Companies using AI in finance: banks, use cases and adoption (2026)
By Paul Okhrem · Updated · Published · Next update: when material evidence changes
Companies using AI in finance include JPMorgan Chase, Bank of America, Goldman Sachs, Morgan Stanley, BlackRock and Mastercard. They use it for fraud checks, document review, wealth research, customer service and software delivery. This reference compares named systems and dated sources. Counts of users, deployments or interactions do not prove a financial return.
This reference describes published company deployments. These companies are not presented as clients of Paul Okhrem's independent consulting.
Industry reference · Last reviewed August 10, 2026
Reading the evidence: Company disclosures establish named deployments, not a sector-wide adoption rate. Survey responses, interaction counts and financial forecasts measure different things. Deloitte's 2024 analysis projects US generative-AI-enabled fraud losses could reach $40 billion by 2027. This is a forecast, not an observed loss total. Read the forecast source.
At a glance
How to read finance AI evidence
Question
What the evidence must show
Limit
How large is the market?
A defined revenue category, geography and period
Vendor revenue estimates do not measure bank adoption.
How common is fraud AI?
A named survey population and a clear definition of use
A company deployment cannot establish sector prevalence.
Is the system in production?
A live workflow, users and operating controls
Tool access and pilot participation are not scaled operation.
Has the investment paid off?
A baseline, measured benefit and full cost
Interactions and time estimates are not enterprise ROI.
What is the forecast fraud risk?
Deloitte projects US losses could reach $40B by 2027.
Banks leading in AI (2026): deployment scale compared
These banks disclose large AI deployments. The table uses figures already covered on this page, with primary sources and dates. It compares selected disclosures, not a complete market ranking. Scale is not value. A large deployment count does not prove a return on investment.
Interaction totals and team adoption have different denominators. Neither measure shows revenue, profit or savings after operating costs. Preserve that distinction when using these examples in a business case.
TL;DR: These examples cover customer service, document review, research, fraud scoring and software work. They include nine financial-services organizations and Walmart as an adjacent retail example. Their measures differ. A deployment announcement does not establish adoption across the bank or a financial return.
Finance AI adoption, by the numbers
89% of respondents said AI helped increase annual revenue and decrease annual costs. NVIDIA's 2026 survey covered more than 800 financial-services professionals. This is self-reported benefit, not audited ROI.
52% reported using or assessing generative AI in 2024, compared with 40% in 2023. NVIDIA's 2025 report labels the comparison years in its workload chart.
Nearly 70% reported AI-related revenue increases of at least 5%; 23% reported increases above 20%. These are respondents in NVIDIA's 2025 report, not its 2026 survey.
As many as 360,000 manual-review hours per year formed JPMorgan's historical comparison for COiN. Its 2016 annual report describes extracting 150 contract attributes in seconds. This is not verified recurring annual savings. Read the original disclosure.
More than 98% of Erica users found the information they needed, according to Bank of America in August 2025. That measure does not mean every inquiry avoided human support.
BlackRock offers Aladdin Copilot for investment-platform assistance. Platform assets are not an AI adoption or return metric. Product source.
Goldman Sachs describes GS AI Assistant and developer copilots in its annual report. Availability does not establish active use or realized savings. Company source.
42% of surveyed issuers reported saving over $5 million in fraud attempts over two years using AI. Separately, 83% of surveyed industry leaders reported reduced false positives and customer churn. These are Mastercard's 2025 research findings, not product-specific results.
Wells Fargo's Fargo supports customer banking tasks. This example describes a banking assistant, not a verified staffing reduction. Product source.
98% of financial advisor teams had adopted AI @ Morgan Stanley Assistant by June 2024. OpenAI separately reports document access rising from 20% to 80%. Company disclosure; supplier case study.
The 2026 state of AI in finance
The named systems show several ways to put AI into financial work. Erica supports customer questions. Morgan Stanley's Assistant retrieves internal research. Aladdin Copilot helps users navigate an investment platform. These applications have different users, tasks and controls.
JPMorgan's machine-learning system for extracting information from commercial credit agreements. The original annual report compares extraction time with the prior manual workload. Source 17.
Aladdin (BlackRock)
An investment-management technology platform. Aladdin Copilot adds a natural-language interface for questions and support within that platform. Source 13.
Erica (Bank of America)
Bank of America's virtual financial assistant, launched in 2018. Its disclosed interaction totals describe customer activity, not staffing savings. Source 4.
GS AI Assistant
Goldman Sachs's generative AI assistant for employees. The firm's annual report also describes coding assistants for developers. Source 8.
AI @ Morgan Stanley
Tools for wealth-management research and meeting support. Debrief prepares notes and draft emails for advisor review, with client consent. Source 11.
Decision Intelligence Pro (Mastercard)
Mastercard's AI-based transaction-risk technology. Its product announcement is separate from Mastercard's survey of payment-fraud practices. Source 5.
Fargo (Wells Fargo)
Wells Fargo's customer assistant for banking questions, account information and selected tasks. Source 12.
Devin (Cognition)
Cognition's AI software-engineering agent. Cognition names Goldman Sachs and Citi as partners. That statement does not establish deployment-wide financial returns. Source 9.
TL;DR: The entries below retain named AI applications and their primary sources. Some report usage or infrastructure outcomes; others establish the application only. Walmart is a retail example, not evidence of an AI banking deployment.
Companies using AI in finance (2026)
1. JPMorgan Chase: contract intelligence and employee tools
COiN extracts information from commercial credit agreements. JPMorgan's 2016 annual report compares this process with the old manual-review workload. It is a historical comparison, not a current annual savings measure. Original COiN disclosure.
JPMorgan's 2024 annual report also describes LLM Suite, an employee tool for writing and summarizing. Company report.
Why it matters: Separate a task-time comparison from verified savings after review and operating costs.
2. Bank of America: Erica at billion-interaction scale
Erica passed 3 billion client interactions by August 2025, with more than 58 million monthly interactions. The disclosure also reported nearly 50 million users since launch. More than 98% of users found the information they needed. Dated company disclosure.
These figures show activity and information access. They do not establish a staffing reduction, cash saving or handoff-free resolution rate.
Why it matters: A useful service measure still needs a separate cost and quality baseline.
3. Goldman Sachs: employee assistants and coding tools
Goldman Sachs's annual report describes GS AI Assistant and developer copilots. Company report. Cognition also names Goldman Sachs as a partner for its Devin software-engineering agent. Supplier disclosure.
The sources establish specific tools, but they do not provide a common measure of realized developer productivity.
Why it matters: Access, accepted code and released software are different measures of adoption and value.
4. Morgan Stanley: research access for wealth advisors
AI @ Morgan Stanley Assistant supports access to internal research. Morgan Stanley reported 98% adoption among financial advisor teams in June 2024. Company disclosure. OpenAI separately describes document access rising from 20% to 80%. Supplier case study.
Debrief creates meeting notes with client consent and drafts follow-up emails. Advisors can review and edit those emails before sending them.
Why it matters: Retrieval and draft preparation can support advisors while keeping client decisions with people.
5. BlackRock: AI assistance inside Aladdin
Aladdin Copilot provides a natural-language interface within BlackRock's investment-management platform. The product supports user questions and access to platform knowledge. BlackRock product description.
The availability of a copilot does not establish that every client uses it. Assets on a platform are not assets managed autonomously by AI.
Why it matters: Measure use of the AI feature separately from the size of its host platform.
6. Citigroup: internal knowledge and document tools
Citi describes Citi Assist for policy questions and Citi Stylus for document insights. Stylus Workspaces supports document drafting and summaries. Citi's July 2025 disclosure. Cognition separately names Citi as a partner. Supplier disclosure.
These applications cover different workflows. A bank's total technology budget does not identify spending on these AI tools.
Why it matters: Track adoption, review effort and cost for each workflow, not one combined tool count.
7. Mastercard: transaction-risk scoring
Decision Intelligence Pro assesses relationships around a transaction to help determine fraud risk. Mastercard product announcement.
Mastercard's separate payment-fraud study reports AI benefits from surveyed issuers and industry leaders. The figures in the statistics section are survey findings, not measured outcomes from Decision Intelligence Pro alone. The same research reported average payment-fraud losses of $60 million per surveyed organization in the previous year. Survey source.
Why it matters: Review fraud losses and false declines together. A blocked legitimate purchase also has a cost.
8. Wells Fargo: Fargo customer assistance
Fargo helps Wells Fargo customers find account information and complete selected banking tasks. Wells Fargo's product page describes the assistant and its features. Fargo product source.
For a comparable deployment, test sensitive-data handling, incorrect answers and access to human help before expanding the service.
Why it matters: Service scope and escalation rules matter alongside the model behind the assistant.
9. PayPal: AI infrastructure efficiency
NVIDIA's 2025 report describes PayPal updating its AI infrastructure for data processing and analytics. It reports cloud-cost reductions of up to 70% and a 35% runtime reduction. Supplier-reported case.
The report does not establish a company-wide cost reduction or isolate a financial return from fraud models.
Why it matters: Record workload, infrastructure and measurement conditions before transferring an efficiency claim to another business case.
10. Walmart: an adjacent retail example
Walmart launched Sparky in June 2025 for shopping questions, review summaries and recommendations. Company launch announcement. That announcement distinguishes available shopping features from planned capabilities.
This is an adjacent commerce example. It does not establish that Sparky handles credit decisions or integrates with OnePay.
Why it matters: Do not turn a shopping-assistant launch into evidence of an autonomous financial service.
Comparison: disclosed AI applications
Company
AI application
Function
Evidence and limit
Source
JPMorgan Chase
COiN; LLM Suite
Document review; employee assistance
Historical manual-work comparison; separate employee-tool disclosure
The company entries above link each application to a primary source. Absence from this graphic does not mean a company lacks a particular use case.
TL;DR: Compare AI by the task it performs. Customer support, fraud scoring and research access need different success measures. The examples below do not establish a sector-wide ROI ranking.
AI use cases in finance by function
1. Fraud detection and payments risk
Mastercard's transaction-risk technology evaluates relationships around a payment. A comparable project should track detected fraud, false declines and review effort. Its value depends on both preventing losses and accepting legitimate transactions. The company entry separates this product from Mastercard's broader issuer survey.
2. Customer service and virtual assistants
Erica and Fargo illustrate customer banking assistants. Interaction volume shows activity, but it does not establish successful resolution or fewer staff. Measure incorrect answers, repeat contacts and access to human help alongside customer use.
3. Wealth advisory and knowledge access
Morgan Stanley's Assistant supports research access. Debrief prepares meeting notes and draft follow-up emails for advisor review. Evaluate source accuracy, retrieval coverage and consent handling. Faster access is a different measure from better investment advice.
4. Software development and engineering
Goldman Sachs describes developer copilots, while Cognition names Goldman Sachs and Citi as partners. A coding-agent evaluation needs accepted changes, test results and review effort. Count released work, not generated code alone. The enterprise AI agent benchmarks provide broader context.
5. Document automation and legal review
COiN and Citi Stylus illustrate different document tasks. Extracting contract fields is not the same as drafting a summary. Define required fields, exceptions and reviewer responsibility before testing either workflow. Retain the historical scope of the COiN comparison.
6. Trading, research and portfolio analysis
Aladdin Copilot adds assistance within an investment platform. That does not establish autonomous portfolio control or investment performance. Evaluate the source, timeliness and permissions of retrieved information before using it in a decision.
7. Internal knowledge and productivity
Citi Assist and employee-facing tools illustrate internal knowledge support. Define which policies and documents the system can access. Track whether staff can verify an answer and find its source. Access to a tool alone is not successful adoption.
TL;DR: A survey can describe reported use or benefit. A company disclosure can establish a named application. Neither replaces a measured baseline for your own workflow.
Adoption and trends in financial services AI (2026)
NVIDIA's 2025 and 2026 reports use different survey periods and questions. Keep the 2023 and 2024 workload comparison separate from the 2026 benefit finding. The former includes assessment; the latter reports respondents' views of revenue and costs.
In the 2025 report, 98% of management respondents planned to increase AI infrastructure spending. This measures intent, not actual spending. Survey source.
Employee assistants and tools that execute software tasks also need different controls. An assistant may return a draft. An agent may change a system. Record its permissions, approval points and recovery process before expanding access.
For cross-industry patterns, see the enterprise AI agent statistics. Compare study definitions before drawing conclusions across industries.
What this means for finance leaders
TL;DR: Start with a defined workflow, its owner and the evidence needed to continue. A large deployment elsewhere is not an investment case for your bank.
Choose a task with a measurable baseline. Record current cost, quality, exceptions and review effort. Then define the evidence that would justify a controlled trial or stop the work.
Keep activity, adoption and financial return separate in board reporting. An interaction count can show demand. A usage measure can show whether staff engage. Neither proves a change in profit after implementation and operating costs.
Customer-service tools are not inherently less valuable than internal systems. The relevant question is whether the chosen workflow improves under acceptable controls.
Questions about AI use in finance
These answers explain the named systems, reported deployment measures and limits on comparing financial institutions.
What is COIN at JPMorgan?
COiN is JPMorgan's machine-learning system for extracting information from commercial credit agreements. Its 2016 annual report compares extraction in seconds with the prior manual workload. The historical 360,000-hour figure is not verified recurring annual savings. Use the original disclosure when quoting it.
Is BlackRock Aladdin an AI?
Aladdin is an investment-management technology platform, not one standalone AI model. Aladdin Copilot adds natural-language assistance within that platform. A product's availability does not establish adoption by every client. Platform assets also do not measure assets managed autonomously by AI.
Does Bank of America use ChatGPT?
Erica is Bank of America's virtual assistant, launched in 2018. It should not be described as ChatGPT. The Erica disclosure on this page does not establish bank-wide use of OpenAI products. Verify each tool and its source separately before making that claim.
Which bank has the most AI use cases in production?
The disclosures collected here do not support a comparable ranking. Banks can count a model, application or workflow as a use case. They also report different periods and deployment stages. Compare the same unit and definition before naming a leader.
What is Devin AI?
Devin is Cognition's AI software-engineering agent for multi-step development work. Cognition names Goldman Sachs and Citi as partners. That supplier statement establishes a relationship, not a verified productivity multiplier. It does not quantify the total share of either bank's engineering work handled by agents.
How does AI detect financial fraud?
AI can assess patterns around a transaction and return a risk score. Mastercard's product disclosure describes this application. Evaluation should include false declines as well as detected fraud. Mastercard's broader issuer survey must not be presented as a result from one product.
How much does JPMorgan spend on AI?
The sources used here do not provide a comparable, standalone AI spending total. Do not treat a technology budget as an AI budget. A useful estimate must separate development, software, infrastructure, data, security and review costs over a defined period.
Which banks use OpenAI?
Morgan Stanley's company disclosure and OpenAI's case study document that relationship. The Assistant supports internal research access, while Debrief supports reviewed meeting follow-up. These primary sources do not establish a complete list of OpenAI's financial-services customers or an exclusive relationship.
What is agentic AI in banking?
Agentic AI can plan and execute multiple steps through tools. Its allowed actions and oversight depend on the workflow. Coding agents provide one example in the sources here. A banking deployment should define permissions, approval points, logs, escalation and recovery before allowing system changes.
What percentage of banks use AI in 2026?
This page does not establish a census of banks using AI. NVIDIA surveys financial-services professionals across a broader sector. Its benefit finding is not an adoption rate. Keep the original respondent population, survey question and year attached to any percentage.
What was AI's biggest impact in finance in 2025?
The collected disclosures cover customer assistance, research access, document work and software tools. They do not share one comparable outcome measure, so they cannot identify the biggest impact across companies. An interaction total, team-adoption rate and workload-cost change answer different questions.
What jobs in finance is AI replacing?
The examples cover contract review, customer service, document drafting and software work. Reported task automation does not by itself show job replacement. Check whether a disclosure measures time, staffing, headcount or service capacity. Company case studies also need evidence about human review and work that moved elsewhere.
Frequently asked questions
Which financial services companies use AI the most in 2026?
The examples include JPMorgan Chase, Bank of America, Goldman Sachs, Morgan Stanley, BlackRock and Mastercard. Their disclosures measure different things, including activity, adoption and product availability. They cannot establish which firm uses AI most. Compare the same function, period and denominator.
What percentage of financial services firms use AI?
The cited NVIDIA reports survey industry professionals rather than every financial-services firm. The workload comparison includes use or assessment, while the benefit question asks about revenue and costs. Those separate results should not be combined into a single production-adoption percentage.
How does JPMorgan Chase use AI?
JPMorgan's annual reports describe COiN for commercial-credit document work and LLM Suite for employee assistance. The COiN comparison is historical. The employee-tool disclosure establishes another application, not a bank-wide financial return. The company entry above links both original annual reports.
What is the AI strategy of large banks in 2026?
The examples combine employee assistance with defined operational tasks such as customer support and document review. They do not establish one strategy shared by all large banks. A useful comparison examines the workflow, owner, data access, review process and acceptance criteria.
How is AI used in fraud detection?
AI-based transaction scoring helps identify suspicious patterns before a payment is approved. Mastercard describes this use in its product announcement. Its separate survey reports respondents' experience with AI across payment-fraud prevention. Neither source proves that all banks use the same system.
Are AI assistants like Bank of America's Erica actually used?
Yes. Bank of America publishes cumulative and monthly interaction figures for Erica, as shown in the comparison above. Its information-finding measure describes users finding what they need. It does not establish cash savings or staffing reductions. It also does not prove that every inquiry avoided human support.
How much do banks spend on AI?
The sources here do not establish a comparable AI spending total across banks. Overall technology and transformation budgets can overlap and include unrelated work. Compare defined AI costs over the same period. Include implementation, licenses, infrastructure, human review and ongoing operations.
Which use cases of AI in finance have the highest ROI?
There is no comparable ROI ranking across the company disclosures on this page. They mix time saved, adoption, interactions and reported productivity. To compare financial returns, use a shared measurement period and count implementation, review and operating costs. A bank's measured baseline matters more than another firm's deployment scale.
Finance AI sources and methods
This reference includes company disclosures, supplier case studies and industry surveys. Each source type answers a different question. Named products establish an application; survey findings reflect their respondents. Forecasts and historical comparisons are labeled separately. No cross-company performance ranking is inferred.
Use the Proof Standard to check a baseline, intervention and measurement window before treating a claim as ROI.
Mastercard and FT Longitude. Payment-fraud research summary. Published February 2026, based on 2025 research. Issuer and industry-leader denominators differ.
This page is published under CC BY 4.0. Free to cite with attribution.
Citation
Okhrem, P. (2026). Companies Using AI in Finance: 2026 Reference.
paul-okhrem.com/companies-using-ai-in-finance/
Embed a stat with attribution
JPMorgan's 2016 annual report compared COiN with manual review
that could require as many as 360,000 hours per year.
Historical workload comparison, not verified recurring savings.
Source: JPMorgan Chase; compiled by Okhrem (2026),
paul-okhrem.com/companies-using-ai-in-finance/.
When quoting a figure, retain the original source, year and measure alongside this page's citation.
Who compiled this finance AI reference
Compiled by Paul Okhrem, AI transformation consultant and fractional Chief AI Officer. Co-Founder and CEO of Elogic Commerce; Managing Partner at Uvik Software since June 2017. Paul compiles this research as part of his fractional CAIO work. He advises CEOs and boards on AI and agentic AI decisions.
Suggested citation: Okhrem, P. (2026). Companies using AI in finance: banks, use cases and adoption. paul-okhrem.com/companies-using-ai-in-finance/. Free to cite under CC BY 4.0.
Related
For sector-wide figures with sources, see the AI in financial services statistics. Check whether each figure measures adoption, deployment or a financial outcome.
Change log
: Reviewed service scope, decision guidance and supporting resources for the October release.
: Reviewed service scope, source evidence, links and buyer answers for the October release.
: Clarified dense passages while preserving source limits and commercial terms.
Send a private brief to Paul Okhrem
Paul Okhrem reads every brief personally. Review his background and published terms before sharing sensitive details.
What happens next
Send the brief: the decision, its owner and the deadline.
Paul replies within two business days: yes, questions, or a no with a referral when possible.
Scope, an NDA where needed, conflicts, availability and terms are agreed in writing before work starts.
