In many sets of financial numbers, a leading 1 appears far more often than a leading 9. When it doesn't, something may have been made up.
Benford's law predicts how often each first digit appears in many naturally occurring sets of numbers: about 30.1% start with 1, falling to about 4.6% for 9. Extract the first digit of each amount in Excel, count each digit, and compare with the expected percentages. Large gaps don't prove fraud, but they show where to look first.
The expected share of numbers starting with digit d is log10(1 + 1/d):
| First digit | Expected share |
|---|---|
| 1 | 30.1% |
| 2 | 17.6% |
| 3 | 12.5% |
| 4 | 9.7% |
| 5 | 7.9% |
| 6 | 6.7% |
| 7 | 5.8% |
| 8 | 5.1% |
| 9 | 4.6% |
=--LEFT(TEXT(ABS(A2),"0"),1)=COUNTIF($B$2:$B$50000,D2)=E2/SUM($E$2:$E$10)=LOG10(1+1/D2)=F2-G2A clustered column chart of F against G makes the result easy to read.
A spike at a particular digit is a lead, not a finding. Look at the amounts behind it. Common innocent causes are a large number of identical recurring charges. Suspicious causes include many amounts just below an approval limit, which often shows up as an excess of a particular leading digit. Combine Benford with a near-threshold test, as in detecting split purchase orders, and with the checks in invoice fraud red flags.
Auditors also use first-two-digit tests and statistical measures such as mean absolute deviation to judge how far a data set strays from the expected pattern. For most finance teams, the first-digit comparison and a review of the biggest gaps is enough to decide where to look.
Good starting points are supplier payments over a full year, expense claims, manual journals and refunds or credit notes. Run each separately: mixing different populations blurs the pattern. Save the results each year, so you can see whether a data set's shape is changing over time.
The test is the same in both countries and in any currency, because it depends only on the leading digit of the amount.
Risk-ranks every journal entry with an on-device AI and 17 audit tests, catches descriptions posted to the wrong account, runs Benford's law and selects a repeatable sample. Runs in your browser on your own computer: nothing is uploaded.
Score a payment request against the warning signs of fraud. Free to use online.
The Invoice & Payment Fraud Defence Kit puts every payment through a Release, Hold or Pulled-from-run decision, with a bank detail change verification log, supplier checks, an incident log, a staff training record and a Fraud Defence Playbook (PDF). Excel, no macros, one-time price.
Yes, if there are enough claims and the amounts vary widely. Exclude fixed allowances, which don't follow the pattern.
An observation that in many naturally occurring sets of numbers, smaller leading digits are more common: about 30% of numbers start with 1 and under 5% with 9.
No. It shows that a data set doesn't follow the expected pattern, which points to where to look. Many legitimate reasons can cause a deviation.
Several hundred at least, and ideally thousands. With small samples, random variation hides any real pattern.
Last updated 21 October 2026. General guidance for finance teams; check it against your own policies, systems and local rules.