Accounting technology / Glossary / Anomaly detection

What is anomaly detection in accounting?

Definition

Anomaly detection is using software to flag transactions, balances or behaviour that differ from what is expected, so people can investigate them. In accounting it highlights unusual journal entries, duplicate or out-of-pattern payments, unexpected movements in account balances and suspicious vendor or employee activity, across every transaction rather than a sample.

Also calledoutlier detectionunusual transaction detection

Compare the softwareBest audit software →All 8 audit vendors →

Methods: Rules, statistics, machine learning
Covers: Every transaction, not a sample
Used in: Audit, close, AP

How anomaly detection works

There are three main methods, usually combined. Rules test for known red flags: entries posted at weekends, round amounts, payments just under approval limits, new suppliers paid quickly. Statistical tests compare values with history or with expected distributions, such as Benford's law for first digits or a balance moving far outside its usual range. Machine learning models learn what normal looks like for a business, then score each transaction by how unusual it is. The output is a ranked list, and its value depends on precision: too many false alarms and reviewers stop looking. Auditors use anomaly scoring to choose where to test; finance teams use it during the close and in accounts payable. MindBridge scores entire ledgers for risk, Numeric flags unusual movements during the close, and Modus and Inflo apply analytics in audit testing.

Common questions

What is Benford's law?+

An observation that in many sets of financial numbers the first digit is 1 about 30% of the time, and larger digits progressively less often. Large departures from that pattern can point to invented or manipulated figures.

Does anomaly detection find fraud?+

It finds unusual items, some of which turn out to be fraud or error. Most flagged items have ordinary explanations, so people still investigate and decide.

What is a false positive in anomaly detection?+

An item flagged as unusual that turns out to be legitimate. Tuning rules and models to reduce false positives keeps reviewers focused on the items that matter.