Which banks, insurers and investment firms are furthest along with AI?

In the Enterprise AI diffusion report, Klarna is at Embedded, Bank of America, DBS Bank, JPMorgan Chase, and Lemonade are at Scaling, BNY, Citigroup, Goldman Sachs, Travelers, and Wells Fargo are at Piloting, and AIG, Allstate, Berkshire Hathaway, Capital One, Chubb, MetLife, Morgan Stanley, and Progressive are at Experimenting. A stage follows from the returns and rollouts a company has publicly reported in the last 18 months, not from its spending or plans.

What return have banks and insurers reported from AI?

Banks and investment firms report AI returns most often in software engineering, where JPMorgan Chase, Wells Fargo, Goldman Sachs, Bank of America and Citigroup each credit AI with a measured gain between Feb 2026 and Oct 2026. Beyond engineering, Klarna credits its AI assistant with about $59 million of customer service savings in 2025 (Feb 2026), and Bank of America credits AI-enabled processes with a 2025 consumer fraud loss rate about 50% below 2018 (Nov 2025). JPMorgan Chase puts AI's savings at about $2 billion a year (Oct 2025) and DBS Bank credits classic AI and machine learning with about SGD 1 billion of economic value in 2025. Among insurers, Lemonade alone credits AI with a return, cutting the cost of handling a Pet claim from $44 in 2021 to $14 in 2025 (Feb 2026); AIG, Allstate, Chubb, MetLife, Progressive and Travelers report no return they credit to AI alone.

Do banks and insurers build or buy their AI platforms?

Banks and insurers mostly build their AI platforms and buy the models that run on them. JPMorgan Chase built LLM Suite, BNY built Eliza, Goldman Sachs built the GS AI Platform, Citigroup built Citi AI, DBS Bank built DBS-GPT and Bank of America built Erica in-house; Wells Fargo pairs its own multi-model platform with Microsoft Copilot for the wider workforce. Among insurers, Travelers built TravAI, MetLife built MetIQ, Allstate is building ALLIE, Lemonade runs on its own Lemonade OS, and AIG built Underwriting by AIG Assist with Palantir and Anthropic. Berkshire Hathaway has no group-wide platform and leaves AI to subsidiaries such as BNSF and GEICO.

Which AI models do banks and insurers use?

Banks and insurers use several model providers at once. JPMorgan Chase puts OpenAI and Anthropic behind LLM Suite alongside open-weight models, Goldman Sachs adds Google to the same mix, BNY routes OpenAI, Google, Anthropic and open-source models by task, Wells Fargo uses Google Gemini, Microsoft Copilot, OpenAI and Meta Llama, and Citigroup uses Google Gemini, Cognition's Devin and Microsoft Copilot. Capital One customizes open-weight models such as Meta Llama, Morgan Stanley uses OpenAI GPT models, and Klarna runs on OpenAI. Among insurers, Travelers uses Anthropic Claude and OpenAI plus its own TravelersLLM, AIG uses Anthropic Claude, and Lemonade used OpenAI for its Maya voice agent in 2025; Chubb, MetLife and Progressive have not named their model providers.

How banks, insurers and investment firms deploy AI

Banks, insurers and payments companies get their clearest AI returns where AI takes over customer service. Bank of America's Erica does work that would otherwise take 10,000 to 12,000 staff, and Klarna's assistant did the work of more than 850 agents, though Klarna has since rebuilt a human service tier, recruiting agents from its own customers, after its CEO said quality had suffered. In engineering, banks take AI's gains as more output rather than fewer engineers: Wells Fargo engineers write 35% more code, and Citigroup cut tech vendor spend 30% by giving vendors' work to an AI coding agent.

Verified Nov 5, 2026See methodology

How banks, insurers and investment firms have deployed AI

Findings

What the financial services profiles show

Software engineering is where banks and investment firms most often report a return they credit to AI: JPMorgan Chase reports coding efficiency up more than 10% in 2025 (Feb 2026), Wells Fargo 35% more code from engineers using generative AI (Mar 2026), Goldman Sachs at least 20% engineering efficiency from AI coding tools (Sep 2026), Bank of America technology development 10% more efficient (Oct 2026), and Citigroup tech vendor spend down 30% after in-sourcing work with the Devin coding agent (Oct 2026).
Insurers mostly credit their gains to AI together with automation and other technology, so the gains do not count as AI returns: Travelers credited the one-third fall in claim call center staff it cited in Jan 2026 to automation, straight-through processing and analytics together, MetLife credited its 11.7% direct expense ratio for 2025 to AI and other emerging technologies (Feb 2026), and Chubb's targets of about 20% fewer staff and about 1.5 combined ratio points of savings (Dec 2025) cover AI with technology, data and process redesign. Lemonade is the one insurer with a return it credits to AI: the cost of handling a Pet claim fell from $44 in 2021 to $14 in 2025 (Feb 2026).
Several large financial firms decline, for now, to put a figure on AI: Capital One's Richard Fairbank said on Oct 20, 2026 that Capital One does not intend to quantify AI's benefits, BNY's Robin Vince said in Oct 2026 that BNY does not break AI out of its expense base, JPMorgan Chase CFO Jeremy Barnum said in Oct 2026 that he is moving away from a dollar figure, Progressive's Tricia Griffith said in Aug 2026 that she would put dollars to AI later, and AIG's Peter Zaffino has said clarity on AI's costs and revenue benefits may come in 2027 or 2028.
Banks and insurers build their own AI platform and buy the models behind it: LLM Suite at JPMorgan Chase, Eliza at BNY, the GS AI Platform at Goldman Sachs, Citi AI at Citigroup, DBS-GPT at DBS Bank, TravAI at Travelers, MetIQ at MetLife, ALLIE at Allstate and Lemonade OS at Lemonade are in-house builds. Berkshire Hathaway is the exception, with no group-wide AI platform: BNSF and GEICO build their own tools, and no executive owns AI across the group.
Bank AI rollouts reach most of the workforce, while most insurers publish no user count: Bank of America counts more than 200,000 employees using AI tools (Oct 2026), JPMorgan Chase about 150,000 weekly LLM Suite users (Feb 2026), Citigroup adoption above 87% (Sep 2026), DBS Bank more than 70% of employees (Aug 2026) and BNY about half of staff using AI daily (Apr 2026). Among insurers only Travelers reports a firmwide count, more than 20,000 regular AI users (Jan 2026); AIG, Chubb, MetLife and Progressive have not said how many employees use their AI tools.

Stages

Where banks, insurers and investment firms stand

Embedded
Klarna
Embedded means the company reports AI's effect on the whole business, in cost, headcount, revenue or margin, and most employees use AI every week in redesigned workflows.
Scaling
Bank of America, DBS Bank, JPMorgan Chase, Lemonade
Scaling means the company reports returns it credits to AI in several functions, or a firm-level AI figure, and runs a company-wide AI platform with a named owner.
Piloting
BNY, Citigroup, Goldman Sachs, Travelers, Wells Fargo
Piloting means the company reports a business return it credits to AI in at least one use case, or a company-wide AI rollout with a user count and a named owner.
Experimenting
AIG, Allstate, Berkshire Hathaway, Capital One, Chubb, MetLife, Morgan Stanley, Progressive
Experimenting means the company has reported neither a business return it credits to AI nor a company-wide AI rollout with a user count. Pilots, partnerships, plans and spending alone keep a company at Experimenting.

Returns

The returns banks, insurers and investment firms credit to AI

Only realized returns the company reported and credited to AI itself, with their dates: revenue, cost, headcount, losses avoided or output per employee.

Company	Use case	Return
Klarna	AI assistant for customer chatsCustomer service	About $59 million in cost savings in 2025, the work of over 850 full-time agents (Feb 2026)
Klarna	AI-made marketing contentMarketing	$10 million saved in 2024 from AI-powered marketing, by Klarna's estimate (Feb 2026)Source
Klarna	Chat log classificationEngineering	About $4.9 million in cost savings in the 12 months to Jun 2025 (Sep 2025)
Bank of America	Erica virtual assistantConsumer Banking	Handles work that would otherwise take 10,000 to 12,000 staff, per Brian Moynihan (Sep 2026)Source
Bank of America	AI fraud detectionConsumer Banking	Consumer fraud loss rate ~50% below 2018, credited to AI-enabled processes (Nov 2025)Source
Bank of America	AI coding assistanceGlobal Technology	Tech development ($4B+ a year) 10% more efficient with AI: more work for the same cost (Oct 2026)Source
JPMorgan Chase	AI coding assistantsGlobal Technology	Coding efficiency up more than 10% from AI in 2025 (Feb 2026)Source
Lemonade	Pet claims and AI medical records reviewPet	Cost of handling a Pet claim down from $44 in 2021 to $14 in 2025, credited to AI (Feb 2026)Source
Citigroup	Devin agentic coding and in-sourcingTechnology	Tech vendor spend down 30% year over year, credited to in-sourcing work with Devin (Oct 2026)
Goldman Sachs	AI coding assistantsEngineering	Engineering efficiency up at least 20% from AI coding tools, per CIO Marco Argenti (Sep 2026)Source
Wells Fargo	AI coding tools and autonomous coding agentsTechnology	35% more code written by engineers using GenAI, with no cut in resources (Mar 2026)Source

Compare

Financial services side by side

Company	Ownership	Platform	Models	Hosting	Reach	Stage
Klarna	CEO Sebastian Siemiatkowski; no chief AI officer disclosed	Hybrid: Assistant and Kiki built in-house on OpenAI; ChatGPT Enterprise bought	OpenAI, plus Google media models and ElevenLabs voice	AWS as primary cloud; Google Cloud trains fraud models; generative AI runs through OpenAI and ElevenLabs services	Vast majority use generative AI daily (Feb 2026); last count 96% (Aug 2024)	Embedded
Bank of America	CTO and CIO Hari Gopalkrishnan builds; Jeff Busconi drives usage as AI catalyst	Hybrid: Erica and generative AI tools built in-house; Copilot and Agentforce bought	Open-source and proprietary models, providers not named; AI in vendor tools	Cloud hosting environments the bank does not name; customer data kept out of outside models	About 200,000 active AI users (Jul 2026)	Scaling
DBS Bank	CEO Tan Su Shan; COO Derrick Goh runs Transformation & Data; CIO Eugene Huang runs technology	Hybrid: ADA, DBS-GPT and the agent platform built in-house; Microsoft Copilot bought	Model-agnostic: US and Chinese models, providers not named	Own primary and secondary data centres plus a hybrid multi-cloud build-out; DBS has not named the clouds that run DBS-GPT or its agents	More than 70% of employees use AI tools (Aug 2026); no weekly figure published	Scaling
JPMorgan Chase	CTO Scot Baldry, under Global CIO Lori Beer; becomes CDAO at the end of 2026	Hybrid: LLM Suite built in-house; coding tools bought	OpenAI and Anthropic behind one portal, plus open-weight models	Own data centers and cloud providers, AWS and Microsoft Azure included; bank data barred from model training	About 150,000 weekly users (Feb 2026)	Scaling
Lemonade	Co-founders: CEO Daniel Schreiber and President Shai Wininger, a former CTO; no chief AI officer disclosed	Built: Lemonade OS: AI Maya, AI Jim, CX.AI, Blender and Cooper, built in-house	Model agnostic: own pricing and LTV models plus frontier LLMs; OpenAI for Maya's voice (2025)	AWS, Lemonade's sole third-party data center provider, runs its site and app; OpenAI's API served Maya's voice (Apr 2025); other model hosting undisclosed	98% of policies sold by AI Maya and APIs (Dec 2025)	Scaling
BNY	Chief Data and AI Officer Sarthak Pattanaik; CIO Leigh-Ann Russell reports to the CEO	Hybrid: Eliza built in-house on Azure; Copilot tools bought	OpenAI, Google, Anthropic and open-source, routed by task	Microsoft Azure and Microsoft Foundry under Eliza, Google Cloud for Gemini Enterprise, and BNY's own on-premises NVIDIA DGX SuperPOD	All employees have access; about 50% daily users (Apr 2026)	Piloting
Citigroup	Group Head of AI David Griffiths, in COO Anand Selvakesari's office	Hybrid: Citi AI suite and Arc built in-house; Devin and Copilot Chat bought	Google Gemini, Cognition's Devin and Microsoft Copilot; Stylus models not disclosed	Google Cloud (Vertex AI, Gemini Enterprise) and Microsoft 365; where Stylus models run is not disclosed	About 180,000 with access, adoption above 87% (Sep to Oct 2026)	Piloting
Goldman Sachs	CIO Marco Argenti, on the Management Committee; OneGS 3.0 developed by President John Waldron	Hybrid: GS AI Platform built in-house; coding tools and agents bought	OpenAI, Google and Anthropic behind one gateway, plus open-weight models	Behind the firm's firewall; GenAI Platform team works on AWS; the clouds and data centers serving the models are not disclosed	About 40,000 weekly users of 46,200 staff (Sep 2026)	Piloting
Travelers	Mojgan Lefebvre, EVP and Chief Technology & Operations Officer	Hybrid: TravAI and TravelersLLM built in-house; Claude and OpenAI bought	Anthropic Claude and OpenAI, plus the in-house TravelersLLM	About 70% of compute in the cloud (Aug 2026), plus its own primary data processing facility; AWS Bedrock and Google Cloud in disclosed projects; TravAI and TravelersLLM hosts unpublished	20,000+ regular AI users (Jan 2026); all 30,000+ staff can use TravAI	Piloting
Wells Fargo	Saul Van Beurden, Head of AI, on the Operating Committee, reporting to the CEO	Hybrid: Internal multi-model platform built; Microsoft Copilot and Google Agentspace bought	Google Gemini, Microsoft Copilot, OpenAI as needed and Meta Llama, chosen by task	Microsoft Azure, Google Cloud and private cloud in new third-party-owned data centers; personal data stripped before Fargo calls Gemini	About 190,000 with Copilot Chat access; about 40,000 use Microsoft 365 Copilot (Mar 2026)	Piloting
AIG	EVP and Chief Digital Officer Scott Hallworth, alongside Chief Technology and Digital Officer Roshan Navagamuwa	Hybrid: AIG Assist, built in-house on Palantir and Anthropic	Anthropic Claude, with AWS Titan and Textract for document work	AWS (Bedrock, Titan, Textract) and Palantir's Foundry and AIP; Google in Japan; no own-data-center inference disclosed	Underwriters in 8 lines of business (May 2026)	Experimenting
Allstate	CIO Zulfi Jeevanjee leads the ALLIE build; CFO Chris Lown owns data and analytics; CEO Tom Wilson sets AI strategy	Hybrid: ALLIE and claims copilot built in-house; Copilot and roof imagery bought	Not named; internal LLMs only (Aug 2026); OpenAI GPT and Microsoft Copilot reported in 2025	Internal LLMs only (Aug 2026); a Jul 2026 job posting names Azure, AWS and Google Cloud AI services; production clouds and data centers not disclosed	No firmwide count; 14,000 claims investigators on a custom copilot (Jan 2025)	Experimenting
Berkshire Hathaway	No group AI chief; subsidiary CEOs own AI, with brk tech (Hari Govind) as shared help	Hybrid: No group platform; subsidiaries build in-house and buy point tools	Not disclosed group-wide; Claude reported in GEICO engineering, in-house models at BNSF	GEICO: private cloud in 2 colocation data centers plus public cloud, reported as mostly Microsoft Azure; brk tech: Azure and Oracle Cloud; others undisclosed	Not disclosed; no group-wide employee AI tool	Experimenting
Capital One	Prem Natarajan, Chief Enterprise AI Officer, within Frank LaPrade's Enterprise Services	Built: Proprietary multi-agent framework built in-house	Customized and distilled open-weight models such as Meta Llama; Anthropic Claude for code security	Principally AWS, with some Azure and Google Cloud resources; Discover's network partly in data centers; inference location not published	AI learning hub launched to 60,000+ associates; tool use not disclosed	Experimenting
Chubb	Chief Scientist Sean Ringsted (AI strategy); EVP Sesh Iyer (operations, technology, transformation)	Built: Enterprise AI platform and Chubb Studio, built in-house	Predictive ML plus open and closed LLMs; providers not named	Own data centers and cloud infrastructure; cloud providers not disclosed	Not disclosed; program to touch 70% of the organization by about 2028	Experimenting
MetLife	Bill Pappas, EVP and Head of Global Technology and Operations, under a CEO-led framework	Hybrid: MetIQ built in-house; foundation models and staff tools bought	Foundation models not named; Microsoft Copilot, Claude Cowork and Cursor for staff	Not disclosed; MetIQ is cloud enabled inside MetLife's secure environment; a 2025 job posting cites Microsoft Azure	Not disclosed; role-appropriate AI tools for employees (2025)	Experimenting
Morgan Stanley	Firmwide AI team; Co-President Andy Saperstein co-led AI strategy in 2025	Hybrid: Tools built in-house on OpenAI models	OpenAI GPT models; Anthropic Claude Mythos in beta	Not disclosed for its OpenAI models; Microsoft Azure is a cloud partner since 2021 in a multi-cloud setup, and Morgan Stanley runs its own data centers	About 98% of employees with access (Jun 2026)	Experimenting
Progressive	Chief Strategy Officer Dan Witalec and the CTO manage AI together under CEO Tricia Griffith (Aug 2026); CIO Steve Broz runs IT	Hybrid: Vision and pricing models built in-house; Azure AI and vendor tools bought	Own supervised vision and predictive models; LLM vendors not disclosed	Own data centers and unnamed third-party clouds; Flo chatbot built on Microsoft Azure; generative AI hosting not disclosed	Not disclosed; about 70,000 employees (Dec 2025)	Experimenting

Questions

AI in financial services, answered

Which banks, insurers and investment firms are furthest along with AI?: In the Enterprise AI diffusion report, Klarna is at Embedded, Bank of America, DBS Bank, JPMorgan Chase, and Lemonade are at Scaling, BNY, Citigroup, Goldman Sachs, Travelers, and Wells Fargo are at Piloting, and AIG, Allstate, Berkshire Hathaway, Capital One, Chubb, MetLife, Morgan Stanley, and Progressive are at Experimenting. A stage follows from the returns and rollouts a company has publicly reported in the last 18 months, not from its spending or plans.
What return have banks and insurers reported from AI?: Banks and investment firms report AI returns most often in software engineering, where JPMorgan Chase, Wells Fargo, Goldman Sachs, Bank of America and Citigroup each credit AI with a measured gain between Feb 2026 and Oct 2026. Beyond engineering, Klarna credits its AI assistant with about $59 million of customer service savings in 2025 (Feb 2026), and Bank of America credits AI-enabled processes with a 2025 consumer fraud loss rate about 50% below 2018 (Nov 2025). JPMorgan Chase puts AI's savings at about $2 billion a year (Oct 2025) and DBS Bank credits classic AI and machine learning with about SGD 1 billion of economic value in 2025. Among insurers, Lemonade alone credits AI with a return, cutting the cost of handling a Pet claim from $44 in 2021 to $14 in 2025 (Feb 2026); AIG, Allstate, Chubb, MetLife, Progressive and Travelers report no return they credit to AI alone.
Do banks and insurers build or buy their AI platforms?: Banks and insurers mostly build their AI platforms and buy the models that run on them. JPMorgan Chase built LLM Suite, BNY built Eliza, Goldman Sachs built the GS AI Platform, Citigroup built Citi AI, DBS Bank built DBS-GPT and Bank of America built Erica in-house; Wells Fargo pairs its own multi-model platform with Microsoft Copilot for the wider workforce. Among insurers, Travelers built TravAI, MetLife built MetIQ, Allstate is building ALLIE, Lemonade runs on its own Lemonade OS, and AIG built Underwriting by AIG Assist with Palantir and Anthropic. Berkshire Hathaway has no group-wide platform and leaves AI to subsidiaries such as BNSF and GEICO.
Which AI models do banks and insurers use?: Banks and insurers use several model providers at once. JPMorgan Chase puts OpenAI and Anthropic behind LLM Suite alongside open-weight models, Goldman Sachs adds Google to the same mix, BNY routes OpenAI, Google, Anthropic and open-source models by task, Wells Fargo uses Google Gemini, Microsoft Copilot, OpenAI and Meta Llama, and Citigroup uses Google Gemini, Cognition's Devin and Microsoft Copilot. Capital One customizes open-weight models such as Meta Llama, Morgan Stanley uses OpenAI GPT models, and Klarna runs on OpenAI. Among insurers, Travelers uses Anthropic Claude and OpenAI plus its own TravelersLLM, AIG uses Anthropic Claude, and Lemonade used OpenAI for its Maya voice agent in 2025; Chubb, MetLife and Progressive have not named their model providers.

Other sectors

Every company in the report