AI in finance statistics: 67 figures on AP, lending, insurance and back-office work

The latest AI in finance statistics show adoption climbing fast and results lagging behind it. KPMG finds active AI use in large companies' finance functions went from 30% to 75% between 2024 and 2026. Yet only 7% of finance organizations report a high impact, and the median ROI finance teams report from AI is 10%.

Most of the early use sits in back-office document work, the invoices, loan files, submissions and claims that people still read and key by hand. The figures below are grouped by where that work happens. Each one is numbered so you can link to a single statistic, and we checked every one on the publisher's own page in September 2026.

Read the spread before quoting any single number. Surveys of large companies report far higher adoption than government data on all firms, and "using AI" can mean one chatbot or a production workflow.

AI in finance teams and accounting firms#

1. Active AI use in the finance function rose from 30% to 75% of organizations between 2024 and 2026, in KPMG's survey of 1,013 senior finance leaders at large companies in 20 countries (KPMG Global AI in Finance Report, 2026).

2. 84% of finance organizations have implemented AI or plan to, yet only 7% report a high or very high impact (Gartner survey of 183 finance leaders, 2026).

3. 44% of CFOs used generative AI for more than five use cases in 2025, up from 7% a year earlier (McKinsey survey of 102 CFOs, 2025).

4. Among CFOs of North American companies with $1 billion or more in revenue, 51% say their finance teams use AI for productivity tasks, 44% for planning and budgeting and 41% to analyze financial data (Deloitte CFO Signals, Q2 2026).

5. Finance leaders say 45% of their AI investment leans toward productivity and only 20% toward better decisions (Gartner survey of 204 finance leaders, 2026).

Report creation is the use case the most finance functions are already pursuing, with accounts payable and data extraction close behind.

6. 65% of finance organizations saw a moderate or significant improvement in close efficiency from AI over the past year (KPMG, 2026).

7. Finance teams that run cloud ERP with embedded AI assistants will close their books 30% faster by 2028, Gartner predicts (Gartner, 2026).

8. Accountants who used generative AI finalized monthly statements 7.5 days faster than peers working the traditional way, in a field study of 79 firms by researchers at Stanford and MIT (Stanford GSB, 2025).

9. 41% of tax and accounting firms used AI in 2025, up from 9% a year earlier, and 72% now use it at least weekly (Wolters Kluwer survey of 2,700+ professionals, 2025).

10. 34% of tax firms say their organization already uses generative AI. Across professional services, organization-wide use nearly doubled in a year, from 22% to 40% (Thomson Reuters Institute, 2026).

AI in accounts payable#

Accounts payable is where finance AI has gone furthest, largely because its input is a document. For cost per invoice, cycle times and exception rates, see our accounts payable statistics.

11. 38% of finance functions are already pursuing AI for accounts payable automation, and another 56% plan to. For numerical data extraction the figures are 35% and 55% (Gartner, 2026).

12. Accounts payable is the most mature finance process for AI, with 33% of organizations already scaling AI there (The Hackett Group, 2026).

13. 58% of accounts payable teams are using or piloting AI, but only 23% have deployed it across several AP functions (Ardent Partners, 2026).

14. Invoice capture and data extraction is the most widely deployed AI use in AP, cited by 58% of the AP organizations surveyed (Ardent Partners, 2026).

Capture comes first because it is easy to measure, in invoices no longer keyed and fields no longer fixed. AI invoice processing walks through how it works, and accounts payable automation shows the full workflow in Docsumo.

AI in lending and banking operations#

Financial firms sit near the top of every adoption survey. The cost of making a loan hasn't come down with it.

15. About 30% of US financial-sector firms use AI, against about 18% of all firms, based on Census Bureau survey data around the end of 2025 (Federal Reserve Board staff, FEDS Notes, 2026).

16. About 63% of US financial-sector workers use generative AI for work, the highest of any sector, compared with about 41% of all workers (Federal Reserve Board staff, 2026).

17. 75% of UK financial services firms already use AI and another 10% plan to within three years, up from 58% using it in 2022 (Bank of England and FCA, 2024).

18. 81% of financial services firms worldwide are adopting AI at some level, and 40% report advanced adoption, meaning they are scaling it or rebuilding around it (Cambridge Centre for Alternative Finance, 2026).

19. Credit risk and underwriting is among the most widely adopted AI uses in financial services, with 54% of firms piloting it or further along (Cambridge Centre for Alternative Finance, 2026).

20. Only 12.5% of US community banks offer automated loan underwriting, and 75.4% neither offer it nor plan to within a year (CSBS Annual Survey of Community Banks, 2025).

21. Independent mortgage banks spent $10,936 to produce each loan in the second quarter of 2026, against a long-run average of $7,945 since 2008 (Mortgage Bankers Association, 2026).

22. In 2025 the most profitable fifth of mortgage lenders spent $10,074 per loan and the least profitable fifth spent $12,603, a $2,529 gap on every loan (Mortgage Bankers Association, 2026).

23. The average US purchase loan took 36.8 days to close in March 2026, the fastest since ICE began tracking in 2019 (ICE Mortgage Monitor, 2026).

24. Among small businesses that applied for credit at a large bank, 26% found the application process difficult and 21% faced a long wait for a decision or funding (Federal Reserve Small Business Credit Survey, 2026).

Much of that cost and waiting sits in the loan file, where statements, pay stubs, tax returns and IDs are checked against the application. Our guide to AI in lending shows where automation fits, and document automation for lending shows how Docsumo reads and checks the file.

AI in insurance operations#

25. 44% of insurers had generative AI in live use by May 2025, up from 28% in 2024 and 8% in 2023 (Celent survey of 55 insurers, 2025).

26. Just 23% of insurers have integrated AI across the enterprise, though 81% report at least a 5% gain in gross written premiums from data and AI (Accenture survey of 263 insurance executives, 2026).

27. 60% of P&C insurers are still exploring AI or running proofs of concept, and about 10% have advanced AI capabilities (Capgemini World Property and Casualty Insurance Report, 2026).

28. 68% of large insurers are adopting automated data entry as part of their generative AI investment (EY-Parthenon survey of 100 insurers, 2025).

29. 31 of 93 US health insurers surveyed by state regulators had AI or machine learning in production for claims adjudication, and 35 had it in production for data processing (NAIC Health AI/ML Survey, 2025).

30. Only 16% of North American P&C insurers use AI to support underwriting decisions today, but 60% plan to prioritize it by 2028 (WTW survey of 59 insurers, 2026).

31. Underwriting executives expect their use of AI to rise from 14% today to 70% within three years (Accenture survey of 430 underwriting executives, 2025).

32. Underwriters spend more than 41% of their time on administrative and operational work such as data entry and record keeping (Capgemini survey of 201 underwriters, 2024).

33. Only 43% of P&C underwriters trust and regularly accept the automated recommendations their analytics tools give them (Capgemini, 2024).

34. AIG's excess and surplus lines insurer Lexington received about 300,000 new business submissions in 2024, 10 times its 2018 volume, and bound about 2% of them (AIG Investor Day, 2025).

35. On average, 59% of US individual life insurance applications qualify for an accelerated underwriting path, across 30 carriers surveyed (Gen Re, 2025).

36. US homeowners claims took an average of 29.6 days to complete repairs and 40.7 days to reach final payment, 2.8 and 3.4 days faster than a year earlier (J.D. Power, 2026).

The submission numbers show where underwriting time goes. Before anyone prices a risk, someone reads the ACORD forms, loss runs and schedules, and that reading is the work document automation for insurance takes on.

Back-office documents and data entry#

37. 66% of organizations planning a new intelligent document processing project intend to replace a system they already have, and 48% expect their paper volumes to grow (AIIM and Deep Analysis, 2025).

38. Only 30% of organizations count reducing headcount as a sign of a successful document processing project, the lowest of seven success measures (AIIM and Deep Analysis, 2025).

39. Office and administrative support tasks made up 13% of business API traffic to Anthropic's Claude in November 2025, up 3 points since August, as companies automate routine back-office work (Anthropic Economic Index, 2026).

40. AI already covers 67% of the tasks of data entry keyers, whose core job of reading source documents and entering data is among the most exposed to automation (Anthropic, 2026).

To turn rates like these into hours for your own team, put in your volumes below. The defaults are starting values, so replace them with yours. For market figures on the software itself, see our intelligent document processing statistics.

Jobs in finance operations#

41. Data entry keyers are among the fastest-shrinking US occupations. BLS projects a 25.5% drop, from 131,800 jobs in 2025 to 98,200 in 2035, against 3.5% growth for all occupations (BLS Employment Projections, 2026).

42. BLS projects bookkeeping, accounting and auditing clerks to decline 6% from 2025 to 2035, to 1,446,800 jobs, and notes that software has automated many of their tasks (BLS Occupational Outlook Handbook, 2026).

43. Accountants and auditors are still projected to grow 5% over the same decade, faster than average, even as BLS expects AI and cloud tools to automate some routine accounting tasks (BLS, 2026).

44. Payroll and timekeeping clerk jobs are projected to fall 16%, while billing and posting clerks hold flat at about 418,000 as automated invoice processing software raises their productivity (BLS, 2026).

45. Loan interviewer and clerk jobs and insurance claims and policy processing clerk jobs are each projected to fall 2%, and loan officer jobs to grow just 1%, with BLS pointing to productivity-enhancing technology in loan processing (BLS, 2026; BLS, 2026).

46. BLS projects claims adjusters, examiners and investigators to decline 5% and insurance underwriters 4%, as software estimates claims from photos and automated underwriting lets fewer people process applications (BLS, 2026; BLS, 2026).

47. 67% of P&C insurers cite a shortage of AI skills (Capgemini, 2026).

48. Nearly half of P&C insurance employees with access to AI tools, 47%, say their workday hasn't changed after 18 months of use (Capgemini, 2026).

AI agents in finance operations#

49. The share of large financial institutions with agentic AI in production rose from 23% in 2025 to 70% in 2026, and 85% of them use it to automate processes and operations (IIF and EY survey of 65 institutions, 2026).

50. 52% of financial services firms are piloting or deploying agentic AI, with 23% already scaling it (Cambridge Centre for Alternative Finance, 2026).

51. 12% of finance functions are building multi-agent AI systems, and another 19% are orchestrating agents across workflows (KPMG, 2026).

52. 54% of North American CFOs say integrating AI agents into finance is a transformation priority for 2026 (Deloitte CFO Signals, Q4 2025).

53. 77% of finance leaders are comfortable letting AI agents make some decisions on their own, but only 14% are comfortable with fully autonomous agents for more critical decisions (Deloitte survey of 1,434 finance leaders, 2026).

54. 79% of US executives say AI agents are being adopted at their companies, yet only 20% trust agents to handle financial transactions, against 38% for data analysis (PwC AI Agent Survey, 2025).

55. 55% of AI use cases at UK financial firms involve some automated decision-making, but only 2% are fully autonomous (Bank of England and FCA, 2024).

56. 22% of insurers plan to have an agentic AI solution in place by the end of 2026 (Celent, 2025).

Our take. The trust numbers above are the right instinct for finance operations. We'd let an agent run the checks that have a clear right answer, such as whether an invoice's lines add up to its total, and send everything else to a person with the reason and the source line in front of them. Your team then reviews a short list of exceptions instead of every file. Docsumo's case management, on the Enterprise plan, works this way.

Returns and what holds AI back#

57. The median ROI finance teams report from AI is 10%, half the 20% many target, and only 45% of finance executives can quantify ROI at all (BCG survey of 280+ finance executives, 2025).

58. 71% of finance leaders say AI is meeting or beating ROI expectations in their function, but only 23% say it is beating them (KPMG, 2026).

59. 86% of large financial institutions report measurable financial returns from predictive AI and 81% from generative AI, against 32% from agentic AI so far (IIF and EY, 2026).

60. 63% of finance organizations said AI implementation moved slower than expected in 2025 (Gartner, 2026).

61. Only about 1 in 4 generative AI use cases that insurers identify becomes a successful implementation (EY-Parthenon, 2025).

62. 42% of P&C insurers track no AI metrics at all (Capgemini, 2026).

63. Technology raised labor productivity in insurance claims, servicing and policy issuance by 14% in P&C and 24% in life, yet insurers' cost ratios are 10% higher than in 2005 (McKinsey, 2026).

64. Data quality is the top barrier to putting AI into production for 84% of large financial institutions, ahead of skilled staff at 77% and integration with existing systems at 58% (IIF and EY, 2026).

65. 77% of organizations name a lack of AI talent as the leading barrier to AI in finance (The Hackett Group, 2026).

66. 50% of insurers name integration with legacy systems among their top obstacles to scaling data and AI, and 45% name access to enough good data (Accenture, 2026).

If you're building a case for your own team, our guide to measuring IDP ROI shows which numbers to baseline first.

Rules for AI in lending and insurance#

Regulators have moved from principles to dated requirements, and two of the biggest take effect in 2027.

67. As of August 31, 2026, 25 states and the District of Columbia had adopted the NAIC model bulletin on insurers' use of AI systems, and California, Colorado, New York and Texas have their own insurance-specific rules or guidance (NAIC, 2026).

How we picked and checked these figures#

Every figure here was published in 2024, 2025 or 2026, and we checked each one on the publisher's own page or report on September 26 and 27, 2026. We left out forecasts whose target date has already passed, and numbers we could only find on sites that repeat each other without a source.

Some sources sell products in the areas they survey. Ardent Partners' 2026 AP research is sponsored by AP software vendors. Wolters Kluwer and Thomson Reuters sell tax and accounting software, EY and PwC sell AI services, and the Anthropic figures come from use of its own model. We kept them because they are the newest data on those questions, and each figure names its source. Samples vary widely too, from 59 insurers in WTW's survey to more than 2,700 professionals in Wolters Kluwer's, so compare figures from different surveys with care.

The BLS figures use its 2025 to 2035 projections, released in August 2026. Many articles still quote the older 2024 to 2034 set.

If your team still keys data from documents#

Most of the returns in these surveys come from narrow work that is easy to measure, and the data entry figures show where that work is heading. Before you buy anything, count how many documents a person touches today and how many fields they fix. That baseline makes any AI result easy to check, including ours.

To see it on your own files, book a demo or start a free trial.

Frequently asked questions#