Statistics
Is expense report automation worth it?
Expense report automation is worth it for checking every claim. The $58 per report everyone quotes dates from 2015 and cannot justify it alone.
By the Foxnut team · Updated
Is expense report automation worth it?
Expense report automation is worth it for capturing and checking claims, but the published evidence does not support justifying that work on a per-report cost saving alone.
We are Foxnut Studios. We are senior people in Bengaluru and Paris who build AI systems and hand them over running. There’s a link to talk to us at the end of this article. Let’s start with what the numbers actually say.
Processing an expense report is one of the few operations tasks with a published price on it, and the number everybody quotes is $58. That figure deserves less weight than it gets. It was published in 2015, it covers a single night’s hotel stay and not an average claim, and a travel trade association produced the study in partnership with a commercial supplier.
Every article on this topic quotes $58 with the confidence of a fact that was checked once, by someone else, a long time ago. The figures that hold up better are less flattering to the usual business case.
One in five reports carries an error or a gap, at $52 and 18 minutes each to fix. Expense reimbursement is the third most common occupational fraud scheme at 13% of cases. It is also the second slowest, losing a median $1,900 a month for a median 18 months before anyone notices.
And the machine that is supposed to read the receipt scores 70% on a blurred image and 92% on a sharp one, which makes photograph quality a larger variable than model choice.
Each record below carries the figure, who was counted, over what period and the publication it came from, so you can check any row without trusting this page. What a build in this area has to hand over at the end is what running an expense system costs after go-live.
$58
It costs an average of $58 to process an expense report covering a single night's hotel stay, and the report takes 20 minutes to complete. The figure covers that one trip shape and is not a per-report average across all expense types
19%
19% of expense reports contain errors or missing information, and each one costs a further $52 and 18 minutes to correct. The same study states an average of 51,000 reports processed per company per year, and puts annual correction cost at roughly half a million dollars and nearly 3,000 hours
1 hour
A US regulator's own filing with the Office of Management and Budget puts the burden at 1 hour per travel voucher: 700 estimated annual responses against 700 estimated annual hours. This is the traveller's time to compile and file a claim, not the employer's cost to process it
2.6 million
The US federal government expects 2.6 million travel and expense transactions a year to run through a single shared service, covering more than 124 civilian agencies and more than one million federal travellers. All agencies are to migrate by June 2027
13%
13% of occupational fraud cases were expense reimbursement schemes, 306 of 2,402 cases, at a median loss of $36,000. Only two scheme types were more frequent: non-cash at 23% and billing at 21%
18 months
An expense reimbursement scheme ran for a median of 18 months before it was detected. That is the second longest of the eleven scheme types measured, behind financial statement fraud at 24 months and against an overall median of 12 months
$1,900
Expense reimbursement schemes lost a median of $1,900 a month, the second slowest of the eleven scheme types, against $9,400 a month across all schemes and $41,700 for financial statement fraud. Expense fraud is frequent and slow, not large and fast
9 months against 16
Where the victim organisation had proactive data monitoring and analysis in place, median time to detection was 9 months against 16 where it did not, a 44% difference. Median loss on the same split was $70,000 against $150,000, a 53% difference. The control was present in 49% of cases
$75
Threshold above which documentary evidence such as a receipt or paid bill is required to substantiate an expenditure under US tax rules, with lodging while travelling away from home requiring documentary evidence at any amount
60 days and 120 days
Safe harbour windows for an arrangement to count as an accountable plan: an expense substantiated within 60 days of being paid or incurred, and an excess amount returned within 120 days, are treated as within a reasonable period. Where an arrangement fails the substantiation or return requirements, all amounts paid under it are treated as paid under a nonaccountable plan and must be reported as wages on the employee's Form W-2, subject to withholding and employment taxes
50%, 70%, 76%
How closely machine-extracted values agreed with the existing labels held by a large multinational, on that organisation's own expense receipts: 50% for merchant name, 70% for receipt amount and 76% for transaction date. The same labelling method scored 94% against crowdsourced exact labels on a public 1,000-receipt benchmark. The authors describe the organisation's own labels as noisy, not exact, so the gap measures agreement, not error
70% to 92%
The same model's extraction accuracy moved with image sharpness: 70% on the blurriest receipts and 90% to 92% on every sharper band. About 90% of the multinational's receipts fell in the sharper bands, against about 60% of the public benchmark's, which is most of why scores differ between the two corpuses
| Number | What it measures | Period | Source |
|---|---|---|---|
| $58 | It costs an average of $58 to process an expense report covering a single night's hotel stay, and the report takes 20 minutes to complete. The figure covers that one trip shape and is not a per-report average across all expense types | Published 20 October 2015 | GBTA Foundation with HRS, October 2015 |
| 19% | 19% of expense reports contain errors or missing information, and each one costs a further $52 and 18 minutes to correct. The same study states an average of 51,000 reports processed per company per year, and puts annual correction cost at roughly half a million dollars and nearly 3,000 hours | Published 20 October 2015 | GBTA Foundation with HRS, October 2015 |
| 1 hour | A US regulator's own filing with the Office of Management and Budget puts the burden at 1 hour per travel voucher: 700 estimated annual responses against 700 estimated annual hours. This is the traveller's time to compile and file a claim, not the employer's cost to process it | Estimate published 28 February 2024 | US Nuclear Regulatory Commission, Federal Register, 28 February 2024 |
| 2.6 million | The US federal government expects 2.6 million travel and expense transactions a year to run through a single shared service, covering more than 124 civilian agencies and more than one million federal travellers. All agencies are to migrate by June 2027 | Announced 27 November 2024 | US General Services Administration, news release, 27 November 2024 |
| 13% | 13% of occupational fraud cases were expense reimbursement schemes, 306 of 2,402 cases, at a median loss of $36,000. Only two scheme types were more frequent: non-cash at 23% and billing at 21% | Cases investigated January 2024 to September 2025, published May 2026 | ACFE, Occupational Fraud 2026: A Report to the Nations |
| 18 months | An expense reimbursement scheme ran for a median of 18 months before it was detected. That is the second longest of the eleven scheme types measured, behind financial statement fraud at 24 months and against an overall median of 12 months | Cases investigated January 2024 to September 2025 | ACFE, Occupational Fraud 2026: A Report to the Nations |
| $1,900 | Expense reimbursement schemes lost a median of $1,900 a month, the second slowest of the eleven scheme types, against $9,400 a month across all schemes and $41,700 for financial statement fraud. Expense fraud is frequent and slow, not large and fast | Cases investigated January 2024 to September 2025 | ACFE, Occupational Fraud 2026: A Report to the Nations |
| 9 months against 16 | Where the victim organisation had proactive data monitoring and analysis in place, median time to detection was 9 months against 16 where it did not, a 44% difference. Median loss on the same split was $70,000 against $150,000, a 53% difference. The control was present in 49% of cases | Cases investigated January 2024 to September 2025 | ACFE, Occupational Fraud 2026: A Report to the Nations |
| $75 | Threshold above which documentary evidence such as a receipt or paid bill is required to substantiate an expenditure under US tax rules, with lodging while travelling away from home requiring documentary evidence at any amount | Current text as at 12 August 2026 | 26 CFR 1.274-5(c)(2)(iii)(A) |
| 60 days and 120 days | Safe harbour windows for an arrangement to count as an accountable plan: an expense substantiated within 60 days of being paid or incurred, and an excess amount returned within 120 days, are treated as within a reasonable period. Where an arrangement fails the substantiation or return requirements, all amounts paid under it are treated as paid under a nonaccountable plan and must be reported as wages on the employee's Form W-2, subject to withholding and employment taxes | Current text as at 12 August 2026 | 26 CFR 1.62-2(c) and (g) |
| 50%, 70%, 76% | How closely machine-extracted values agreed with the existing labels held by a large multinational, on that organisation's own expense receipts: 50% for merchant name, 70% for receipt amount and 76% for transaction date. The same labelling method scored 94% against crowdsourced exact labels on a public 1,000-receipt benchmark. The authors describe the organisation's own labels as noisy, not exact, so the gap measures agreement, not error | Submitted 22 November 2024, last revised 25 November 2024 | Bhattacharyya and Tripathi, arXiv:2411.14957 |
| 70% to 92% | The same model's extraction accuracy moved with image sharpness: 70% on the blurriest receipts and 90% to 92% on every sharper band. About 90% of the multinational's receipts fell in the sharper bands, against about 60% of the public benchmark's, which is most of why scores differ between the two corpuses | Submitted 22 November 2024, last revised 25 November 2024 | Bhattacharyya and Tripathi, arXiv:2411.14957 |
When is an automated expense reporting system worth it?
An automated expense reporting system is worth it when a cheap check on every claim pays back across your claim volume. Six structural dimensions decide whether a process is worth automating, and expense reports sit at an unusual corner: the individual amounts are trivial, the individual errors are cheap, and the total is neither.
This is a high-frequency process with low stakes per event, wrapped in a tax rule that makes a procedural failure expensive for the whole arrangement and not just the single claim.
| Dimension | What it reads on expense reports | Source |
|---|---|---|
| Exception variance | High, and driven by the input rather than the transaction. One in five reports arrives with an error or missing information. The receipts themselves are heterogeneous in format, language and image quality, and sharpness alone moves machine extraction from 70% to 92%. The exception is usually a bad photograph of an ordinary purchase, not an unusual purchase. | GBTA Foundation with HRS, October 2015; Bhattacharyya and Tripathi, arXiv:2411.14957 |
| Volume | High and continuous, and it scales with headcount rather than with revenue. One study puts the average company at 51,000 reports a year; the US federal government expects 2.6 million travel and expense transactions a year across more than a million travellers. Unlike a reporting cycle, nothing about this arrives on a calendar, which is why throughput arguments apply here and payback tracks claim count directly. | GBTA Foundation with HRS, October 2015; US General Services Administration, 27 November 2024 |
| Cost of an error | Small per claim, and structural in aggregate. A single correction costs $52 and 18 minutes. A single overpayment risk found by comparing a receipt against the amount an employee typed was worth about $2 a document. The number that is not small is the median $36,000 loss on an expense reimbursement fraud case, and the compliance failure, which is priced against the whole arrangement rather than the claim. | GBTA Foundation with HRS, October 2015; Bhattacharyya and Tripathi, arXiv:2411.14957; ACFE, Occupational Fraud 2026 |
| Reversibility | Reversible in principle, on a clock, and usually missed in practice. Tax rules give an employee 60 days to substantiate an expense and 120 days to return an excess before an arrangement falls outside the safe harbour. Set against that, the median expense reimbursement fraud ran 18 months before detection, so the window that governs correction is far shorter than the window in which problems are actually found. | 26 CFR 1.62-2(g)(2)(i); ACFE, Occupational Fraud 2026 |
| Regulatory exposure | Real, and it attaches to the arrangement rather than to the individual claim. Documentary evidence is required for lodging at any amount and for any other expenditure of $75 or more. Where an arrangement fails the substantiation or return requirements, all amounts paid under it are treated as paid under a nonaccountable plan and must be reported as wages, subject to withholding and employment taxes. | 26 CFR 1.274-5(c)(2)(iii)(A); 26 CFR 1.62-2(c)(3) and (c)(5) |
| Vendor market maturity | Mature enough to be bought as a single service by a government. The US federal government awarded a 15-year contract for one travel and expense shared service covering planning, authorising, booking, vouchering, audit and reporting for more than 124 civilian agencies, with every agency to migrate by June 2027. A buyer of that size consolidating onto one platform is a market that has settled on a shape. | US General Services Administration, news release, 27 November 2024 |
Three of those rows point the same way, and they are not the rows the usual pitch is built on. The exception-variance row says the hard input is a photograph, not a transaction. The cost-of-an-error row says no single claim is worth much attention. The reversibility row says the correction window is short and the detection window is long.
Together they describe a process where automation pays by doing a cheap check on every claim, continuously, and not a careful check on any particular one.
The volume row separates expense reports from the reporting and reconciliation processes alongside them. Those arrive on a calendar at a size known in advance. Claims arrive whenever somebody travels, in a stream that grows with headcount, so throughput reasoning applies and a per-claim saving multiplies. That is the honest half of the standard business case, and it is the half worth keeping.
What is the expense report approval process actually for?
The expense report approval process is for judging whether the spend should have happened. It looks like a control and mostly is not one. A manager approving a claim sees a total and a category, rarely the receipt, and almost never the comparison between the receipt and what the claimant typed.
The paper behind our extraction figures found 89 invoices in one multinational’s corpus where the amount read off the receipt was lower than the amount the employee had entered, in a system where reimbursement follows the typed figure. No approval step catches that, because the approver is not reading the image either.
This is why the regulatory rows matter. The tax rule does not ask whether a manager approved the claim. It asks whether the expense was substantiated with documentary evidence, whether it had a business connection, and whether excess amounts came back within a reasonable period.
An approval workflow that records a decision without recording the evidence behind it satisfies your own process and not the rule you are actually exposed to.
So approval and verification are different jobs, and only one of them is worth automating first. Verification compares a document with a claim, on every claim, and machines do it at a cost per document low enough to be uninteresting. Approval is a judgement about whether the spend should have happened, which needs context no receipt contains.
Automating the first does not remove the second. It gives the second something to look at.
What do most claims about expense report automation get wrong?
They quote the $58 figure as a current, universal cost per expense report, and it is neither. It is eleven years old, its own source limits it to a report covering a single night’s hotel stay, and it comes from a trade association study run with a commercial partner whose interest was in the answer.
We cite it because it is the only widely circulated per-report cost with a traceable publication and date behind it, and because you deserve to see what it actually is. A number that cannot be traced to a live primary publication with a date is not a statistic however often it is repeated. Most of the cost-per-report figures in circulation fail that test entirely.
The second error is selling expense automation on fraud prevention while quoting the wrong fraud numbers. Expense reimbursement fraud is common, at 13% of cases. It is also one of the two slowest schemes measured, at a median $1,900 a month against $9,400 across all schemes. The organisations that lose spectacular sums to fraud do not lose them here.
What stands out in that data is the duration: a median 18 months undetected, second only to financial statement fraud. The problem with expense fraud is not size but delay, and those call for different systems. A control that reduces the size of a loss is not the same as a control that shortens the time to finding it.
The third error is reading benchmark extraction scores as a forecast for your own receipts. The same labelling method that scored 94% against exact labels on a public benchmark of 1,000 receipts scored 50%, 70% and 76% on merchant name, amount and date against a real multinational’s own expense documents.
Some of that gap is the reference labels being imperfect and not the extraction being wrong, and the authors say so. Most of the rest is image sharpness: about 90% of that organisation’s receipts were sharp enough to score above 90%, against about 60% of the public benchmark’s. What decides whether receipt reading works is how the photographs are taken, which is a policy and interface question and not a modelling one.
What can’t these expense figures tell you?
They cannot tell any particular organisation what its own expense processing costs. The published per-report figure is old and narrowly scoped, the regulator’s burden estimate measures the traveller’s time and not the employer’s, and neither replaces counting the hours a specific finance team spends on a specific claim volume.
Important note: nothing here is tax advice, and we take no position on whether any particular expense is deductible, substantiated or reimbursable under any arrangement. We cite the regulations for what they require a system to keep, not for how any transaction should be treated.
They also cannot say whether the link between proactive data monitoring and shorter fraud duration is causal. The study compares organisations that had the control against organisations that did not, across cases that were all eventually detected. Organisations that monitor their data proactively differ from those that do not in many ways beyond the monitoring, and the study does not try to separate them. The direction is worth knowing, and the size of the effect is not a prediction.
How were these expense figures compiled?
We read every record from its primary source, as of August 2026. We read the trade association’s article on its own site and took its publication date from the page’s own timestamp. We downloaded the fraud report as a PDF and read it in full text, locating the scheme table, the duration figure, the velocity figure and the two anti-fraud control tables separately and not from a summary.
We read both regulation sections through the Electronic Code of Federal Regulations renderer and not a secondary restatement, and the paragraph references in the records point at the specific provisions we read. We read the Federal Register notice in the government publishing office’s text, and the extraction figures in the HTML version of the paper, which carries the results tables the abstract does not.
We dropped two sources instead of approximating them. The most frequently cited cost-per-report benchmark from a process benchmarking body could not be retrieved, because its host refuses server-side requests. No figure from it appears here, in paraphrase or taken from a search result quoting it.
The fraud report’s detection-method data is not broken out at the level of an individual scheme, so we make no claim about how expense fraud in particular tends to be found, although that would have been the most useful thing the report could say for this page.
Important note: the fraud study is a sample of cases that were detected and investigated by professionals who chose to report them. It cannot describe fraud that was never found, which matters more here than on most subjects, because the same study measures this scheme’s median time to detection at 18 months.
So should you automate expense reports?
Yes for capture and checking, but not on a per-report cost saving alone. The check is worth automating because it is cheap enough to run on every claim, and there is no other point in the process where anyone compares the document with the claim.
The overpayment risk measured on one real corpus was about $2 a document, which is uninteresting per claim and not uninteresting across 51,000 of them. The cost saving is real too, but the figure normally used to size it is eleven years old and scoped to one trip shape. A business case resting on it is resting on the weakest number on this page.
The stronger argument is about time, not money. Expense fraud is frequent, small per month and long-lived, running a median 18 months before detection. The only control in the data associated with a materially shorter detection window is continuous monitoring of the transactions themselves, at 9 months against 16, and that association is not proof of cause.
Even discounted for that, it points at the same design as the cost argument: a system whose job is to look at everything cheaply and continuously, not to look at anything carefully once. The audited artefact that build has to produce is the comparison it made on each claim and what it did about the difference. That is the record the accountable-plan rules ask for, and the record an investigation 18 months later has to reconstruct.
The brake is the receipt. Extraction accuracy on real corporate expense documents tracks image sharpness far more than it tracks anything a buyer is being sold, moving from 70% to above 90% across that one variable.
A finance team weighing this should establish three things before choosing anything: how many claims it actually processes and what the last quarter’s corrections cost, whether its own receipt images are sharp enough for a machine to read, and whether its current approval step ever compares the document against the claimed amount.
A team that can answer those is buying on evidence. A team that cannot is being asked to automate a check nobody has yet performed manually, and performing it once on a sample is cheaper than any of the alternatives.
At Foxnut Studios you would do that reading with the people who would build the system. There is one tier, so the person who signs the proposal does the work. We designed, made and launched our own products first, and what we build we hand over running, so your finance team owns the check and the record it leaves.
The first step is a free expert audit: a senior practitioner reviews your setup, with no RFP and no pitch deck. You can see what we cover at Foxnut Consulting first. If you have run that sample and want a second read on it, send it to us.
Sources
- GBTA Foundation with HRS, 'How Much Do Expense Reports Really Cost a Company?', published 20 October 2015 - read on gbta.org. A trade association study produced in partnership with a commercial travel supplier Retrieved
- Association of Certified Fraud Examiners, 'Occupational Fraud 2026: A Report to the Nations', 14th edition, published May 2026 - scheme table, duration and velocity figures read at report pages 16 to 19, anti-fraud control tables read later in the same report Retrieved
- 26 CFR 1.62-2, 'Reimbursements and other expense allowance arrangements' - current text read through the Electronic Code of Federal Regulations renderer, paragraphs (c), (e), (f) and (g) Retrieved
- 26 CFR 1.274-5, 'Substantiation requirements' - current text read through the Electronic Code of Federal Regulations renderer, paragraph (c)(2)(iii) on documentary evidence Retrieved
- US Nuclear Regulatory Commission, 'Information Collection: NRC Form 64, Travel Voucher', Federal Register document 2024-04078, published 28 February 2024 - read on govinfo.gov Retrieved
- US General Services Administration, 'GSA selects IBM to provide technology-managed service for governmentwide travel and expense management', news release dated 27 November 2024 Retrieved
- Bhattacharyya and Tripathi, 'Information Extraction from Heterogeneous Documents without Ground Truth Labels using Synthetic Label Generation and Knowledge Distillation', arXiv:2411.14957, submitted 22 November 2024 and last revised 25 November 2024 - results read in the HTML version of v2 Retrieved
Foxnut Studios takes on work like this from Bengaluru and Paris. If you want to see how we do it before you talk to anyone, here is what an AI engagement costs before you ask.