Spotting unusual transactions
Listen to this lesson
Nigerian voice narration
Fictional training case. Internal audit asks you to review 60 payment vouchers for anything unusual before external auditors arrive next week. This lesson shows how AI can surface patterns worth investigating, while making clear these are leads, not conclusions. Work through the example, identify the evidence needed and practise the method with cleared or synthetic inputs. Any additional figures in the explanation are separate practice examples, not claimed rows in a downloadable file.
Worked example
Olu feeds the 60 anonymised voucher records into AI and asks it to look for split payments just below the approval limit, duplicate payees under different account numbers, weekend payment dates, and suspiciously round-number amounts. The AI returns an exceptions schedule flagging four vouchers, including two payments of ₦499,000 each to the same vendor paid a day apart — just under a ₦500,000 approval threshold. Olu does not treat this as proof of wrongdoing; he pulls the original vouchers, checks the approval trail, and documents his findings in the exceptions schedule for the audit file, since the pattern alone only justifies deeper investigation.
The method
Ask AI to scan for named red-flag patterns explicitly — split payments near approval limits, duplicate payees, weekend dates, round numbers — rather than a vague 'find problems' request. Treat every flagged item as a lead requiring document-level investigation, never as proof on its own. Record findings in a structured exceptions schedule for the audit file.
Worked practice and common mistakes
The approval limit in a practice case is fictional; actual limits come from current policy. Similar amounts and close dates justify investigation, not a verdict. A weekend posting may differ from the authorisation date. Document voucher IDs, the observation, source checks and conclusion, including cases cleared after investigation. Preserve the original evidence rather than rewriting it to match the flag.
Review before using the result
AI can scan 60 vouchers faster than manual review for known red-flag patterns, but flags are leads, not verdicts. Two payments just under an approval limit deserve investigation, not automatic suspicion. Document every finding in an exceptions schedule.
Your workplace practice
Use the recommended free assistant with this lesson's fictional example. Ask for the output described in the worked example. Keep a copy of the input and draft. Compare the result with the verification steps above, record any correction, and retain placeholders where evidence is missing. The expected result is a checked draft for human review, not an automatically approved decision. Course examples and practice responses are fictional; amounts, dates, rules and thresholds are not official guidance or supplier quotations.
Why this tool: Can suggest possible outliers from a small fictional table; verify every flag against the records.
Hit a free limit? Switch to ChatGPT or Gemini and continue — the prompt works in any assistant.