Field notes · 11 May 2026
When a journal looks ordinary but fails a tie-out
How we use AI-assisted anomaly detection to catch recurring journals that break reconciliation tie-outs before audit fieldwork.
Auditors rarely start with the dramatic fraud story. They start with a recurring accrual that no longer matches the subledger, or a payroll clearing account that drifted by a few basis points each month until the year-end balance looked “immaterial” — until sampling landed on it.
In our pre-audit anomaly reviews we ask the model to surface journals that are routine in description but irregular in amount, timing, or counterpart account. The human review then asks a simpler question: does this still tie to the reconciliation binder?
Last spring a Kaohsiung trading client had twelve months of “rent accrual true-up” journals. Descriptions were identical. Amounts were not. Three of them posted to a prepaid account that the reconciliation treated as zero. The anomaly pass ranked those three above larger one-off entries because the pattern had broken, not because the absolute value was huge.
If your close team is polishing variance narratives while the tie-outs still wobble, fix the wobble first. Narratives written on top of broken clearings become the questions auditors ask twice.