The same bill, twice

Every payables team pays something twice eventually: a resent bill with a new reference, the same invoice by mail and portal. May’s scan flagged five pairs, a person spent ten minutes, and 8,940.00 stayed in the bank instead of becoming a recovery.

RuleFlag duplicate candidates by rule before payment: same supplier, same amount, close dates, similar references, then let a person keep or kill each pair.

Neverrely on the audit to find duplicates. Money recovered a year later is a favor asked of a supplier; money never paid twice is a control.

The control earns its keep on the pairs it kills. The pairs it clears cost minutes; the ones it catches cost nothing at all.

◆ May, the flagged pairsfive candidates, two real.

Review outcome	Pairs
Cleared, legitimate recurring charges	3
Confirmed duplicates, blocked	2
Pairs flagged	5

The two confirmed pairs totalled 8,940.00: one supplier re-sent an unpaid bill with a new reference, one bill arrived by both mail and portal. Neither was fraud, both would have paid twice, and the recovery email that never had to be written is the whole return on this control. Sample values are illustrative, never client data.

◆ What the rule looks atsimilarity, not identity.

Exact-duplicate checks catch almost nothing, because resent bills get new references. The rule scores similarity: same supplier and amount, dates within a window, references that differ by a suffix. Wide enough to catch the re-send, narrow enough that the review list stays five pairs, not five hundred.

◆ The owned answerthe check as a nightly query.

With invoice lines landed on the supplier star, the candidate scan is a nightly self-join, the review takes minutes, and the outcome lands back as data, so the rule’s precision is itself measured month over month.

Use case

Problem

Duplicate payments leak out through resent bills and dual channels, surface in audits a year later, and get recovered through awkward supplier conversations, sometimes.

What we build

A nightly similarity scan over landed invoice lines flagging candidate pairs for a minutes-long human review, with outcomes kept as data.

What you get

Duplicates stopped before payment instead of chased after: two pairs and 8,940.00 this month, a review list short enough to actually work, and a rule whose precision is tracked.

When did you last catch a duplicate before paying it?

We build the nightly scan that makes the recovery email unnecessary.

Talk to us

Terms on this page

AP: Accounts payable, the money owed to suppliers.
duplicate candidate: A pair the rule flags for human review.
similarity rule: Same supplier and amount, close dates, near references.
confirmed duplicate: A pair a person blocked before payment.
recovery: Asking a supplier for money back; the failure mode.