The AI tool flagged 46 entries. Is journal-entry testing done?
A tool can score every entry it was given. The interview tests whether you check what it was given.
The question
You’re the audit senior on Wrenhollow, a fictional distributor of heating and plumbing parts with a December 31, 2025 year end. The firm’s AI tool scored the FY25 journal entries and flagged 46 as high risk. The manager wants to sign off journal-entry testing today.
Analysis complete. 182,000 journal entries scored. 46 entries exceed the risk threshold and are listed for testing. All remaining entries scored low risk. No further entries require testing.
| Reason | entries | Share |
|---|---|---|
| Posted on a weekend | 18 | |
| Round amount over $50k | 12 | |
| Posted by a senior finance user | 9 | |
| Unusual account pairing | 7 | |
| Total | 46 |
Invented data. One reason per entry, the highest-scoring one.
| Entries | |
|---|---|
| General ledger, per the client’s export | 184400 |
| Received and scored by the tool | 182000 |
Invented data. Counts from the ledger export and the tool’s run log.
“The manager wants to sign off journal-entry testing today. What do you check before you agree?”
Not saved by this site. Your browser may restore text when you navigate back.
A strong answer, step by step
Open each step in turn. This is one strong answer, not the only one.
Step 1: Check the population before the results
What would you do here? Open the move you’d make.
Test the 46 flagged entries; the tool scored everything else as low risk.
Tempting, but it misses something. The tool scored 182,000 entries, not the ledger’s 184,400. Low risk on what it saw says nothing about the 2,400 it never received.
Reconcile the tool’s 182,000 entries to the ledger’s 184,400 and find out what the 2,400 are.
The strong move. Completeness first: a result is only as good as its population, and here 2,400 entries were never scored.
Ask how the vendor trained the risk model before relying on it.
Tempting, but it misses something. Worth knowing for the firm, but it can’t be answered today, and it misses the gap you can see now: 2,400 entries the model never scored.
See how a strong candidate says it
“Before the 46, I’d check what the tool tested. The ledger has 184,400 entries and the tool scored 182,000, so 2,400 never reached it. I need to know which.”
Why it works: A tool can only be right about the data it was given, so the population comes first.
Step 2: Find out what is missing
What would you do here? Open the move you’d make.
Treat them as low risk: they’re only 1.3% of the entries.
Tempting, but it misses something. Count isn’t risk. 2,400 of 184,400 is 1.3%, but entries posted after close, outside the normal system, are where management override would show up.
Identify them as post-close top-side entries and treat them as high risk for management override.
The strong move. Entries made after the ledger closes, outside routine processing, carry the override risk the standards expect you to test.
Ask the finance team whether the 2,400 entries matter.
Tempting, but it misses something. Asking management to judge the risk in management’s own entries is the problem skepticism exists for. Get the entries and judge them yourself.
See how a strong candidate says it
“The 2,400 are consolidation and top-side adjustments posted after the ledger closed, in a separate reporting system. Those are the entries management could use to override controls.”
Why it works: Auditing standards treat the risk of management override as present on every audit, and post-close top-side entries are where it would show.
Step 3: Test the flags for what they are
What would you do here? Open the move you’d make.
Clear the 18 weekend entries in bulk: the finance team works month-end weekends.
Tempting, but it misses something. Maybe, but that’s an explanation, not evidence. Check that each one has normal support before you clear it.
Test all 46 against support, starting with the nine posted by senior finance users.
The strong move. Every flag gets support, and entries by senior users carry the most override risk, so they go first.
Record all 46 as exceptions because the tool flagged them.
Tempting, but it misses something. A flag isn’t a misstatement. Recording them unexamined overstates the problem and hides which, if any, are wrong.
See how a strong candidate says it
“I’d test all 46 against support, starting with the nine posted by senior finance users. A weekend posting at month end may be routine; a round entry to revenue by the CFO is not.”
Why it works: A flag is a reason to look, not a finding; the support decides.
Step 4: Tell the manager before signing off
What would you do here? Open the move you’d make.
Sign off the 46 now and note the 2,400 for next year’s audit.
Tempting, but it misses something. The gap is in this year’s population. Deferring it leaves FY25’s highest-risk entries untested.
Tell the manager what the tool missed, what you’ll test and when you’ll finish.
The strong move. It’s specific, early and comes with a plan the manager can act on.
Test the 2,400 yourself over the weekend before raising it.
Tempting, but it misses something. Effort isn’t the issue: the manager is about to rely on a conclusion that’s wrong. Say so now.
See how a strong candidate says it
“I can’t sign off today. The tool saw 182,000 of 184,400 entries, the 2,400 top-side entries need testing, and I’ll have them done by Thursday.”
Why it works: Bad news early, with a plan, is what a manager can act on.
The answer, in short
- Decision
- Don’t sign off journal-entry testing yet.
- Because
- The tool scored 182,000 of the ledger’s 184,400 entries; the missing 2,400 are post-close top-side adjustments, the entries most exposed to management override.
- What would change it
- Support for the 2,400 and the 46 flagged entries, with no unexplained items.
Common traps
- Treating the tool’s “no further entries require testing” as a conclusion.
- Checking the results before checking the population.
- Letting management say which of its own entries matter.
- Holding bad news until it’s fixed.
Check your answer
- Did I reconcile the tool’s population to the ledger?
- Did I say why the missing entries are high risk?
- Did I treat flags as reasons to look, not findings?
- Did I tell the manager early, with a plan?
Last reviewed Nov 9, 2026