FRAUD CATEGORY 6 OF 6
Data mismatch detection. When the receipt says one thing and the claim says another.
The most common fraud pattern in expense management is not outright fabrication. It is small discrepancies between what the receipt shows and what the employee claims. Amount off by a few percent. Vendor name slightly different. Date shifted. Category miscategorized. Each individually looks minor. Added across a company, they cost real money. REME's AI catches them at submission, before approval.
How data mismatches actually happen
Data mismatches come in four patterns. Each pattern is common. Each pattern typically slips past manual review because the discrepancy is small. Each pattern is caught by REME's engine because we compare every field extracted from the receipt against every field submitted in the claim, in real time.
Receipt amount does not match claim amount
The most common data mismatch pattern. Receipt shows $47.50. Employee claims $57.50. Off by ten dollars, small enough to overlook in manual review, especially if the receipt is handwritten or low-resolution. Our engine compares the OCR-extracted amount against the submitted amount and flags any discrepancy above your configured tolerance threshold.
Vendor on receipt does not match vendor in claim
Employee submits a claim listing 'Marriott Hotel Singapore' as the vendor. Receipt shows 'Grand Copthorne Waterfront'. Sometimes an honest mistake (employee filled the wrong dropdown). Sometimes deliberate (employee is trying to categorize under a different vendor for policy reasons). Our engine cross-checks the OCR-extracted vendor name against the submitted vendor and flags mismatches.
Receipt date does not match claim date
Receipt shows August 15. Employee claims for August 12. Sometimes just careless data entry. Sometimes trying to fit the expense within a policy window (before the fiscal cutoff, within a business trip authorization period). Our engine compares receipt date against claim date and flags discrepancies of more than one day (configurable).
Receipt content does not match expense category
Employee submits a receipt clearly showing a restaurant meal, but categorizes it as 'Client entertainment' when your policy allows client entertainment reimbursement at a higher rate than personal meals. Or submits an alcohol receipt categorized as 'Food'. Our engine analyzes receipt content against submitted category and flags category mismatches.
SEE IT IN ACTION
See data mismatch detection in action
How small data mismatches slip past manual review and how REME's engine catches every discrepancy between receipt and claim in under 200 milliseconds.
How our engine actually detects data mismatches
Data mismatch detection runs four parallel field-level comparisons on every claim submission. Each comparison has a configurable tolerance threshold your finance team can adjust. Any comparison exceeding threshold flags the claim for finance review.
Amount comparison
OCR-extracted receipt total compared against submitted claim amount. Default tolerance: 2 percent variance allowed (to account for tips added after receipt or foreign transaction fees). Configurable per company policy. Discrepancies above tolerance flag for review.
Vendor comparison
OCR-extracted vendor name compared against submitted vendor. Uses fuzzy matching for slight variations (Grand Copthorne vs Grand Copthorne Waterfront). Confidence-scored. Low-confidence vendor matches are flagged even if the primary comparison passes.
Date comparison
OCR-extracted receipt date compared against submitted claim date. Default tolerance: 1 day variance allowed (to account for late-night receipts crossing midnight or different-timezone travel). Configurable per company policy.
Category-content analysis
Receipt content analyzed (item types, merchant category, context clues) against submitted claim category. Uses AI classification against your company's expense category taxonomy. Mismatches (alcohol receipt categorized as food, personal meal categorized as client entertainment) flagged for review.
What happens when we flag a data mismatch
Finance sees the flag with the specific discrepancy. Examples:
"Amount mismatch. Receipt total: $47.50. Claim amount: $57.50. Discrepancy: 21%. Exceeds tolerance threshold of 2%."
Finance reviews. If the discrepancy is legitimate (tip added, foreign transaction fee), finance approves with a note. If unexplained, finance messages the employee for clarification.
"Vendor mismatch. Claim vendor: 'Marriott Hotel Singapore'. Receipt vendor: 'Grand Copthorne Waterfront'. Match confidence: 12%."
Low confidence match. Finance reviews to confirm this is an honest error (wrong dropdown selection) versus deliberate mis-categorization.
"Category mismatch. Claim category: 'Client entertainment'. Receipt content analysis: personal dining pattern detected (party of 1, no wine, weekday lunch). Recommended category: 'Personal meal'."
Category mismatches often reveal policy gaming attempts. Finance reviews the pattern and can push back on the category or approve if there is a legitimate reason (client canceled, employee still attended).
FAQ
Common questions about data mismatch detection
Amount comparisons: 2 percent variance allowed. Date comparisons: 1 day variance allowed. Vendor comparisons: fuzzy match confidence above 80 percent required. Category-content analysis: confidence above 75 percent required. All thresholds are configurable per company during onboarding.
The 2 percent default tolerance covers most tip and service charge scenarios. For larger tips (dining in cultures where 15 or 20 percent tips are standard), finance can raise the tolerance threshold. Legitimate discrepancies get approved with a note. The pattern is: flag, review, approve if legitimate.
For our paid business customers, yes. Receipts are stored in your company's REME account with encryption in transit and at rest. This enables cross-claim comparison (matching a receipt against previously submitted claims) and audit trail reconstruction. Retention period is configurable per company policy.
OCR errors are extraction problems (our engine misreads the receipt). Data mismatches are content problems (the receipt says one thing, the claim says another). We use confidence scoring on OCR extraction to reduce false-positive data mismatch flags (if OCR confidence on the amount is low, we do not flag as data mismatch just because it does not match the claim, we flag as OCR uncertainty instead).
The other five fraud categories
Data mismatch detection is one of six. Explore the others.
Handwritten claim validation
Coming soon
Out-of-country claim flagging
Coming soon
See it work on your own expense data
A twenty-minute demo walks through how data mismatch detection would work on your specific claim volume, typical discrepancy patterns, and configured tolerance thresholds.