk factors | Generative AI and LLMs as an information-security risk; AI as a potentially transformative competitive change. | In production | [21] |
| Agentic fraud-claims and AML checks | Fraud platforms rebuilt in the cloud with machine learning for suspicious-activity reports and security checks; agentic AI piloted for straight-through processing of small-dollar claims, with AML checks to follow. | Pilot | [2][7] |
| Complaint classification and automated call QC | Agentic AI classifies complaints and drafts responses; gen AI piloted to review 100% of call transcripts for tone and accuracy against 5% reviewed by people today. | Pilot | [2][13] |
| AI lending onboarding platform | Built for student lending, parts reused for home equity and planned for credit cards, so new scores such as cash-flow underwriting deploy faster. | Rolling out | [2] |
| Engineers as supervisors of coding agents | Citizens is moving engineers from writing code to managing code-writing agents; average engineer productivity up 30% to 40% in a year. | Rolling out | [22][23] |
| Models hosted inside the bank's VPC | For some use cases Citizens runs models in its own cloud network so nothing leaves its firewall, trading engineering effort for control. | In production | [10] |
| Date | Move | Detail | Source |
|---|---|---|---|
| Sep 29, 2026 | Coughlin: engineers now 30-40% more productive with AI | American Banker reports the average Citizens engineer is 30% to 40% more productive than a year earlier under the three-year, $300 million Reimagine the Bank project; Coughlin says the role is evolving toward engineers overseeing multiple AI agents, and that AI-existential-risk warnings have 'not really' changed the project's direction. | [23] |
| Aug 3, 2026 | Q2 2026 10-Q names Reimagine the Bank but carries no AI text | The quarterly report attributes higher outside-services expense to costs of the Reimagine the Bank programme; neither the Q1 nor Q2 2026 10-Q contains the words AI, artificial intelligence, generative or agentic. | [24] |
| Jul 30, 2026 | CDAO Swamy: LLMs as a translator for customer intent; models inside the bank's VPC | On an American Banker panel Krish Swamy said LLMs map what callers say to a small set of intents, that some models run inside Citizens' own VPC, that chatbots answer questions but do not yet execute transactions, and that commercial underwriters share a Copilot prompt library. | [10] |
| Jul 16, 2026 | Q2 2026: AI agents in the back office, a 10x engineering pilot | Back-office re-architecture with AI agents orchestrating end-to-end workflows, AI case classification, AI in credit-card contact centres and a first enterprise-wide AI-chat agent; about $7 million of program costs in the quarter. | [13][25][26][27][28] |
| Jun 1, 2026 | The mainframe goes: CIO's cloud-first claim | 700 applications migrated; CiZi −44% call volume; +55% output for engineers on advanced agents. | [5] |
| Apr 16, 2026 | Q1 2026: LLMs in the call centre, 25% of calls without a human by year-end | About 50% over time; GenAI automating private-company credit research; AI coding pilots showing 30% to 10x productivity gains; $6 million of implementation costs. | [19][20] |
| Mar 5, 2026 | Proxy statement: AI and agentic AI in how the bank is run | Reimagine the Bank described to shareholders. | [29] |
| Mar 3, 2026 | 2025 Annual Review: first super-regional fully in the cloud; CiZi use up 40% | The shareholder review says Citizens became the first super-regional bank fully in the cloud in 2025, set up a transformation office for Reimagine the Bank, and that CiZi usage rose 40% year over year in 4Q25. | [30] |
| Feb 12, 2026 | 10-K flags generative AI and LLMs as a security risk | Risk factors name generative AI and large language models among evolving threats, and AI as a change that could transform how banks offer products. | [21] |
| Jan 21, 2026 | Q4 2025 deck: ~50 initiatives in five themes | Software productivity with AI, AI fraud detection, an 'AI Powered Banker' and KYC/AML streamlining among them; two-thirds of the benefit from expenses, about 5% of the 2025 expense base; call-centre modernisation and voice AI discussed on the call. | [31][32] |
| Nov 20, 2025 | GitHub Copilot's coding agent to ~800 developers | Half the software developers; fintech partners rewriting legacy systems for the cloud. | [11] |
| Oct 15, 2025 | Q3 2025 deck: agentic AI call centre and agentic infrastructure on the Reimagine list | The deck listed an agentic-AI-enabled call centre, building agentic AI infrastructure, AI-driven pricing, credit and collections, AI-driven fraud, KYC and AML tools, and further upskilling of AI/ML talent, with details promised for January. | [33] |
Does Citizens build its own AI models?
Not large language models. CIO Michael Ruttledge said in September 2024 that 'our plans right now are not to build our own LLM models' and that Citizens uses 'best-in-class models from our technology providers'. It does build its own applications, agents and some machine-learning models, such as new fraud-detection models, and hosts certain models inside its own VPC.
What AI has Citizens built in-house?
A gen-AI portal for employees, a contact-centre product model (2024), the CiZi app assistant (November 2024), the Digital Butler commercial assistant (patented 2024), a cloud fraud platform with new detection models, an AI lending onboarding platform, agentic complaint handling, LLM voice agents, an enterprise AI-chat agent and AI phone support for credit-card calls (October 2026).
Which AI models does Citizens use?
Its gen-AI portal lets employees choose OpenAI, AWS Bedrock, Anthropic or another LLM provider, per American Banker in December 2025. In 2024 the CIO named AWS and Microsoft as the providers it works with to deploy large language models. The bank has not named the model behind CiZi or its voice agents in its own materials.
What is Citizens' gen-AI portal?
An internal gateway where Citizens employees pick among OpenAI, AWS Bedrock, Anthropic and other LLM providers. Ruttledge said it gives the bank 'the freedom to adopt different ones where we think someone's got a leap on the market and we want to get there'. Usage figures for the portal have not been published.
How much of Citizens' call volume will its AI agents handle?
President Brendan Coughlin said in March 2026 that the goal is for 50%-plus of phone calls to be answered by a non-human, with the bank halfway there by the end of 2026. The agents can reorder a debit card, start a fraud claim or read recent transactions, and AWS says in-app voice agents took adoption from 0% to about 60%.
- Citizens sees up to 20% efficiency gains through gen AI
- Inside Citizens' plan to reimagine itself with AI
- Citizens CIO: 'Human in the loop' still key for banks using genAI
- Citizens Enhances Consumer Onboarding Experience, Rolls Out New Digital Features
- Citizens Financial's Michael Ruttledge is #43 on the Most Innovative People in Finance ranking
- Citizens details how AI is speeding everything up
- Citizens on pace for 100% cloud migration by Q3
- Citizens receives patent for its Digital Butler
- Citizens Financial Group 3Q26 Financial Results presentation (Form 8-K, Exhibit 99.2)
- The AI-Enabled Bank: Reimagining Retail Banking from Branch to Mobile
- How Citi, Citizens, TD deploy AI agents without upsetting humans
- Customer Contact Week Sessions On Demand: How Citizens Bank is building the AI-native customer experience
- 2Q26 Earnings Presentation (Form 8-K, Exhibit 99.2)
- Revenue crosses $20 billion mark with resilient growth of 3.1% in FY 26 in constant currency (Client Wins: Citizens AI-first Innovation Hub)
- Citizens (CFG) Q3 2026 Earnings Call Transcript
- CiZi (product page)
- When you're busy, use CiZi
- What Treasurers Need to Know About Payment Fraud In An Increasingly Digital World
- 1Q26 Earnings Presentation (Form 8-K, Exhibit 99.2)
- Citizens Financial Group, Inc. Q1 2026 Earnings Call Summary
- Citizens Financial Group, Inc. Annual Report on Form 10-K for 2025
- Citizens' Coughlin talks 'overwhelming' pace of AI change
- Tech leaders say AI could end humanity. Banks still love it
- Citizens Financial Group, Inc. Quarterly Report on Form 10-Q for June 30, 2026
- Citizens Financial Group Q2 Earnings Call Highlights
- Citizens Bank Mobilizes AI to Target $450 Million Operational Savings by 2028
- Citizens to exit in-store branches, build standalones nearby
- Citizens Financial (CFG) Q2 2026 Earnings Call Transcript
- 2026 Proxy Statement
- 2025 Annual Review
- 4Q25 Earnings Presentation (Form 8-K, Exhibit 99.2)
- Citizens Financial Group Q4 Earnings Call Highlights
- 3Q25 Earnings Presentation (Form 8-K, Exhibit 99.2)
When Citizens 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