The errors that survive review

Ask a controller where errors come from and you’ll usually hear about the things nobody checked. In practice, the misstatements that make it all the way to a signed set of accounts are more often the ones that were checked — and looked fine.

That’s an uncomfortable observation, because it means more review, done the same way, doesn’t reliably help.

Why reasonable-looking errors survive

Review is a sampling exercise with a human attention budget. It works well against errors that announce themselves: a transposed figure that blows out a total, a missing accrual that leaves an obvious hole. Those get caught early, usually by the preparer.

What survives is anything that lands inside the range a reviewer already expects:

• Correct amount, wrong period. The invoice is real, the figure is right, it’s booked a month late. Nothing looks odd in either month in isolation.
• Correct amount, wrong account. Expense classified one line away from where it belongs. The total is unaffected; the analysis and the tax treatment aren’t.
• Consistently wrong. An error repeated every month becomes the baseline. Variance analysis compares this month to a wrong number and reports no variance.
• Right in aggregate, wrong underneath. Two offsetting mistakes. The control total ties, so the reconciliation passes.
• Inherited. A mapping or an assumption set up years ago by someone who has left, carried forward because it has always been there.

Each of these passes the test a reviewer is actually applying, which is does this look about right. They don’t look wrong. That’s precisely why they’re still there.

Tie-outs confirm consistency, not correctness

Most routine controls check that two representations of the same data agree. The subledger ties to the general ledger. The schedule foots to the trial balance. The bank reconciles.

These are worth doing — they catch a real class of problem. But they verify internal consistency. If the underlying treatment is wrong, a consistent system reproduces the wrong answer faithfully everywhere, and every tie-out passes.

A reconciliation that has balanced perfectly for eleven months is evidence the process is stable. It is not evidence the process is right.

What actually finds this class of error

The common thread is that these errors are invisible in a single period and visible across many. The techniques that find them share that shape:

Test the whole population, not a sample. Sampling is designed for a world where reading every transaction is impossible. That constraint has largely gone. Every journal, every month, tested against the rules you care about, finds the consistent error that a sample of forty items will miss by construction.

Look at the edges of the period. Cut-off errors cluster in the days either side of a close. Entries posted after the period end, backdated entries, and unusually large items in the final days are a short, high-yield list.

Ask who, not just what. Entries posted by someone who posts no others. Entries posted outside working hours. Round-number manual journals to accounts that are normally system-fed. These aren’t accusations — they’re a sensible place to spend attention.

Challenge the inherited assumption. Pick a mapping or an accrual policy each quarter and ask why it is the way it is. If nobody can explain it, that’s the finding.

Track what changed in the model, not just the numbers. When a figure moves, the first question should be whether the business changed or the treatment changed. Those need very different responses, and only one of them is a real result.

The compliance connection

The same errors that distort management reporting tend to be the ones that create tax exposure later. A misclassification between categories with different deductibility doesn’t affect profit, so nobody notices — until a filing depends on the split.

Fixing them in the month they occur is a bookkeeping correction. Finding them two years later, across multiple filed returns, is a very different conversation.

Where this leaves the review process

None of this argues for less review. It argues for pointing review at the things review is good at — judgement, estimates, unusual transactions, anything requiring context — and handing the exhaustive, repetitive, whole-population testing to something that doesn’t get tired at four in the afternoon on the fifth day of close.

Reviewers are the most expensive and most capable resource in the process. Spending that capability re-reading transactions that a rule could have tested is the real cost of doing it the old way.

---

BookAudit automates auditing and tax compliance so finance teams can spend review time where judgement is actually required. More at bookaudit.net.