Fraud Detection and Data Analytics
Where detection really comes from
Detection means finding fraud you were not told about. Tips and internal reporting are consistently the most common initial detection source, followed by internal audit and management review; external audit is comparatively rare. Know the difference between reactive detection (triggered by a complaint or anomaly) and proactive detection (continuous monitoring, analytics, surprise audits).
Analytics that matter
Core tests: duplicate payment and vendor testing (normalise names, fuzzy match, test invoice-number/amount/date combinations), gap detection in cheque or invoice sequences, Benford's Law digit tests, round-number and just-below-threshold testing, trend and ratio analysis, segregation-of-duties conflict analysis, and text mining of journal-entry narratives. Analytics flags an anomaly, not proof - every hit needs corroboration and a benign explanation ruled out.
Data quality first
Before testing, confirm completeness and integrity: reconcile record counts to source systems, check date ranges, look for blank fields. A sophisticated test on incomplete data is worthless.
Traps. Benford's Law does not apply to assigned or constrained numbers (sequential invoice numbers, capped amounts, small populations - a common rule of thumb is several hundred records minimum). Candidates also wrongly treat audit sampling as detection: sampling is built for assurance, not for finding rare fraud.
Worked example. Take 40,000 payments, count first digits, compare with Benford expectations, then investigate any excess of 4s or 9s by vendor, approver and posting date - quietly, before anyone is confronted.