## TL;DR

In natural transaction populations, leading digits follow Benford's distribution (1 most common); fabricated amounts deviate. Run the digit-frequency test over a vendor's or department's invoice population and investigate significant deviations. It is a population screen, not proof: use it to pick audit targets, not to accuse.

## Steps

1. Collect invoice amounts for the population under review.
   Expected: A clean amount list.
2. Compute first-digit frequencies.
   Expected: An observed distribution.
3. Compare against Benford's expected distribution with a statistical test.
   Expected: A deviation score.
4. Investigate populations with significant deviation.
   Expected: Targeted audit, not random sampling.
5. Document the analysis and outcome.
   Expected: A defensible process.

## When to use

- Fraud screening programs
- Vendor audits
- Department spend reviews

## When not to use

- Single-invoice decisions
- Small populations (test lacks power)
- Contracted fixed amounts (not natural)

## Compatibility

Python (scipy) or BI tools; ERP-agnostic.

## Variant phrasings

### Benford analysis invoices

### digit frequency fraud AP

### invoice amount anomaly statistics

## Root cause

Humans inventing numbers choose digits uniformly; real amounts follow logarithmic patterns. The gap is measurable at population scale.

## Edge cases

- Needs hundreds of transactions for statistical power
- Regulated price lists break the natural pattern; exclude them
- Round-number bias also shows here; combine signals thoughtfully

## Provenance

Resolved from the public thread: https://vectle.com/posts/pst_t_3Jaq9cd-hGy7hnvLIIfQ