No AP process is duplicate proof. Even with strong controls, somewhere between 0.05% and 0.5% of payments at most companies are duplicates: the same invoice paid twice, often weeks or months apart.
Duplicates happen because invoices arrive through multiple channels, get processed by different AP staff, sit in two systems after an acquisition, or are issued by vendors with slight variations in invoice number or date.
Catching duplicates before they pay is the prevention angle, which the existing AP automation literature covers well. This article is about the other side: what to do when duplicates have already paid and the money needs to come back.
How Duplicate Payments Happen
Knowing the patterns helps with both detection and prevention.
Multi channel invoice arrival
The vendor emails the invoice to AP and mails a paper copy. Different AP staff process each one, treating them as separate transactions. Both pay.
Invoice number variation
The vendor issues an invoice numbered 12345. A reminder version arrives with the number 12345R or 12345A. Standard duplicate detection only matches on exact invoice numbers and misses the variation.
Date or amount variation from a credit
The original invoice was for $5,200. A credit reduced it to $4,800. The original gets paid, and the corrected invoice (with the same underlying transaction) gets paid separately.
Cross system duplication after acquisitions
The same vendor exists in two ERPs after a merger or acquisition. The vendor sends invoices to both buyer entities, which were once separate companies. Both entities process and pay before the AP consolidation is complete.
Wire payment and check payment combinations
A vendor receives a check and a wire for the same invoice, often because the vendor sent reminders to a different channel after the initial invoice and AP responded with a second payment via a different method.
Detecting Duplicates That Already Paid
Post payment duplicate detection requires running matching algorithms across historical payment data. The standard approaches:
- Exact match on vendor, amount, and invoice number across a defined time window (most basic)
- Fuzzy match on vendor, amount within a tolerance, and invoice number with character variations
- Pattern matching on identical line item descriptions and amounts even when invoice numbers differ entirely
- Cross system matching when vendor masters have been merged from multiple source systems
The fuzzy matching approach surfaces materially more duplicates than exact matching. Recovery firms use it heavily, which is one reason their findings often exceed what internal teams can identify with standard ERP duplicate detection.
The Recovery Process
Once a duplicate is identified, the recovery process has three stages.
Validation
Confirm that the two payments actually represent the same underlying obligation. Pull the supporting documentation for both payments, including the original invoices and any related POs. False positives are common in fuzzy matching, and pursuing a false positive damages the vendor relationship.
Vendor outreach
Contact the vendor with the documentation, explain the duplicate, and request a refund or credit for the duplicate amount. Most vendors will acknowledge once shown clear documentation, though the response time varies significantly.
Resolution
The vendor offers one of three outcomes: a refund check or wire, a credit note to be applied against future invoices, or a dispute if the vendor's records do not match. Each path requires different follow up.
Credit Versus Cash Refund
Vendors often prefer to issue a credit rather than refund cash, because the credit keeps the cash with them and reduces future invoiced amounts.
For the buyer, credit versus cash matters depending on the vendor relationship. With an active vendor and consistent future invoice volume, a credit is fine and may even be preferable because it avoids the cash inflow processing.
With a vendor whose relationship is winding down or whose future invoice volume is unpredictable, a cash refund is more reliable. A credit from a vendor whose next invoice may be 18 months away effectively defers the recovery indefinitely, with all the aging risks discussed earlier in this series.
The vendor will often default to offering a credit. The buyer can request a cash refund and most vendors will accommodate, particularly for older duplicates where the vendor would rather close the matter than carry the credit on their books.
Statistical Recovery Rates
Recovery audit firms publish industry benchmarks for duplicate payment recovery. The numbers vary by industry but the typical ranges for mid market companies:
- Duplicate payment rate: 0.05% to 0.5% of total payments
- Recovery rate on identified duplicates: 70% to 90% (the remainder either get disputed or written off after pursuit)
- Average duplicate amount recovered: ranges widely, but $500 to $5,000 per duplicate is typical
For a company with $200M in annual payments, that implies $100K to $1M in identifiable duplicate payments per year, of which 70% to 90% may be recoverable. The numbers are larger than most finance teams expect.
Building the Prevention Loop
Recovery only matters if the same patterns do not produce the same duplicates next year. Three controls reduce the duplicate rate meaningfully.
- Single intake channel per vendor, with vendor master flags identifying which channel each vendor uses. Multi channel arrival is the largest source of duplicates.
- Cross system duplicate checking when multiple ERPs exist post acquisition. Until the systems are consolidated, the duplicate check needs to run across both source systems.
- Periodic post payment audit, not just pre payment matching. The pre payment check catches what it is configured to catch. A periodic post payment audit catches the patterns the pre payment check missed and feeds those patterns back into the rule set.
Start Here
Run a post payment duplicate scan against the last 24 months of payment data, using fuzzy matching not just exact matching. The output will surface candidates for recovery action. Expect false positives in the 30% to 50% range, which is why validation matters before any vendor outreach.
After working through the candidates, the patterns you find go directly into the prevention control set. The recovery is the immediate cash benefit. The pattern feedback is the durable benefit.
