Expense Fraud Statistics October 2026

A sales manager adds a $94 client dinner to this month's expenses. The dinner never happened. Nobody checks, because $94 is the kind of number nobody checks. So next month there's another one, and another the month after that.

That's expense fraud: an employee claiming repayment for business costs that were invented or inflated. It also goes by expense report fraud and expense claim fraud, and the expense fraud statistics in this report show the same pattern every time. It isn't a heist. The typical case costs an employer $36,000 and runs for 18 months before anyone notices, a pace of about $1,900 a month.

Then the receipt got easy to fake. In March 2025, none of the fraudulent receipts flagged on one large audit platform had been made with AI. By mid-May 2026 the share was 70.8%. Small, slow and now easy to fake: that combination runs through every section below.

We took each figure back to the report, survey or release that published it, using the most recent edition available on 5 October 2026. Where the arithmetic is ours, we say so.

Key findings at a glance

• Expense fraud appeared in 13% of occupational fraud cases, 306 of the 2,402 studied, with a median loss of $36,000 (ACFE, Occupational Fraud 2026).
• The typical scheme ran for 18 months before it was caught, against 12 months for occupational fraud as a whole (ACFE, 2026).
• It drains about $1,900 a month, a fifth of the $9,400 a month that a typical fraud costs (ACFE, 2026).
• Small organisations see it nearly twice as often: in 19% of fraud cases at employers with fewer than 100 staff, and 10% at larger ones (ACFE, 2026).
• AI-generated images went from 0% to 70.8% of flagged fake receipts in 14 months, from March 2025 to mid-May 2026 (AppZen, 2026).
• AI-made fakes are smaller: roughly $100 each, against just over $180 for fakes built from a template (AppZen, 2026).
• 32% of finance professionals say they wouldn't recognise a fake expense report if it crossed their desk (Medius and Censuswide, July 2025).
• 43% of occupational fraud is caught by a tip, and 5% by examining a document (ACFE, 2026).
• Nobody measures how much expense fraud happens in total. Every figure here comes from cases that were caught, receipts that were flagged, or people who answered a survey.

How much does expense fraud cost?

The median loss is $36,000 a case. That comes from the Association of Certified Fraud Examiners (ACFE) and its Occupational Fraud 2026 report. It studied 2,402 fraud cases investigated between January 2024 and September 2025, in 143 countries and territories. Expense reimbursement schemes appeared in 306 of them, or 13%.

The ACFE's definition is narrow, and it helps to have it in front of you. An expense reimbursement scheme is one where an employee claims repayment for "fictitious or inflated business expenses". Its examples are personal travel and meals that never took place.

Median loss by type of occupational fraud, 2026

Median loss per case in US dollars, for the nine kinds of asset misappropriation the ACFE tracks

1. Cheque and payment tampering$114,000
2. Billing$90,000
3. Theft of non-cash assets$75,000
4. Payroll$63,000
5. Skimming$45,000
6. Cash larceny$45,000
7. Expense reimbursements$36,000
8. Cash on hand$20,000
9. Register disbursements$14,000

Scheme	Share of cases	Median loss	Lasted	Per month
Theft of non-cash assets	23%	$75,000	12 months	$6,300
Billing	21%	$90,000	14 months	$6,400
Expense reimbursements	13%	$36,000	18 months	$1,900
Cheque and payment tampering	10%	$114,000	13 months	$8,800
Cash on hand	10%	$20,000	9 months	$2,200
Payroll	10%	$63,000	13 months	$4,800
Skimming	9%	$45,000	12 months	$3,800
Cash larceny	8%	$45,000	12 months	$3,800
Register disbursements	3%	$14,000	12 months	$1,200

All four columns are the ACFE's own figures, from Figures 5, 8 and 9 of the 2026 report. "Lasted" is the median time before detection, and "Per month" is the median loss for each month a scheme ran. The shares add up to more than 100% because one case can involve several schemes.

The median hides a wide spread. The report's statistical table puts a quarter of expense fraud cases under $10,000 and a quarter over $100,000. The average is $113,000, pulled up by a few large cases.

Now set that against the whole study. The median occupational fraud cost $104,000, and the examiners surveyed estimate that organisations lose 5% of revenue to fraud each year. Expense fraud is about a third the size of the typical case. It's one of the cheapest frauds an employer can suffer, and one of the most common.

Why does expense fraud go unnoticed for so long?

Because each claim is small. Expense schemes ran for a median of 18 months before detection, longer than every other type the ACFE tracks except financial statement fraud.

How long each type of occupational fraud lasts before detection

Median months from the start of a scheme to its detection

1. Financial statement fraud24
2. Expense reimbursements18
3. Billing14
4. Cheque and payment tampering13
5. Payroll13
6. Corruption12
7. Theft of non-cash assets12
8. Skimming12
9. Cash larceny12
10. Register disbursements12
11. Cash on hand9

The ACFE also works out how fast each scheme does its damage. Across every case in the study, a fraud cost its victim $9,400 for each month it went undetected. Expense schemes cost $1,900 a month, the second slowest of any type.

Put the two numbers side by side and the logic of the scheme is plain. Someone tampering with cheques takes $8,800 a month and is caught in 13 months. Someone padding expenses takes about a fifth of that and lasts five months longer. Small amounts buy time.

They also add up. Eighteen months at $1,900 comes to about $34,000, close to the $36,000 median loss. That's our arithmetic, and medians don't strictly multiply, but it shows where the money goes. No single claim would trouble an approver. The total would.

The length matters for record-keeping as well. A scheme that surfaces after 18 months can only be investigated if the claims and receipts from month one still exist. That's one more reason to know how long to keep receipts.

Who commits expense fraud, and where?

Senior people, more often than junior ones, and small employers more often than large ones.

Share of each department's fraud cases that involved expense claims

Percentage of occupational fraud cases in each department that included an expense reimbursement scheme

1. Executive and upper management29%
2. Administrative support17%
3. Finance16%
4. Operations14%
5. Accounting13%
6. Sales13%
7. Purchasing11%
8. Customer service5%

Nearly three in ten fraud cases traced to an executive or upper manager involved expense claims. For customer service staff it was one in twenty. Senior people travel more, entertain more and often sign off their own team's claims. Our reading, which the ACFE doesn't state, is that few organisations audit the boss's dinner.

Size matters as much as rank. Expense schemes featured in 19% of fraud cases at organisations with fewer than 100 employees, and 10% at those with 100 or more. The ACFE points to weaker internal controls at small organisations as one likely reason.

Industry is the widest split of all:

Industry	Fraud cases involving expense claims
Education	28%
Food service and hospitality	25%
Religious, charitable or social services	24%
Agriculture, forestry, fishing and hunting	21%
Energy	18%
Technology	15%
Manufacturing	15%
Health care	14%
Transportation and warehousing	12%
Government and public administration	11%
Construction	11%
Insurance	11%
Retail	10%
Banking and financial services	7%

Banking sits at the bottom and education at the top, a fourfold gap. The ACFE lists only industries with at least 50 cases in the study. In the United States and Canada, which supplied 876 of the 2,402 cases, expense schemes appeared in 15%.

What does a real expense fraud case look like?

Two cases from US federal court records show the pattern at its largest. We've left the names out. Both press releases are linked.

A company president and his wife: $1 million over seven years. Between 2002 and 2009, the president of a Maryland electronics firm and his wife, its director of sales and marketing, charged personal spending to company cards and described it as business. One $3,007 claim for meals at business meetings was luxury bedding. According to the US Attorney's Office for the District of Maryland, he "repeatedly forged receipts and invoices", and the couple billed two arms of the same company for the same expenses. Both pleaded guilty in December 2014.

A healthcare company president: $763,887 over three years. From December 2013 to January 2017, the president of a Nashville company filed reimbursement requests for supplies he never bought and trips he never took. The US Attorney's Office for the Middle District of Tennessee says he "fabricated receipts to include with his fraudulent reimbursement requests." About $203,748 of it was for flights he didn't take. He pleaded guilty in May 2020.

Both were company presidents. Both schemes ran for years. And both ran on forged paperwork made long before any AI tool existed.

They're also far bigger than the $36,000 median. The second case works out at about $20,000 a month, on our arithmetic. A case reaches a federal court because it's large. The typical one never does.

Is expense fraud getting worse?

Not on the best long-run evidence. Across five editions of the ACFE report, expense schemes have appeared in 11% to 14% of cases, and the 2026 median loss is lower than in 2022 or 2024.

Median loss from expense fraud, by ACFE report edition

Median loss per expense reimbursement case in US dollars. The figure under each year is the share of all cases in that edition that involved an expense scheme.

1. 2018$31,00014%
2. 2020$33,00014%
3. 2022$40,00011%
4. 2024$50,00013%
5. 2026$36,00013%

Edition	Cases studied	Share with expense fraud	Median loss	Lasted
2018	2,690	14%	$31,000	24 months
2020	2,504	14%	$33,000	24 months
2022	2,110	11%	$40,000	18 months
2024	1,921	13%	$50,000	18 months
2026	2,402	13%	$36,000	18 months

Read this as five snapshots. Each edition covers the cases that certified fraud examiners chose to report, so a change between two editions can come from the sample as much as from the world.

Two things hold across all five. The share barely moves. And detection got faster once: 24 months in 2018 and 2020, then 18 months in every edition since.

There's a comparison we haven't seen made anywhere else. Set each edition's expense fraud median against that edition's median for all fraud:

Edition	Expense fraud	All occupational fraud	Expense fraud as a share
2018	$31,000	$130,000	24%
2020	$33,000	$125,000	26%
2022	$40,000	$117,000	34%
2024	$50,000	$145,000	34%
2026	$36,000	$104,000	35%

The two medians are the ACFE's. The share is our arithmetic. In 2018 and 2020 an expense scheme cost about a quarter of the typical fraud. In every edition since, it has cost about a third. So the fall to $36,000 in 2026 isn't expense fraud shrinking. The typical fraud fell with it, and expense fraud held its place.

What the series can't show yet is the AI receipt. The 2026 cases were investigated between January 2024 and September 2025, and a scheme that takes 18 months to surface began well before that. On the timeline in the next section, most of them started before AI-made receipts existed in any number.

How common are AI-generated fake receipts?

On one large audit platform, they went from none to most in 14 months. AppZen makes software that audits expense reports for large companies. In a 2026 white paper, it says AI-generated images were 0% of the fraudulent receipts its customers flagged in March 2025 and 70.8% by mid-May 2026.

When	AI-generated share of flagged fake receipts
March 2025	0%
April 2026	Passed template-made fakes for the first time
Mid-May 2026	70.8%

Before that, the usual fake was a template: a receipt layout filled in by hand. In AppZen's words, "a previously marginal receipt-faking method replaced the dominant one."

The sample behind the 70.8% was reported by PYMNTS in June 2026: 1,471 AI-generated fake receipts, submitted by 745 employees at 174 companies, claiming $148,143 between them. Divide those figures and you get about $101 a receipt, two fake receipts per employee, and about $199 per employee. That's our arithmetic. These are not big-ticket frauds.

They're getting smaller, too.

Average value of a fake receipt, by how it was made

Average amount claimed per flagged fake receipt, US dollars

1. Made from a template$182
2. Generated with AI$100

An AI-made fake claims a little over half what a template-made one does. That fits everything in the first half of this report. A claim near $100 is small enough to be waved through, and large enough to need a receipt at all. US tax rules don't ask for one on most business expenses under $75, and many employers set their own no-receipt limit lower still. The fakes cluster just above the line where paper starts to matter. That last point is our reading of the numbers, not a finding AppZen published.

Why does another credible source say under 5%?

Because it's counting different documents. Inscribe checks the paperwork people send to banks and lenders, such as bank statements and pay stubs. In its 2026 State of Document Fraud Report, AI-generated files were less than 5% of the fraudulent documents it detected in 2025.

Source	What it screens	AI-generated share	Period
AppZen	Expense receipts flagged as fraudulent	70.8%	Mid-May 2026
Inscribe	Fraudulent documents sent to financial institutions	Under 5%	2025

Templates were the story at Inscribe too, in the other direction. Template-based manipulation showed up in one in five flagged documents in 2025, up from one in 14 the year before. Through the first half of 2026, Inscribe says template fraud still ran two to three times higher than AI-generated fraud.

So both numbers are right, and the gap between them is the finding. A bank statement has to agree with itself across every page and with the bank's own records. A receipt only has to look right.

We can put a number on how little there is to copy. When we measured 632 real receipts, the average one held 21.1 elements, and 37.3% carried a barcode. Nothing on a receipt has to add up against anything except its own total. Our layout research describes a receipt as a text document with a handful of ornaments. That makes it the easiest financial document there is to imitate.

The same measurements say what a reviewer can't rely on:

On a real receipt	Share of the 632 we measured
A logo	52.8%
A barcode	37.3%
A QR code	0%

A receipt with no barcode isn't suspicious. Most real ones don't have one, and nearly half carry no logo. There's no single feature whose absence gives a fake away, which is the conclusion the detection figures further down reach from the other side.

Two practical consequences:

• Don't quote 70.8% as the share of document fraud that is AI. It's the share of flagged fake receipts on one expense platform in one month.
• Don't read "under 5%" as reassurance about expenses. It describes documents sent to financial institutions in 2025, before the receipt figures turned.

How many employees admit to faking a receipt?

In one 2026 survey, four in ten US workers. Treat that number with care.

Emburse, an expense software company, commissioned Atomik Research to survey 1,000 professionally employed adults in the US and 1,000 in the UK between 5 and 8 May 2026. Among the US respondents:

What they said they had used AI to do	Share
Invent a purchase that never happened	19%
Inflate the value of a real purchase	15%
Replace a lost receipt for a genuine purchase	6%
Any of the three	40%

Of those who had, 40% used an AI tool their employer pays for. And 9% of respondents said they had built their own tool to make receipts.

The survey also asked the older question. 27% of US respondents admitted passing off a personal purchase as a business expense because of their finances. In Emburse's 2024 survey the figure was 24%. A different pollster ran that one, with a different sample, so the three-point rise is weaker evidence than it looks.

Three reasons for the care. The answers are self-reported, from an online panel. The release doesn't say how many of the respondents file expenses at all. And a figure as high as 40% sits oddly beside AppZen's sample, where the AI fakes came from 745 employees across 174 companies. The two measure different things, and we haven't tried to reconcile them.

The third row of that table deserves a second look. Six percent used AI to replace a receipt for something they really bought, and the survey counts it as a fake receipt. That's the right call. A real purchase doesn't make a made-up document genuine. If the original is gone, the honest route is a lost-receipt declaration, which finance teams accept every day.

One more figure explains some of the motive. Among respondents who had passed off personal purchases, 51% had paid overdraft fees, late fees or card interest while waiting for their employer to reimburse them.

Can finance teams tell a fake receipt from a real one?

Many say they can't. In a Censuswide survey for Medius of 1,000 finance professionals in the US and UK, run in July 2025:

• 32% said they wouldn't recognise a fake expense report if it crossed their desk
• 30% had seen a rise in faked receipts since the start of 2024
• 42% had suspected a colleague of submitting a fake or altered receipt
• 34% had been pressured to approve an expense that didn't seem legitimate
• 29% admitted bending the rules themselves, by rounding up figures or reclassifying personal costs

That last line is finance staff describing their own claims.

The people who approve expenses are also far surer that AI is involved than the people who file them. SAP Concur's 2026 Global Business Travel Survey asked three groups whether it's at least a bit likely that employees are using AI to falsify expenses or receipts.

Who thinks employees are using AI to falsify expenses

Share saying it is at least a bit likely, by role

1. CFOs92%
2. Travel managers86%
3. Business travellers66%

What actually catches expense fraud?

People, mostly. 43% of the frauds in the ACFE's 2026 study came to light through a tip, and more than half of those tips came from employees.

How occupational fraud is first detected

Share of cases by initial detection method, all scheme types

1. Tip43%
2. Internal audit15%
3. Management review13%
4. Document examination5%
5. Account reconciliation5%
6. By accident4%
7. Automated data monitoring3%
8. External audit2%

Look at the highlighted bar. Only 5% of frauds were caught by someone examining a document. Meanwhile 39% of the fraudsters in the study had created fraudulent physical documents to cover their tracks, and 33% had altered real ones.

Faking paperwork was the normal way to hide a fraud long before AI. And looking at the paperwork was never what caught it.

Retail learned the same lesson at the returns desk. A shop doesn't judge a receipt by its look. It matches the receipt to a sale in its own system, and when there's no receipt at all it falls back on ID checks and store credit. Our retail returns report has the fraud figures, and our guide to returning something without a receipt shows the controls from the shopper's side.

What does work shows up in the ACFE's comparison of organisations with and without each control:

Control	In place at	Loss with	Loss without	Lasted with	Lasted without
Management review	71%	$84,000	$186,000	10 months	18 months
Proactive data monitoring	49%	$70,000	$150,000	9 months	16 months
Surprise audits	44%	$74,000	$149,000	8 months	16 months
Hotline	73%	$100,000	$150,000	12 months	17 months

"In place at" is the share of victim organisations that had the control. Losses and durations are medians, across every kind of fraud in the study.

Organisations where a manager reviews transactions lost 55% less than those without. Those running proactive data monitoring lost 53% less and caught the fraud in 9 months, not 16.

For an expense process, that points away from staring harder at receipts. It points towards checking each claim against something the claimant can't edit: the card feed, the calendar, the hotel folio, the same receipt turning up twice. That's our reading of the ACFE figures, and it fits the AI data. A fake receipt can look perfect and still describe a dinner that no card ever paid for.

That check gets harder as money leaves the card networks. A dinner settled over Zelle or Venmo leaves a transfer log and no receipt, a gap we cover in our peer-to-peer payment report.

Where a receipt tool stands in this

Repceipt makes receipt templates, so part of this report is about tools like ours. The template-made fakes in AppZen's data come from our product category. We'd rather say that plainly than leave it for a reader to notice.

Our terms put the line at deception. Nothing made with Repceipt may be given to an employer, a client, an insurer or a shop as if it were the original. What the tool produces is a mockup, and it isn't proof of purchase.

Our guides hold the same line. The advice in what to do when you lose a receipt is to ask the merchant for a duplicate, use the statement, or file a declaration. A rebuilt copy is for your own records and nobody else's.

What it all adds up to

Expense fraud was never dramatic. It's $36,000 taken at $1,900 a month over a year and a half, most often by someone senior, most often where nobody double-checks.

The receipt was the weakest part of the claim long before 2025. Two in five fraudsters were already creating fake documents, and one fraud in twenty was caught by examining one. What changed is the cost of the fake. On AppZen's platform, the template gave way to the AI image in 14 months, and the amounts got smaller as it did.

None of the numbers suggest the answer is sharper eyes. They point to three things. Check claims against records the claimant doesn't control. Pay people back quickly enough that fewer are tempted. And keep an honest route open for the person who really did lose the receipt.

How we compiled this report

Every source was read on 5 October 2026. Here is what that involved.

• We read the ACFE's tables, not summaries of them. We downloaded all five editions as the ACFE published them and took each figure from the report's own figures and tables. Where a chart's labels were jumbled in the text of the file, we checked the page image before using a number from it.
• Vendor figures were traced to the vendor. Each AppZen, Inscribe, Emburse, Medius and SAP Concur number comes from that company's own page or release. Where we could only reach press coverage of a figure, the source is marked secondary in the list below.
• We read each survey's method before its headline. The pollster, the sample and the fieldwork dates are recorded with every survey we cite.
• Our arithmetic is labelled as ours. That covers the per-receipt and per-employee averages, the 18 months multiplied out, the share-of-typical-loss column and the monthly figure for the court case.
• What we couldn't trace is left out and listed. The next section is that list.

What we could not verify

• The total size of expense fraud. No source measures it. The ACFE studies cases that were detected and investigated, and its 5% of revenue figure is examiners' estimate for all occupational fraud, not for expenses.
• A real trend line. The five ACFE editions use different case samples, so we show them as snapshots and haven't calculated growth rates between them.
• The ACFE's two medians. Figure 5 of the 2026 report gives a $36,000 median across 306 expense cases. Its statistical table gives $35,000 across 169. We use the first and haven't found the reason for the difference.
• AppZen's denominators. Its white paper doesn't say how many flagged receipts or customers sit behind the 70.8%, and it publishes no monthly series between March 2025 and April 2026. The sample counts above reached us through PYMNTS, which cites AppZen and Accounting Today.
• How many fake receipts are made in total. AppZen says more than 3.5 million were created on four receipt-generator sites in six months. It gives no method, so the figure isn't used above.
• Templates' share before 2025. Some coverage says template-made receipts were 95% to 100% of flagged fakes before AI. We couldn't trace that to a release we could read, so it isn't here.
• Emburse's sample. The release doesn't say how many respondents submit expense claims, and we haven't used its UK figures, which sit in a separate post.
• When Medius published. Its release carries fieldwork dates of 4 to 14 July 2025 and no publication date.
• How well detection software works. Several vendors quote accuracy rates for spotting AI-made receipts. None has been independently tested that we could find, so none appears here.
• Figures by country. Outside the ACFE's regional split, we found no reliable national figures for expense fraud.

Corrections and updates

Published 5 October 2026. No corrections have been made since.

If a figure here is wrong, or one of these sources has published a newer edition, tell us. We'll correct it and record the change here.

<a href="https://www.repceipt.com/reports/expense-fraud-statistics">Expense Fraud Statistics October 2026</a> — Repceipt