Ask a reviewer where year-end goes wrong and the answer is depressingly consistent: the gap was there all along. The clearing account that would not clear since October. The W1 totals that stopped matching payroll in the third quarter. The GST variance that grew a little each month. Nobody found them earlier because nobody was looking earlier, and nobody was looking because checking everything continuously used to be economically impossible for an Australian firm billing time.
The classic gaps, and when they are actually created
Bank feeds coded but never reconciled, so the ledger agrees with itself and nothing else
Payroll reports drifting from the general ledger after a mid-year software change or a manually keyed pay run
The GST audit report disagreeing with lodged BAS because a prior quarter was amended in the file but never re-lodged
Intercompany and related-party loans that never square because the two files are done by two people months apart
Clearing and suspense accounts used as a parking lot in March and forgotten by June
Every one of these is created months before it is discovered, and every month of delay makes the archaeology more expensive. A variance investigated in the week it appears is one transaction to inspect; the same variance found at year end is a needle in twelve months of hay.
Sampling under pressure versus checking everything
Deadline-week review is sampling. A senior with forty files and five days spot-checks the balances that usually misbehave and trusts the rest, because that is all the hours allow. It mostly works, which is why the failures that get through are the expensive kind: unusual, material and discovered by the ATO or the next accountant rather than by the firm. An agent flips the economics. Claude Cowork reads every transaction, every reconciliation report and every lodged figure across the client base, and checking the hundredth file costs the same as the first.
What the morning exceptions run covers
Unreconciled bank lines older than 14 days, by client, with the offending transactions listed
Payroll-to-ledger variances above a dollar threshold you set, checked against lodged activity statements
GST control account movements that do not trace to a BAS label
Suspense and clearing balances that have stopped moving
Quarter-on-quarter swings sharp enough to deserve a question, even when everything technically reconciles
Each item arrives with the source references attached, so the staff member who picks it up starts at the answer, not the search. The run is a scheduled task; nobody remembers to do it because nobody has to.
A worked example: the payroll drift
A real pattern from firm files: a client migrates payroll platforms in February. The old platform's final pay run posts to the ledger; the new one starts its own journal convention. From March, W1 on the lodged BAS runs slightly above the payroll report because a termination payment was mapped to the wrong account. By June the cumulative variance is a few thousand dollars, small enough that a deadline-week sample never catches it, large enough to matter at finalisation. The continuous check flags it in March as a two-line variance with both source reports attached. The bookkeeper fixes the mapping the same week, the Q3 BAS goes out right, and finalisation in July is uneventful. The alternative timeline ends with an STP amendment in August, employee myGov queries, and a partner writing off the remediation hours.
Getting the thresholds right
The first fortnight of continuous checking is noisy by design. Every firm carries harmless standing oddities: the clearing account that always holds the merchant settlement lag, the client who reconciles monthly rather than weekly. The skill learns your materiality lines and your known-benign patterns, so by week three the report is short and every line on it deserves attention. Tuning is a one-hour conversation with a senior, not a project, and it is the difference between a report people read and a report people filter.
The cost asymmetry
A gap caught early is minutes of correction inside normal workflow. The same gap caught after lodgement is an amendment, typically $400 to $900 of unrecoverable staff time per affected client, plus the compliance exposure conversation nobody wants and the interest and penalty maths if the ATO found it first. Firms do not need many avoided amendments for continuous checking to pay for itself; most need exactly one.
Where the accountant stays in charge
The agent finds and evidences; it does not decide. Whether a variance is a coding error, a timing difference or a client conversation is judgment, and the exceptions report is built to feed that judgment rather than replace it. That boundary is what keeps the workflow inside professional standards and what makes reviewers trust the output. A fixed-fee Claude Cowork setup is $3,500 and the exceptions run is typically live in the first week. Book a brainstorm call and we will run it across three of your files as the proof.
