In short
- Typical rules: same vendor and amount within a date window; same normalized number across vendors; same amount and date across related vendors.
- Each rule finds a different kind of duplicate, so several run together.
- Fuzzy rules flag honest repeats too, so results need a quick review.
- Run on paid history first to find what already went out, then on each payment run.
- Tuning the date window and amount tolerance to your own data keeps the list short.
What the check does
An exact duplicate check asks whether the same invoice number from the same vendor is already in the ledger. Fuzzy matching loosens that question. It looks for invoices that are close enough to be the same bill entered twice, or a bill that came through two vendor records.
The usual rule set has a few shapes. Same vendor and same amount inside a date window catches a re-keyed invoice where the number was mistyped. Same normalized invoice number across vendors catches a supplier that exists under a slightly different vendor name. Same amount and same date across related vendors catches a bill that came in from a parent and a subsidiary, or from an old and a new vendor record for the same company. Normalizing means ignoring spaces, punctuation and letter case so that two versions of the same number land in the same bucket.
Each rule finds a different kind of duplicate, so you run several together. A packaging supplier bills the same amount twice in a short window with invoice numbers that differ by a character. A temp staff agency sends a weekly invoice with the same amount and a nearby date. A freight broker bills the same accessorial charge on similar loads under a related vendor record. Those are candidates for review, not findings.
Setting it up
Start with paid history. Run the rules over invoices you have already paid, then look at the hits. You will see the duplicates you missed and the honest repeats that will clog the list. That tells you where to set the date window and amount tolerance before you point the rules at live payments.
Tune the date window to your own billing rhythm. If your packaging supplier sends several loads a week, a wide window will flag every one of them. If your rent or retainer bills monthly, a narrow window may miss a payment posted late. The right window is the one that catches the double payment without burying the reviewer.
Set amount tolerance with care. A zero tolerance catches exact repeats. A small tolerance catches invoices that were re-keyed with a rounding difference, but it also pulls in legitimate fuel surcharges and freight adjustments that land near the same amount. Widen it only after you see what the paid history returns.
Decide how to treat related vendors. A parent and a subsidiary, or a vendor that changed names, can share an invoice number or an amount and date without any duplicate. Keep a short list of related vendor records and review those hits by hand. Then run the rules on each payment run.
Where it falls short
Fuzzy matching is a net, not a verdict. It catches near misses, so it also catches honest repeats: a monthly retainer from a temp staff agency, a standing order for packaging supplies, a freight broker that bills the same accessorial charge on similar loads. None of those are duplicates, and a reviewer has to say so.
The rules also miss duplicates that do not look alike. A supplier can reissue an invoice with a new number, a new date and a slightly different amount after a credit is applied. Fuzzy matching may not connect the two records. The same goes for an invoice entered under the wrong vendor record with no shared number or amount.
A long review list is the main failure mode. When the list gets noisy, clerks start clearing hits without looking, and the check stops working. Keep the rules tight, review the list daily, and log the misses so you can see which rule needs a change.
Checking that it works
Test with duplicates you already know. Pull a duplicate that someone caught by hand, or one that a vendor statement shows was paid twice, and see whether your rules would have flagged it. If they would not, find out which field broke the match: vendor name, invoice number, amount or date.
Then test the other direction. Sample the hits from a paid-history run and confirm each one with the vendor record and the payment history. Note which hits are honest repeats and which are real duplicates, but keep the sample small enough to review by hand.
Watch the list length on live runs. If reviewers can clear it in a few minutes and the hits are mostly real, the setup is working. If they are clearing without checking, the rules are too loose.
Keep a simple log: rule, hit, decision, reason. Review it with the AP manager each month. When a new supplier or a vendor name change creates noise, adjust the related vendor list or the date window. The goal is a short list that a clerk will actually read.
Questions
What is fuzzy matching for duplicate invoices?
It is a check that looks for invoices that are close rather than identical. It compares vendor, invoice number, amount and date with room for small differences. The output is a list of candidates for a clerk to review, not a final answer.
How wide should the date window be?
Wide enough to catch a late-posted duplicate, narrow enough to keep recurring bills out. Look at your paid history and your suppliers' billing rhythm, then test a window against known duplicates and honest repeats. Change it when the review list gets noisy.
Can fuzzy matching flag legitimate recurring invoices?
Yes. A monthly retainer, a standing packaging order or a freight accessorial charge can hit the same vendor and amount rule. That is why the review step matters. Keep those hits short and clear them with a reason, or tune the rule so they do not swamp the list.
Does fuzzy matching replace an exact duplicate check?
No. Exact checks on vendor and invoice number are still the first line. Fuzzy rules add a second pass for near matches, such as the same amount and a nearby date or a similar number under a related vendor. Run both, and review what the fuzzy rules return.
Updated Nov 14, 2026. Figures on this page are calculated by ERPSav from the definitions shown. The explanatory text is drafted with AI from those figures and checked automatically before it is published. General information only, not legal, tax or financial advice.