AI in banking statistics: 41 figures on banks and lenders, traced to the source
For writers who need a number they can quote, and for lending leaders planning an AI project. Each statistic is numbered, dated and linked to the regulator, survey or bank that published it.
Key takeaways
- 92% of EU banks deploy AI, says the European Banking Authority (EBA). The European Central Bank (ECB) says more than 85% of the large banks it supervises use AI.
- US numbers depend on what each survey asked. 49% of US banks and 59% of credit unions surveyed have deployed generative AI (Cornerstone Advisors, 2026). Only 13% of community lenders use AI in credit and lending (Cornerstone Advisors for Zest AI, 2024).
- In loan work, lenders use AI on documents first. 61% of the mortgage lenders and servicers surveyed use AI to pull data from documents, compared with 6% for credit scoring. Only general tools for writing and coding are more common (AARMR, MBA and BCG, 2026).
- Many use AI, but fewer see results. Only 13% of the mortgage lenders surveyed say AI has cut their costs significantly. Of the generative AI uses banks have put in place, 40% fall short of what the banks expected (EY-Parthenon).
- Some widely quoted numbers are weaker than they look. The $447 billion in AI savings by 2023 was a 2019 forecast from a paid report, and the free pages show no method. The claim that AI lending models approve 23% to 29% more borrowers came from one lender's simulation.
On this page
- How many banks use AI
- What banks use AI for
- AI in lending and credit decisions
- AI in mortgage lending
- Documents come before decisions
- Generative and agentic AI in banks
- Bank jobs and AI
- Rules and supervision
- Popular banking AI numbers, checked at the source
- Where these numbers come from
- If your loan files still arrive as PDFs and scans
- Frequently asked questions
The latest AI in banking statistics show wide use at large banks and slower use in lending decisions. The European Banking Authority says 92% of EU banks deploy AI. The European Central Bank says more than 85% of the large banks it supervises use AI. In a 2026 survey of mortgage lenders and servicers, 61% were using AI to pull data from documents. Only 6% used it for credit scoring.
Below are 41 figures on banks, credit unions and mortgage lenders, starting with regulators and central banks. Each figure has its own number and link. We checked every figure at its source on September 29, 2026. For AI in accounts payable and insurance, see our AI in finance statistics.
The gap comes from what each survey asks. European supervisors count any AI use at large banks. The US surveys ask about generative AI, or about lending alone. When you quote a number, say who was surveyed and what they were asked.
How many banks use AI#
A use case is one task an AI system does, such as reading documents or answering customer questions.
1. The EBA reports that 92% of EU banks are deploying AI, and that the other 8% are testing it or discussing use cases (EBA, 2025).
2. More than 85% of the large banks under European supervision use AI in some form, and the share rises every year (ECB, 2026).
3. 94% of the international banks in the survey use AI in their UK operations. Only insurers, at 95%, report a higher share (Bank of England and FCA, 2024).
4. The typical large UK bank reports 39 AI use cases, and the typical international bank 49. The typical firm in the survey reports 9. These are medians, so half the firms report more and half fewer (Bank of England and FCA, 2024).
5. 49% of banks and 59% of credit unions with $250 million to $50 billion in assets have deployed generative AI. That is the kind of AI that writes text or creates images. The survey covered 416 executives (Cornerstone Advisors, 2026).
What banks use AI for#
Most AI at banks still works on operations and internal tasks that customers don't see.
6. Operations and IT account for about 22% of all AI use cases at the UK financial firms surveyed. That is twice the next area, retail banking, at 11% (Bank of England and FCA, 2024).
7. Improving internal processes is the most common use. 41% of the firms surveyed use AI for it, ahead of cybersecurity at 37% (Bank of England and FCA, 2024).
8. Bank of America's assistant, Erica, had 20.6 million users and nearly 700 million interactions in 2025. It has passed 3.2 billion client interactions since 2018 (Bank of America, 2026).
9. 47% of the community bankers surveyed see AI for customer support as a promising opportunity, up 16 points since 2024. Nearly 62% see fully integrated loan processing systems as a promising opportunity. The Conference of State Bank Supervisors (CSBS) surveyed 268 community banks (CSBS, 2025).
10. JPMorgan Chase leads the Evident AI Index of 50 large banks. Capital One, Royal Bank of Canada, CommBank and Morgan Stanley follow (Evident, 2025).
AI in lending and credit decisions#
Banks move more slowly where AI would help decide who gets a loan. Many lenders let software decide some loans with fixed rules and credit scores. Few use AI to make the credit decision.
11. Only 13% of the community banks and credit unions surveyed have deployed AI in their credit and lending processes. Another 29% don't have it on their radar (Cornerstone Advisors for Zest AI, 2024).
12. About 8 in 10 of these lenders make automated underwriting decisions. But at 40% of them, software decides no more than 20% of their loans (Cornerstone Advisors for Zest AI, 2024).
13. Only 26% say their loan origination staff are very confident in automated decisions. Another 59% are somewhat confident (Cornerstone Advisors for Zest AI, 2024).
14. Underwriters handle more loans where decisions are automated. Each full-time underwriter reviews 3.5 times as many loan applications a month as at lenders that don't automate (Cornerstone Advisors for Zest AI, 2024).
15. Regulatory scrutiny is the top concern about AI in credit models, named by 70%. Yet 80% believe AI will improve credit and lending (Cornerstone Advisors for Zest AI, 2024).
16. In a commercial lending study, nearly 60% of bankers expect AI to have a moderate to strong impact on underwriting and lending. Actual use is still limited (Cornerstone Advisors for Baker Hill, 2026).
17. In a 2026 survey, just 6% of mortgage lenders and servicers have AI in live use for credit scoring and credit risk analytics. 29% have it in live use for underwriting decision support (AARMR, MBA and BCG, 2026).
For where AI fits in each stage of a loan, see our guide to AI in lending.
AI in mortgage lending#
The mortgage figures come from Fannie Mae and STRATMOR, and from a 2026 survey by AARMR, the association of state mortgage regulators. AARMR ran the survey with the Mortgage Bankers Association (MBA) and Boston Consulting Group (BCG). Most of its 31 respondents are independent mortgage companies, not banks. In production means in live use, not a pilot.
18. In Fannie Mae's 2023 survey, 73% of lenders using AI said their main goal was to work more efficiently, up from 42% in 2018. In that survey, 30% of lenders had deployed AI or were trying it (Fannie Mae, 2023).
19. 38% of the lenders in STRATMOR's study used AI or machine learning in 2024, up from 15% in 2023 (STRATMOR, 2025).
20. The 2026 survey covered 31 lenders and servicers that together make about 40% of US mortgage loans by volume. Nearly all had at least one AI use case in production. About 80% had fully scaled at least one, meaning in wide use rather than a limited rollout (AARMR, MBA and BCG, 2026).
21. The average respondent had about 10 of the 38 AI use cases in production, about a quarter of them. Every respondent expects to spend more on AI in the next 12 months (AARMR, MBA and BCG, 2026).
22. 59% name worries about rules and compliance as one of the biggest barriers to using AI more widely. No other barrier was named as often. 45% say the return on AI tools is still unclear. Data quality and a lack of products on the market tie for third, at 24% each (AARMR, MBA and BCG, 2026).
23. 87% have written AI policies, and 81% have a person check the AI's work. But only 58% keep checking their AI models after they go into live use (AARMR, MBA and BCG, 2026).
24. Smaller lenders in the survey are less likely to monitor their AI. 80% of the large lenders, which closed 25,000 or more loans in 2025, monitor their AI models once they are in use. The share is 60% at mid-sized lenders and 40% at small ones, which closed fewer than 5,000 loans (AARMR, MBA and BCG, 2026).
25. More than 1 in 4 respondents say AI is being used outside their approved tools. The survey calls this shadow AI (AARMR, MBA and BCG, 2026).
Documents come before decisions#
In loan work, the AI that reaches production mostly reads documents. Only general tools for writing and coding are more common. AI rarely makes the credit decision. The chart shows 8 of the 38 use cases in the 2026 mortgage survey.
26. Data extraction from documents is in production at 61% of the lenders and servicers surveyed, and 42% have fully scaled it. It is the most common AI use in making new loans, also called origination (AARMR, MBA and BCG, 2026).
27. In the same STRATMOR study, 63% of lenders use AI to classify and index documents, and 54% use it to read them. STRATMOR doesn't say whether these shares are of all lenders or only of those using AI (STRATMOR, 2025).
28. 63% of mortgage lenders using AI rely on outside vendors for it, and 21% are building their own (STRATMOR, 2025).
29. 35% of the community lenders surveyed use patterns in borrowers' deposit account transactions in their credit models. Among data from outside the credit report, deposit patterns come second, after job history at 41%. Just 13% are very satisfied with the credit data they can get (Cornerstone Advisors for Zest AI, 2024).
30. Only 13% of the mortgage lenders and servicers surveyed say AI in production has cut their costs significantly. 29% report a significant gain in employee productivity (AARMR, MBA and BCG, 2026).
Deposit patterns can also come from the bank statements an applicant provides. Before anyone studies a statement's cash flow, someone has to confirm that its transactions match its balances. Commercial loans add financial statements and tax returns. One community banker gave an estimate in the CSBS survey report. Lloyd Hamm Jr. of River Run Bancorp said AI could do 80% to 90% of the work in his bank's yearly commercial credit reviews. A credit analyst would still check the result. Our financial spreading page shows that part of the file.
| Field | Extracted value | Confidence |
|---|---|---|
| Account holder | Harbor Street Bakery LLC | |
| Address | 118 Harbor St, Portland, ME | |
| Bank name | First Midwest Bank | |
| Account number | •••• 4821 | |
| Account type | Business checking | |
| Statement period | 2026-07-01 → 2026-07-31 |
| Field | Extracted value | Confidence |
|---|---|---|
| Opening balance | 42,180.55 | |
| Total deposits | 88,412.10 | |
| Total withdrawals | 91,686.44 | |
| Closing balance | 38,906.21 | |
| Transaction count | 142 |
| Field | Extracted value | Confidence |
|---|---|---|
| Date | 2026-07-14 | |
| Description | ACH DEPOSIT STRIPE PAYOUT | |
| Amount | +3,284.10 | |
| Debit / credit | Credit | |
| Running balance | 51,902.44 | |
| Category | Card processor payout |
Docsumo reads bank statements, pay stubs, W-2s and tax returns for lenders. It joins transaction tables that run across pages and checks each row against the running balance. Values it isn't sure of go to a person, who can click a value to see its line on the page. On the Enterprise plan, cross-document validation checks one document against another in the same file. Cash flow analytics is on the Enterprise plan too. Docsumo reads the statements an applicant provides and doesn't connect to bank accounts. The data goes to your loan system through the API and webhooks, and your underwriters still make the credit decision. For income documents, see income verification.
Our take. Surveys count use cases in production, and vendors quote accuracy from their own tests. Neither number tells a lender how much work is left on its own files. We'd judge any extraction tool, ours included, by how many values a person had to fix on a sample of recent loan files. That count is easy to track each month, and it shows where the manual work still is.
Generative and agentic AI in banks#
31. 77% of the retail and commercial banks surveyed have launched or soft-launched generative AI tools, up from 61% in 2023. A soft launch releases a tool to a limited group first (EY-Parthenon, 2025).
32. Fewer banks, 47%, say they have rolled out generative AI tools, up from 10% in 2023. The survey covered 100 retail and commercial banks (EY-Parthenon, 2025).
33. Around 40% of the large EU banks in an EBA survey used generative AI at the end of 2024. By the first quarter of 2025, the share was over 60% (BaFin, 2026).
34. 55% of EU banks surveyed use general-purpose or agentic AI in work that deals directly with customers. General-purpose AI can do many kinds of tasks, and agentic AI can take steps on its own (EBA, 2025).
35. Executives or the board have discussed agentic AI at more than half of the US banks and credit unions surveyed (Cornerstone Advisors, 2026).
36. Only 16% of banks' generative AI use cases reach full deployment. Of the use cases banks have implemented, 40% fall short of expectations (EY-Parthenon, 2025).
37. Regulatory compliance is the top barrier to generative AI at banks, at 26%. Data privacy follows at 22%, and access to good data at 21% (EY-Parthenon, 2025).
Bank jobs and AI#
Projections from the Bureau of Labor Statistics (BLS) don't single out AI. They do show which bank jobs automation affects most.
38. BLS projects teller jobs to fall 13% from 2025 to 2035, a loss of 44,700 jobs. It expects fewer branch visits as people bank online, and ATMs that handle more teller tasks (BLS, 2026).
39. New accounts clerk jobs are projected to fall 6%, and jobs for credit authorizers, checkers and clerks 7%. BLS points to online tools that let customers do more themselves (BLS, 2026).
| Occupation | US jobs, 2025 | Projected change to 2035 |
|---|---|---|
| Tellers | 339,200 | −13% |
| Credit authorizers, checkers and clerks | 12,200 | −7% |
| New accounts clerks | 37,900 | −6% |
| Loan interviewers and clerks | 169,100 | −2% |
| Loan officers | 283,000 | +1% |
Loan jobs change the least. BLS still expects loan officer jobs to grow 1%, though it says technology in loan processing will slow that growth (BLS, 2026).
Rules and supervision#
40. In April 2026, the Federal Reserve, FDIC and OCC replaced their 2011 guidance on model risk. Model risk is the chance that a wrong or misused model leads to a bad decision. The new guidance doesn't cover generative or agentic AI. The agencies plan to ask the industry for input on banks' use of AI (OCC, 2026; Federal Reserve, 2026).
41. State regulators, who supervise 79% of US banks, published a common AI framework for their examiners in September 2026. Its examiner guide says it creates no new legal obligations or supervisory requirements (CSBS, 2026; CSBS examiner guide, 2026).
For the AI governance rules that Fannie Mae and Freddie Mac set for lenders, see the rules table in our AI in finance statistics.
Popular banking AI numbers, checked at the source#
A few numbers about AI in banking appear in almost every round-up. Some are forecasts whose target year has passed. Others have no source at all. We traced each one as far as it goes.
| Often quoted | What the source says | Cite instead |
|---|---|---|
| AI will save banks $447 billion by 2023 | A 2019 forecast in a paid Insider Intelligence report, now published on eMarketer's site. The free pages show no method. Its 2023 AI in finance guide repeated the figure. 2023 has passed, and we found no follow-up that measured actual savings | A survey of results, such as the 13% of mortgage lenders in a 2026 survey reporting a significant cost cut (figure 30) |
| Chatbots will save banks $7.3 billion a year by 2023 | A 2019 Juniper Research forecast of growth from an estimated $209 million in 2019. The forecast year has passed | We found no measured savings figure. For chatbot use, cite a bank's own numbers, such as Erica's nearly 700 million interactions in 2025 (figure 8) |
| 37% of Americans have used a bank chatbot (CFPB) | The CFPB's 2023 chatbot report quotes an Insider Intelligence estimate for 2022. The CFPB didn't measure it | Cite it as Insider Intelligence's estimate, reported by the CFPB |
| Generative AI will add $200 billion to $340 billion a year to banking | A 2023 McKinsey estimate of potential value, equal to 2.8% to 4.7% of revenue, if the use cases were fully put in place | Keep it, but call it potential value, not money banks have made |
| AI lending models approve 23% to 29% more borrowers at 15% to 17% lower APRs | In a 2019 update, the CFPB reported results from one lender's own simulations. Upstart compared its model with one built only on traditional credit data. The CFPB didn't repeat the analysis itself. The ranges come from separate results for each race, ethnicity and sex group. For all applicants together, the model approved 27% more at 16% lower average APRs | Cite it as Upstart's 2019 simulation, not a general result for AI lending |
| 92% of global banks use AI in at least one core function | We found it only in stat round-ups that give no source | 92% of EU banks deploy AI (EBA, 2025, figure 1) |
| AI is involved in over 70% of loan underwriting decisions at top US banks | No source given, and we found no survey that measures it | 6% of the mortgage lenders surveyed use AI for credit scoring (figure 17), and 13% of the community lenders surveyed use AI in credit (figure 11) |
| 54% of customer interactions at US banks are fully automated | No source given, and no bank or regulator we checked reports such a share | McKinsey's 2023 finding that roughly half of customer contacts at North American banks, telecom and utility companies were already handled by machines, not all of them AI |
| A market size for AI in banking, such as $15 billion or $34 billion for 2025 | Paid market reports disagree. For 2025, The Business Research Company says $15.32 billion and Polaris Market Research says $34.40 billion. Neither page shows how the number was built | Adoption or spending surveys that name their sample |
Where these numbers come from#
We started with regulators and central banks. The EBA, the ECB, the Bank of England and the FCA publish survey data on AI at banks. The OCC, the Federal Reserve and state supervisors publish the rules.
The rest come from industry surveys, and the sponsor matters. Cornerstone Advisors' two lending studies were commissioned by Zest AI and Baker Hill, which sell lending software. EY-Parthenon and BCG sell consulting, STRATMOR advises mortgage lenders, and Evident sells research on banks' use of AI. We kept these surveys because they are the newest data on these questions, and each figure names its source.
Samples vary widely, from 31 mortgage companies in the AARMR, MBA and BCG survey to 416 executives in Cornerstone's banking study. Where AARMR's press release or news coverage differs from the report, we follow the report's own slides. The definitions differ too. Some surveys count any use of AI, some count only generative AI, and some count only tools in production. Compare figures from different surveys with care.
Most figures here are from 2025 and 2026, and older ones carry their year. Anyone may use these figures. Please link to the original source, and to this page if it helped.
If your loan files still arrive as PDFs and scans#
Apart from general office tools, most of the AI that lenders run today reads documents. So far, most of the gains are staff time saved. Documents are a sensible place to start. Pick one document type, such as bank statements for small business loans, and count how many values a person corrects today. Then test any tool on the same files and compare the counts.
Book a demo with a few of your own loan files, or start a free trial.
Frequently asked questions#
What percentage of banks are using AI?
It depends on who is surveyed. The European Banking Authority (EBA) reports that 92% of EU banks deploy AI. In a 2024 survey by the Bank of England and FCA, 94% of the international banks in the UK that responded use it. In the US, 49% of the banks surveyed with $250 million to $50 billion in assets have deployed generative AI (Cornerstone Advisors, 2026).
How is AI being used in banking?
Most often in operations. Operations and IT make up about 22% of AI use cases at the UK financial firms surveyed. No other area has more (Bank of England and FCA, 2024). The most common use there is improving internal processes. In mortgage lending, 61% of the lenders surveyed use AI to pull data from documents, compared with 6% for credit scoring.
Which banks are leading in AI?
JPMorgan Chase leads Evident's October 2025 AI Index of 50 large banks. Capital One, Royal Bank of Canada, CommBank and Morgan Stanley follow. The index scores banks from public data on talent, innovation, leadership and transparency.
Will AI replace banking jobs?
The US Bureau of Labor Statistics projects teller jobs to fall 13% from 2025 to 2035, as more banking moves online. It expects loan officer jobs to grow 1%. At mortgage lenders, the most common AI tools help staff write and summarize. 87% of the lenders and servicers surveyed have these tools in live use.
How are lenders using AI?
In loan work, mostly on documents so far. In 2024, 38% of the mortgage lenders in STRATMOR's study used AI or machine learning, and classifying and reading documents were common uses. In 2024, only 13% of community banks and credit unions had deployed AI in credit and lending (Cornerstone Advisors for Zest AI). Our guide to AI in lending shows where it fits in the loan process.
Sources
- EBA, Rising application of AI in EU banking and payments sector, digital finance factsheet (Sep 25, 2025)
- ECB Banking Supervision, Pedro Machado, Technology is neutral, governance is not: AI adoption in the banking sector (Feb 24, 2026)
- Bank of England and FCA, Artificial intelligence in UK financial services 2024 (Nov 21, 2024)
- Cornerstone Advisors, What's Going On in Banking 2026 (Jan 29, 2026)
- Bank of America, BofA AI and digital innovations fuel 30 billion client interactions (Mar 10, 2026)
- CSBS, 2025 Annual Survey of Community Banks (2025)
- Evident, Evident AI Index for banks (Oct 2025)
- Cornerstone Advisors, commissioned by Zest AI, Achieving High-Performance Lending: The Impact of AI on Lending Efficiency (2024)
- Cornerstone Advisors, commissioned by Baker Hill, What's Going On in Commercial Lending 2026 (Mar 25, 2026)
- AARMR, MBA and BCG, The state of AI among mortgage lenders and servicers (Sep 2026)
- AARMR, Mortgage industry survey finds AI adoption growing but regulatory uncertainty remains key barrier (Sep 28, 2026)
- Fannie Mae Economic and Strategic Research, Mortgage Lender Sentiment Survey special topics report, Artificial intelligence and mortgage lending (Oct 2023)
- STRATMOR Group, Nicole Yung, digital innovations results from the 2024 Technology Insight Study (Apr 2025)
- EY-Parthenon, AI in banking: 2025 Generative AI in Banking survey insights (Sep 2025)
- BaFin, Risks in focus 2026, trend 1: digitalisation (Jan 28, 2026)
- BLS, Occupational Outlook Handbook, tellers (Aug 27, 2026)
- BLS, Occupational Outlook Handbook, financial clerks, job outlook (Aug 27, 2026)
- BLS, Occupational Outlook Handbook, loan officers (Aug 27, 2026)
- OCC, OCC issues updated model risk management guidance, NR 2026-29 (Apr 17, 2026)
- Federal Reserve, SR 26-2, Revised guidance on model risk management (Apr 17, 2026)
- CSBS, CSBS announces AI supervisory framework (Sep 16, 2026)
- CSBS, AI supervisory framework, core examiner guide v1.0 (Sep 2026)
- Insider Intelligence (Eleni Digalaki), AI in banking: artificial intelligence could be a near $450 billion opportunity for banks, paid report page on eMarketer (Jun 21, 2019)
- eMarketer (Insider Intelligence), AI in finance guide, Internet Archive copy of a page dated Jan 2, 2023
- Juniper Research, Bank cost savings via chatbots to reach $7.3 billion by 2023 (Feb 20, 2019)
- CFPB, Chatbots in consumer finance (Jun 6, 2023)
- McKinsey, The economic potential of generative AI: the next productivity frontier (Jun 14, 2023)
- CFPB, An update on credit access and the Bureau's first No-Action Letter (Aug 6, 2019)
- The Business Research Company, AI in banking global market report 2026 (Sep 2026)
- Polaris Market Research, Artificial intelligence in banking market (Jul 2026)
First published .