Back to blog
    Duplicate expense reports: the most mundane fraud — and the most costly

    Duplicate expense reports: the most mundane fraud — and the most costly

    By Anthony Osmar10 min read

    In short. Duplicate expense detection means comparing every receipt with every other one, across reports, employees, months and sources (corporate card transaction, out-of-pocket expense, supplier invoice), on amount, date, merchant, label and the receipt image itself. It has to run before reimbursement: a duplicate found after payment turns into a recovery procedure.

    An honest employee can submit the same expense twice without realizing it. A less honest employee knows it — and counts on nobody noticing. In both cases, your accounting department pays the same invoice twice, without ever knowing.

    The duplicate: 30% of T&E fraud, yet nobody talks about it

    According to an AppZen 2025 study covering 4.5 million audited expense reports, duplicate submissions represent 30% of detected T&E fraud cases — ahead of fake receipts (22%) and unauthorized cap overruns (18%).

    Yet this is the hardest category to spot with the naked eye. No suspicious round number. No exotic vendor. No obvious date discrepancy. The duplicate hides in the mass, exactly where a human auditor no longer has the energy to look.

    30%of T&E fraud is duplicates
    70%of duplicates are unintentional per field data
    €70average cost of a reimbursed duplicate
    0.8-4.2%of total volume detected by AI audit

    The 5 duplicate patterns your team misses

    1. The same receipt submitted two months in a row

    The employee photographs his restaurant ticket on March 28, submits it, then re-submits the same image on April 4, forgetting he already sent it. In 70% of cases per field feedback, it's unintentional. It still costs you €70 on average — in triple: the reimbursed expense, the wrongfully deducted VAT, and the tax risk.

    2. The receipt shared between colleagues

    Three salespeople dine with a client. All photograph the bill "for their file." Two submit it for reimbursement. Manual control almost never cross-checks them — they are in different files, validation dates, and sometimes different teams.

    3. Mileage + parking on the same trip

    A salesperson declares 240 km round trip Paris-Reims at standard mileage rate. He also attaches the parking ticket from Champagne-Ardenne TGV station. If he took the train, the mileage is fraud. If he drove, the TGV station parking makes no sense. Separate line control never sees the inconsistency.

    4. VAT recovered twice on the same invoice

    A subtler case: the invoice is entered in supplier accounting (VAT deducted), and the same receipt is mistakenly attached to an expense report (VAT re-deducted). Nobody sees it, because the accounting entry and the expense report live in two systems that don't talk to each other.

    5. Restaurant receipt + per diem combined

    A business trip entitles the employee to a flat per diem for meals. When the employee receives the per diem AND also submits the restaurant bill, it constitutes a duplicate — tolerated in 8 out of 10 companies, because nobody actively controls it.

    ⚠ The common thread

    In every case, isolated manual control — one report, one month, one employee — cannot spot the anomaly. Only a cross-view across all sources, all months, and all employees succeeds.

    Why manual control fails

    The unit of measurement for T&E control is the month. A batch of expense reports arrives at month-end, the auditor validates it, and moves to the next. Yet the most frequent duplicates only reveal themselves when comparing across two months, between two employees, or between an expense report and a supplier invoice.

    No human brain sustains this cognitive load across 8,000 lines per month. Nobody. Not even your best controller. That's why duplicates remain invisible: not because your teams are bad, but because the nature of the problem exceeds human scale.

    Catching after the fact: an operational cost nobody quantifies

    Some companies surface duplicates through monthly reporting or annual audits. It's better than nothing — but it's not enough. Because detecting a duplicate after reimbursement means opening a procedure, not closing an anomaly.

    Specifically, this involves:

    • Validating the duplicate with the parties. The controller must contact the employee, find the original receipts, and establish with certainty that the double submission is indeed a duplicate. For cases involving multiple employees, each file must be processed individually. In practice: 30 to 90 minutes per case.
    • Recovering wrongfully paid amounts. Once confirmed, recovery must be organized — salary deduction, incoming transfer, or accounting adjustment. Each option involves HR or accounting instruction, regulatory traceability, and often managerial validation. For 50 duplicates detected at year-end, that's dozens of hours of administrative processing — to recover amounts that should never have left.

    This operational cost is systematically absent from ROI calculations — yet it is real, it mobilizes qualified resources, and it generates social friction. Upstream detection, before reimbursement, eliminates this cost entirely: no transfer goes out, no recovery procedure opens.

    AI sees what humans cannot

    Modern duplicate detection algorithms don't just compare invoice numbers. They compute a digital fingerprint (hash) of each receipt, then perform fuzzy matching across five dimensions:

    • Amount (within €0.10)
    • Date (± 2 days)
    • Vendor SIRET
    • First 4 characters of the label
    • The image itself (perceptual hash)

    Result: a perfect duplicate, a cropped-photo duplicate, a timestamp-shifted duplicate — all surface. The engine compares the entire scope in seconds: all teams, all months, all sources. And most importantly: it does so before reimbursement.

    Concrete ROI: what you recover

    Across 18 SMEs and mid-caps that deployed comprehensive AI expense auditing in 2025, field feedback is consistent:

    0.8-4.2%duplicate detection rate by volume
    €7,200-41,000average amount recovered per year
    6-12 weeksimplementation time
    < 4 monthspayback on > 5,000 reports/year
    35-60%drop in opportunistic submissions within 6 months

    And most importantly: once employees know that 100% of reports are screened, opportunistic submissions — real fraud — drop by 35% to 60% on average within 6 months. This is the structural deterrent effect of comprehensive control.

    FAQ

    How does duplicate expense detection work?

    Each expense is compared with all the others, not only with the lines of the same report: across employees, across months and across sources (card transaction, out-of-pocket expense, supplier invoice). An exact match on the invoice number is not enough. Detection relies on fuzzy matching on amount, date within a few days, merchant, label and the receipt image, so that cropped photos and shifted dates are caught too.

    How do you detect duplicate travel reimbursements?

    Look for inconsistencies between the lines of the same trip: mileage and train station parking for the same journey, a restaurant bill claimed on top of a meal per diem, the same ride paid with the corporate card and claimed again as an out-of-pocket expense in another report. These cases only show up when lines, reports and employees are cross-checked.

    Can AI detect duplicate or fraudulent expense submissions?

    Yes. An AI audit engine computes a fingerprint of each receipt and compares the whole scope in seconds, which no controller can do by hand across thousands of lines a month. It also catches near-duplicates (cropped photo, shifted date) and, combined with other checks, amounts that do not match the receipt and AI-generated fake receipts.

    What should duplicate expense detection software check?

    At a minimum: duplicates across employees and not only within one report, month-over-month comparison, matching against card transactions and supplier invoices, a tolerance on amount and date, image comparison, and a check before reimbursement rather than after-the-fact reporting. EVA runs these checks on expense reports from SAP Concur, N2F, Lucca and Cegid.

    Is a duplicate expense always fraud?

    No. Based on the field data quoted above, around 70% of duplicates are unintentional: a receipt submitted twice by mistake, a shared bill photographed by several diners. The cost to the company is the same, which is why they should be stopped before payment without presuming intent.

    Why catch duplicate expenses before reimbursement?

    A duplicate found after payment has to be confirmed with the employee and then recovered (payroll deduction, transfer or accounting adjustment), which takes 30 to 90 minutes per case. Before reimbursement, the expense simply goes back to the employee with the reason, and no money leaves the company.

    Conclusion: the most profitable fraud to detect

    Duplicates are the most mundane, invisible, and profitable fraud to detect. They are also the category with the greatest human/machine gap: an auditor cannot do this work, AI can.

    It is also the project that best prepares your reliable audit trail. The duplicates you don't see today are the ones the tax authority will see tomorrow.

    TEVASOFT supports CFOs in deploying 100% AI T&E auditing. Free assessment of your duplicate exposure on request.