Best AI for CFOs

Research notes on AI in finance work, by task

Index
W/P 3
Built
4 November 2026
Volume
1
This page
4 exhibits
  1. Best AI for CFOs
  2. Use cases
  3. Expense audit

3of 12 workpapers

Expense audit

How do AI products audit employee expenses, and what is left for a person to decide?

Published . Vendor pages read 29 October 2026.

How AI products review employee expense reports, receipts and card spend against policy, what they flag, who decides, and questions for a vendor.

Products listed
9
Makers in the SEC ticker file
2
Filing excerpts
4
Notes
19

Lead schedule.Data the makers in this workpaper name

One stroke per record, in fives. Systems named by 4 or more records; every system is in Appendix C.

The schedule scrolls sideways; the workpaper column stays in place.

W/PBanksNetSuiteERPSAPCardsIntacctWorkdayClaudeCRMQuickBooksDynamicsOraclePaymentsPayrollSalesforceNo data named
3Expense audit11212112

Answer sheet.What the makers' pages in this workpaper answer

One stroke per record: solid where the maker's own page answers the question, dotted where the pages read are silent. A count of what the pages say, not a score.

W/PData it namesWhere a person reviewsModel or data handling
3Expense audit7 of 94 of 92 of 9

1The work

AI expense audit products read every expense line and receipt against the company's policy and flag the ones that look wrong; a person decides what is a violation and what happens next. The makers in this note say they check receipts, categorise spend, enforce policy at the time of purchase or submission, and rank flagged items by risk so auditors start with the worst.

Expense audit used to mean sampling. A small team reviewed a percentage of reports, checked receipts against claimed amounts, looked for duplicates and personal spend, and chased missing documentation. Sampling misses things by design. Software that reads every receipt changes the economics, but it does not change who is accountable for the policy or for the decision to recover money from an employee.

The work has a tax side. In the United States, how reimbursements are treated depends on whether the company's arrangement meets the IRS rules for an accountable plan; IRS Publication 463 sets out the accountable and nonaccountable plan rules.1 Receipts, business purpose and timely substantiation matter for that, not only for policy.

A person must stay in the loop for three things: setting and interpreting the policy (a model cannot decide whether a dinner was reasonable for a client meeting in a given city), deciding on a finding about an individual, and handling anything that could become a disciplinary or legal matter. Products differ on whether they only flag, or also approve low-risk items, reject items or contact employees automatically. Those are design choices a finance team should make, not defaults to accept.

2Where AI is applied

Makers describe receipt reading and matching, automatic categorisation, real-time policy checks, risk ranking of flagged spend, automatic approval of low-risk items and follow-up with employees. Several sit on top of an existing expense system rather than replacing it.

Exhibit 3.1.What the makers say, in their words, alphabetical by maker

  • AppZen adds 100% audit coverage on top of your expense management system, receipt-level analysis, and autonomous resolution without replacing your system of record.8
    AppZen, Expense Audit
  • Your audit agent monitors company expenses according to your internal processes and categorizes potential violations by risk level.9
    Brex, Brex AI agents
  • Emburse AI interacts with the Emburse Expense Enterprise workflow to automatically approve in-policy expenses or flag risky ones before they become a problem.10
    Emburse, Emburse AI
  • Concierge AI catches expense policy violations in realtime and follows up automatically.11
    Expensify, Concierge AI
  • With access to global inventory sources and every transaction type, Navan’s AI agents identify savings opportunities and give finance and travel teams a complete, real-time view of T&E.12
    Navan, Navan Intelligence
  • Surface high-risk T&E activity first, including repeat misuse, suspicious patterns, duplicate reimbursements, fake receipts, and policy violations.13
    Oversight, Travel and expense monitoring
  • AI expense reviews that catch every out-of-policy spend14
    Pleo, AI expense reviews
  • Let the agent handle low-risk and routine approvals. Any risky or ambiguous transactions are flagged to you, fast.15
    Ramp, Ramp Intelligence, Ramp agents
  • Rydoo’s AI agents and assistants strengthen your control, take on real work and increase productivity across your expense management and accounts payable processes.16
    Rydoo, Rydoo Intelligence

Source: the makers' own pages, read 29 October 2026; notes 8 to 16.

3Limits and control questions

Thresholds are a judgment, not a setting. The SEC staff's view on materiality is that "exclusive reliance on this or any percentage or numerical threshold has no basis in the accounting literature or the law."2 A tool that flags only what crosses a percentage leaves the qualitative items to a person.

False positives have a cost: each one is an employee asked to explain a legitimate expense. Ask for the rate at which flags are overturned by reviewers, measured on the company's own data.

Decisions about named employees deserve care. A model's risk ranking is an input to a person's judgment, and the record of who decided should be kept with the evidence.

For a listed company, any automated step that feeds the financial statements sits inside internal control over financial reporting, which management must evaluate and report on; the SEC's interpretive guidance for that report is Release 33-8810.3 COSO describes its Internal Control, Integrated Framework as originally issued in 1992 and refreshed in 2013.4

NIST's AI Risk Management Framework is a common reference for those questions. NIST says the framework is "intended for voluntary use and to improve the ability to incorporate trustworthiness considerations into the design, development, use, and evaluation of AI products, services, and systems."5 Its first version is published as NIST AI 100-1.6

Vendor claims about AI are claims, not findings. The SEC has brought settled charges against two investment advisers "for making false and misleading statements about their purported use of artificial intelligence (AI)."7 A finance team that relies on a vendor's description should ask for evidence of what the feature does with its own data.

4What filers say

Exhibit 3.2.Filers on their own finance work

Companies describing AI in their own operations, from their SEC filings. A filing is not evidence that the filer uses any product on this page unless it names one.

  • For example we use AI Technologies to optimize our internal processes, including, but not limited to, analyzing and processing receipts, bills, and invoices, responding to customer support requests, and drafting personalized marketing communications.17
    Expensify, Inc. (EXFY), Form 10-K filed 26 February 2026
  • In accounting and finance, we are developing AI agents intended to automate invoice processing, reconciliation, residual calculations, expense categorization and routine financial close activities.19
    OLB GROUP, INC. (OLB), Form 10-Q filed 14 August 2026

Source: SEC EDGAR full-text search; notes 17 to 19.

Exhibit 3.3.Finance-software filers on their AI products

Listed software companies describing their own AI products to investors. These are the makers' statements, made in filings.

  • Concierge, our proprietary customer support engine powered by artificial intelligence (“AI”), helps new members set up an account, connect teammates, assign company policies and immediately sync Expensify with existing accounting, HR and travel systems.17
    Expensify, Inc. (EXFY), Form 10-K filed 26 February 2026
  • We offer a comprehensive, all-in-one, AI-powered travel, payments and expense management solution designed to streamline the entire travel lifecycle, from booking and policy enforcement to payment processing, expense reconciliation, and reporting.18
    Navan, Inc. (NAVN), Form 10-K filed 2 April 2026

Source: SEC EDGAR full-text search; notes 17 to 18.

5Products

Exhibit 3.4.Products addressing expense audit, by the data they name

9 product records whose makers say they address this task, alphabetical by maker. Open a row for our summary, the data it names and the maker's words on review; the marks show which questions the maker's pages answer.

  1. Expense AuditAppZenData it names: quoted.Where a person reviews: not stated.Model or data handling: not stated.
    Our summary
    An expense audit product that reviews employee expense reports and receipts on top of an existing expense system. The maker says it covers every report rather than a sample.
    Data it names (our list)
    SAP Concur, Workday, Emburse, Oracle, Coupa, Rydoo, Cytric
    Where a person reviews (quoted)
    Not stated on the pages read.
  2. Brex AI agentsBrexData it names: quoted.Where a person reviews: quoted.Model or data handling: not stated.
    Our summary
    AI agents in a corporate card and spend platform that audit, approve or escalate employee expenses against the customer's policy. The maker says customers upload and edit the policy the agents enforce.
    Data it names (our list)
    Expense policy, Receipts
    Where a person reviews (quoted)
    escalates exceptions, and follows up with anyone out of compliance.
  3. Emburse AIEmburseData it names: not stated.Where a person reviews: not stated.Model or data handling: not stated.
    Our summary
    AI in Emburse's travel and expense software that extracts receipt data, categorises expenses and approves or flags them against policy.
    Data it names (our list)
    Not stated on the pages read.
    Where a person reviews (quoted)
    Not stated on the pages read.
  4. Concierge AIExpensifyData it names: quoted.Where a person reviews: not stated.Model or data handling: not stated.
    Our summary
    An AI assistant in an expense app that categorises expenses, answers policy questions and flags violations. The maker says each action has an Explain button showing the reasoning.
    Data it names (our list)
    QuickBooks, Xero, NetSuite, Sage Intacct
    Where a person reviews (quoted)
    Not stated on the pages read.
  5. Navan IntelligenceNavanData it names: quoted.Where a person reviews: quoted.Model or data handling: not stated.
    Our summary
    AI agents in a travel and expense platform that use trip, card and policy data to check spend. The maker says administrators see only exceptions that need attention.
    Data it names (our list)
    Trip details, Card data, Travel and expense policies
    Where a person reviews (quoted)
    admins only focus on the exceptions that truly need attention.
  6. Travel and expense monitoringOversightData it names: quoted.Where a person reviews: not stated.Model or data handling: quoted.
    Our summary
    Risk monitoring for travel and expense and card spend that prioritises high-risk items for auditors. The maker says its models are trained on anonymised transactions and auditor decisions.
    Data it names (our list)
    SAP Concur, Workday, Emburse, Mastercard, Bank card issuers
    Where a person reviews (quoted)
    Not stated on the pages read.
  7. AI expense reviewsPleoData it names: quoted.Where a person reviews: not stated.Model or data handling: not stated.
    Our summary
    Expense management software with AI review of spending against company policy. The maker names ERP integration.
    Data it names (our list)
    ERP
    Where a person reviews (quoted)
    Not stated on the pages read.
  8. Ramp Intelligence, Ramp agentsRampData it names: quoted.Where a person reviews: quoted.Model or data handling: quoted.
    Our summary
    AI agents in a spend management platform that code expenses, enforce policy and approve routine transactions. The maker says risky or ambiguous items go to a person.
    Data it names (our list)
    Expense policy, Controls
    Where a person reviews (quoted)
    Any risky or ambiguous transactions are flagged to you, fast.
  9. Rydoo IntelligenceRydooData it names: not stated.Where a person reviews: quoted.Model or data handling: not stated.
    Our summary
    Expense management and accounts payable software with AI agents and assistants. The maker says its AI handles routine work and flags decisions that need judgement.
    Data it names (our list)
    Not stated on the pages read.
    Where a person reviews (quoted)
    AI that handles routine work and flags judgement calls

Source: the makers' own pages, read 29 October 2026; notes 8 to 16.

6Questions to put to a vendor

  1. Does the product review every report and receipt, or a sample, and can we see the coverage for a period?
  2. Which policy rules are configured by us, and which are the product's own?
  3. Can it approve, reject or recover anything on its own, and can we limit that by amount or category?
  4. How are flags explained to the reviewer and to the employee?
  5. What share of flags did reviewers overturn in a pilot on our data?
  6. Is our employees' data used to train models for other customers?
  7. How does it detect duplicate, altered or generated receipts, and how is that shown?
  8. What audit trail does it keep of each decision and who made it?

Notes

  1. Internal Revenue Service, Publication 463, Travel, Gift, and Car Expenses. www.irs.gov/publications/p463. Read 29 October 2026.
  2. U.S. Securities and Exchange Commission, Staff Accounting Bulletin No. 99: Materiality. www.sec.gov/interps/account/sab99.htm. Read 29 October 2026.
  3. U.S. Securities and Exchange Commission, Commission Guidance Regarding Management's Report on Internal Control Over Financial Reporting, Release 33-8810. www.sec.gov/rules/interp/2007/33-8810.pdf. Read 29 October 2026.
  4. COSO, Internal Control, Integrated Framework. www.coso.org/guidance-on-ic. Read 29 October 2026.
  5. National Institute of Standards and Technology, AI Risk Management Framework. www.nist.gov/itl/ai-risk-management-framework. Read 29 October 2026.
  6. National Institute of Standards and Technology, Artificial Intelligence Risk Management Framework (AI RMF 1.0), NIST AI 100-1. nvlpubs.nist.gov/nistpubs/ai/NIST.AI.100-1.pdf. Read 29 October 2026.
  7. U.S. Securities and Exchange Commission, SEC Charges Two Investment Advisers with Making False and Misleading Statements About Their Use of Artificial Intelligence. www.sec.gov/newsroom/press-releases/2024-36. Read 29 October 2026.
  8. AppZen, AI Expense Audit Software | 100% Receipt Verification | AppZen. www.appzen.com/ai-for-expense-audit. Read 29 October 2026.
  9. Brex, Agents on Brex: Welcome to intelligent finance October is our biggest drop of Early Access features yet. See what’s new →. www.brex.com/platform/intelligent-finance. Read 2 November 2026.
  10. Emburse, Emburse AI | Emburse. www.emburse.com/platform/emburse-ai. Read 3 November 2026.
  11. Expensify, AI Expense Management Software | Automate Expenses with Expensify. use.expensify.com/ai-expense-management. Read 3 November 2026.
  12. Navan, Navan Intelligence | AI Travel and Expense Management Agents. navan.com/intelligence. Read 31 October 2026.
  13. Oversight, AI-powered Travel & Expense Monitoring | Oversight. oversight.com/ai-travel-expense-monitoring. Read 31 October 2026.
  14. Pleo, Smarter spending for your business - Pleo. www.pleo.io/en. Read 2 November 2026.
  15. Ramp, 10X your team's impact with AI for finance | Ramp Intelligence. ramp.com/intelligence. Read 31 October 2026.
  16. Rydoo, Rydoo | Expense Management & AP Automation. www.rydoo.com/. Read 2 November 2026.
  17. SEC EDGAR, Expensify, Inc., Form 10-K filed 26 February 2026. www.sec.gov/Archives/edgar/data/1476840/000147684026000009/exfy-20251231.htm. Read 29 October 2026.
  18. SEC EDGAR, Navan, Inc., Form 10-K filed 2 April 2026. www.sec.gov/Archives/edgar/data/1639723/000162828026023250/navn-20260131.htm. Read 29 October 2026.
  19. SEC EDGAR, OLB GROUP, INC., Form 10-Q filed 14 August 2026. www.sec.gov/Archives/edgar/data/1314196/000118518526003553/olbgroup10q063026.htm. Read 29 October 2026.

Other use cases: Forecasting, Accounts payable, Collections and cash application, Reconciliation, Close automation, Variance analysis, Contract and lease data, Tax research, Treasury and FX, Payment anomalies and fraud, Audit preparation.