How cross-card duplicates happen
Most duplicate expenses are accidents produced by the plumbing. A corporate card charge flows into the expense tool from the card feed, and the employee separately forwards the email receipt, which the tool turns into a new out-of-pocket line. Now the same ride exists twice, once as a card transaction and once as a reimbursement claim.
The other common source is shared expenses. A manager pays for a team dinner on their corporate card, and a colleague who kept the itemized receipt submits it as an out-of-pocket claim, believing they were helping. Neither person did anything wrong on purpose, and the company still pays twice.
Intentional duplicates exist too. The ACFE Report to the Nations finds expense reimbursement schemes make up 14% of asset misappropriation cases, with a $40,000 median loss and a median of 14 months before detection. Multiple reimbursement is one of the classic patterns.
Why exact-match rules miss most of them
Many expense tools flag duplicates only when merchant, date and amount match exactly. Real duplicates rarely do. The card posts a day or two after the transaction, the merchant descriptor reads "UBER *TRIP HELP.UBER.COM" while the receipt says "Uber Technologies", and the amounts differ because of tips or currency conversion.
Restaurant charges are a good example. The card authorizes the pre-tip amount, then settles at the final amount after the tip is added. An employee who photographs the printed check before writing the tip will submit a different number than the card shows.
Signals that actually match
Good duplicate detection scores candidate pairs on several weak signals rather than requiring one strong one. No single signal is decisive, but three or four together are hard to explain away.
Weight the signals by how hard they are to fake by accident. A matching folio number or order ID is close to proof, while a similar amount on the same day is only a reason to look. Tune the tolerance bands on your own data: a team that travels internationally needs wider FX bands than one that drives to customers in the same state.
- Normalized merchant: map card descriptors and receipt names to the same entity
- Amount within tolerance: allow for tips of up to about 25% and for FX conversion
- Date window: transaction date versus posting date, usually zero to three days
- Last four card digits printed on the receipt versus the card on file
- Trip identifiers, folio numbers or order numbers that appear on both documents
- Near-identical receipt images, even when cropped or rotated
A worked example at Halvorsen Biologics
Halvorsen Biologics ran a matching pass over one quarter of reports from its 220-person commercial team. Three pairs illustrate the patterns. In the first, Daniel Okafor's corporate card shows an Uber charge of $47.83 posted 10 July, while his report also contains an out-of-pocket Uber claim of $41.20 dated 9 July. The receipt email in the claim shows the same pickup time and route, and $41.20 was the pre-tip fare.
In the second, Colin Ashby paid $286.14 for a team dinner on his card, and Wen Li submitted the itemized receipt for $286.14 as her own out-of-pocket expense two days later. Same amount, different employees, so a per-employee check would never pair them.
In the third, Nadia Rahman settled a Frankfurt hotel folio of €412.60 on her corporate card, which posted as $447.92, and also attached the folio to an out-of-pocket line in euros. The folio number matched exactly. Across the quarter, the three patterns together accounted for $3,186.55 in duplicate claims.
Handling a hit without accusing anyone
Treat every match as a question until proven otherwise. Two Uber rides at similar amounts on the same day can be a legitimate round trip, and two dinners at the same restaurant can be two different nights in a long conference week.
The message to the employee should show both lines side by side and offer a one-click way to remove the duplicate or explain the difference. Escalate to internal audit only when a pattern repeats across periods or the employee deletes evidence instead of answering.
Where detection runs in Phyto Nex AI
Phyto Nex AI compares every new out-of-pocket line against card transactions and claims across all employees, not just the submitter, using the signals above. When it finds a likely pair, it shows finance both documents and drafts a neutral message. Finance decides whether to send it and whether any amount is adjusted.
The fastest win, though, is upstream. Configure the expense tool so forwarded receipts attach to an existing card line when one matches, rather than creating a new out-of-pocket line by default. That single setting removes the most common source of accidental duplicates before anyone has to review them.
Rohan Mehta, Staff Engineer, Duplicate Detection
Rohan builds the matching systems that pair card transactions, receipts and reimbursement claims across employees and payment methods. Before Phyto Nex AI he worked on payments reconciliation at a card issuer.