Benford's Law Fraud Detector

What is Benford's Law Fraud Detector?

Benford's Law Fraud Detector tests whether the leading digits of a numeric column follow the distribution that naturally occurring numbers tend to follow — about 30% starting with 1, falling away to under 5% starting with 9. It reports a chi-squared conformity verdict, charts expected against observed, and lists the rows driving the deviation. It is a screening tool used in forensic accounting and audit: a failed test is a reason to look closer, never evidence of anything on its own.

How to use Benford's Law Fraud Detector

1Load the data and pick the numeric column to test.
2Check the sample size first — Benford analysis is unreliable below a few hundred values.
3Compare the observed first-digit distribution against the expected curve.
4Read the chi-squared statistic and the conformity verdict.
5Review the rows contributing most to the deviation to see whether there is an innocent explanation.

Features

First-digit frequency analysis against the Benford distribution
Chi-squared test with a stated conformity verdict
Expected-versus-observed chart
Drill-down to the rows driving the deviation
Sample-size warning when the test would not be meaningful
Entirely browser-based, so financial data stays local

Common uses

Screening expense or invoice data for anomalies worth auditing
Teaching Benford's law with real data rather than a textbook table
Checking whether a numeric dataset looks naturally generated
Prioritising which accounts to examine manually

Used by

Data analysts
Researchers
Business analysts
Students
Analysts

Common errors

Running it on data that has no reason to follow Benford — assigned numbers like invoice ids, zip codes, phone numbers, and values with enforced minimums or maximums all fail legitimately.
Testing too small a sample; below a few hundred values the chi-squared result is dominated by noise.
Treating a failed test as proof of fraud. It is a screening signal, and the overwhelmingly common explanation is that the data simply is not Benford-distributed.
Including negative numbers or zeros without deciding how to handle them, which distorts the first-digit counts.

How it works

1The leading significant digit is extracted from each value, ignoring sign and leading zeros.
2Observed frequencies for digits 1 through 9 are compared against Benford's expected proportions, which follow log10(1 + 1/d).
3A chi-squared statistic summarises the total deviation, and per-digit contributions identify which digits — and therefore which rows — are responsible.

Recommended workflow

1
Check the column's statistics
Distribution, range and outliers for the same column.
Open tool
2
Find the extreme values
Outlier detection by IQR and z-score.
Open tool

Frequently asked questions

Why should 1 be the leading digit 30% of the time?: Because many real quantities grow multiplicatively and span several orders of magnitude. Under that kind of growth a value spends longer between 1 and 2 than between 8 and 9, so leading 1s accumulate. Formally the expected proportion is log10(1 + 1/d).
Which data should not be tested?: Anything assigned rather than measured or accumulated — invoice numbers, ids, postcodes, phone numbers — and anything with imposed bounds, like values capped at an approval limit. These fail the test for entirely innocent reasons.
Does failing the test mean fraud?: No. It means the digit distribution is not Benford. That is a prompt to look at why, and the most likely answer is that this data was never expected to conform.
How many values do I need?: A few hundred at minimum, and ideally over a thousand. The tool warns you when the sample is too small for the chi-squared result to mean much.