Journal entry testing on every entry, not a sample — agents do the work, the auditor decides
Every entry tested for override, not a sample, with each conclusion cited to its support.
See one case, screen by screen ↓Journal entry testing is the audit procedure that tests journal entries and other adjustments for management override of controls. External auditors must perform it under AS 2401 and ISA 240, and internal audit performs it too. Journal Entry Testing proves the ledger complete, risk-scores every entry against the audit team's own criteria and checks the risky ones against their support — seven agents do the work, and the internal auditor decides and signs.
How to choose AI solutions for back-office and enterprise work →
Updated
Every financial statement audit has to test journal entries for management override (AS 2401.58–.62, ISA 240). Today the team waits for a general ledger extract, pulls it into a spreadsheet or an analytics tool, runs a dozen filters — weekend, round amount, period end, unusual accounts — picks 25 to 60 entries and spends weeks e-mailing preparers for support. Often the support that comes back is a spreadsheet the preparer made themselves.
PCAOB inspectors keep finding the same gaps: the population not tested for completeness, top-side and consolidation adjustments left out, entries that met the criteria left untested, and criteria with no written rationale. Support is often ticked as attached without being read. Reversals after the year end and the comparison with last year are rarely done — so an entry that moves revenue across the year end and reverses a week later can pass two years running.
Estimated split for one year-end test at a large group, by hand and with the solution.
Seven agents prove the year complete, score every entry, fetch and judge the support, and draft the questions; the auditor decides and signs.
Kestrel & Vane Group — 41 legal entities on SAP S/4HANA and Oracle Fusion, financial year ended Sunday 28 June 2026. The external auditor, Brennick Hale LLP, wants internal audit's year-end journal entry testing by Monday 19 October; the audit committee meets on Thursday 22 October. Here is how the work runs, screen by screen, in the working solution.
Leah's home is a conversation, not a report. The agents scored all 186,412 entries of FY2026 and worked through the 372 selected; 41 need her decision, 7 of them with a proposal to escalate. Each agent writes what it found, with numbered citations and a proposal: the reasonableness agent on JE 900481, the pattern agent on 14 entries reversed in full after the year end, the inquiry agent with 3 drafted questions and 2 overdue replies.
Reasonableness agent: “I propose escalating it to Tomasz, and not asking Graham first.”
Before anything was scored, the population agent extracted the year from S/4HANA (33 entities) and Oracle Fusion (8) and rolled forward all 2,318 accounts. KV Treasury B.V. (KV-NL03) was out by $1,284,220.00: 212 consolidation adjustments are made only in the consolidation system, so no ledger extract can see them. It pulled them in and the difference went to $0.00. The 41 skipped document numbers in Brazil are explained by an approved IT change. Leah confirmed the population at 07:30.
186,412 entries · 1,912,604 lines · debits = credits · roll-forward ties for 2,318 of 2,318 accounts.
Twelve criteria, version 3, approved by Tomasz on 1 July — each with its points, its rule, which entries it applies to, and the paragraph of AS 2401 or ISA 240 behind it. Asked “what if just below a limit were 3% instead of 2%?”, the scoring agent re-scores the whole year: 41 more entries carry the criterion and 23 move into High. It can save that as draft 3.1 for Tomasz; nothing changes until he approves.
Scoring agent: “I re-scored all 186,412 entries with a 3% window.”
Every entry sits on a time ribbon by when it was posted, raised by its risk score, with the quarter ends and the last three days of the year shaded. The highest score of the year, 93, is on the last night: JE 900481, $249,500.00, posted Sunday 28 June at 23:12 by Graham Pellow, Assistant Controller, Americas. Its panel explains every point, starting with +12 just below an approval limit ($500 under $250,000), +11 unusual account pairing, +11 senior finance poster, +10 period end, +10 reversed after period end.
Debit 400100 Product revenue — Industrial pumps · credit 240300 Deferred revenue — current.
The support agent read 6 sources and found one attachment: Q4_rev_true-up_v3.xlsx, six customer invoices marked “pull to Q1”, “pull”, “timing” and “per GP”, totalling exactly $249,500.00. There is no contract, no acceptance term, no customer reference and no reviewer. The file was created at 22:58 by GPELLOW — 14 minutes before the entry was posted, by the person who posted it.
What the marks show: “No contract, acceptance term or customer reference.”
JE 900977 reversed the entry in full on Monday 6 July at 08:41, putting $249,500 of revenue into FY2027. A year earlier, JE 771204 for $246,000 was posted on Sunday 29 June 2025 at 22:47 by Graham Pellow and reversed by JE 771690 on 7 July — not selected by last year's top-25 test. All six orders were delivered by 26 June and no contract has an acceptance clause. The support explains neither the pairing, the amount nor the timing. Confidence 96%.
“Because the preparer is a senior finance officer and this may be an override, I have not drafted a question to him.”
The agents propose; Leah decides. She reads the cited passages, picks X4 · Possible management override from the reason codes — close, exception, escalate, or disagree with the agents — keeps the note on the record and signs as Senior Internal Auditor.
“Support does not explain pairing, amount or timing; reversed 6 Jul; same pattern FY2025. Preparer not asked.”
Tomasz Wierzbicki has it, with Leah's note and the five citations. As she chose, nobody in finance has been asked: Tomasz decides who is, and whether Brennick Hale and the audit committee are told before Thursday 22 October. Her signed decision and the notice to Tomasz are both on the audit trail.
“You escalated JE 900481 · $249,500.00” — Escalated · X4 · Signed by you.
For entries with no support, the inquiry agent drafts the question in plain words — to Daniel Kovač on JE 900118, $249,100.00, posted on a Saturday — and says why it is asking. Leah can approve and send, edit or leave it. Replies are due in 3 working days; the agent reads each one against the support. 19 questions: 3 drafted, 9 waiting, 7 answered.
“No question is drafted for JE 900481 … the audit manager decides who is asked.”
A random 25 of the 318 entries the agents explained are re-checked by a person, drawn with a recorded seed so nobody can steer them. Devika agreed with all 22 she has checked; 3 are left. Over the last four quarters people agreed with the agents on 214 of 216. Closing the 318 explained entries in one step stays locked until the sample is done.
“Do you agree, having read the support?” — I agree · I disagree…
The same year-end test, seen by the five people who carry it or rely on it — what it looked like, and what it looks like now.
Seven agents prove the year complete, score every entry, fetch and judge the support, and draft the questions; the auditor decides and signs.
Extracts the year from every ledger and the consolidation system, rolls forward each account from opening to closing balance, and checks document-number gaps, duplicates and balance.
Applies the approved criteria to every entry and writes the reason for each point; draws the random selection and the quality sample with a recorded seed.
Fetches the backup for each selected entry from the ledger attachments, the close library, the billing system and the contract repository.
Reads the support and judges three questions — does it explain the account pairing, the amount and the timing? — citing the passage for each, then proposes explained, needs the auditor or escalate.
Pairs year-end entries with what reversed them after the year end, finds the same entry by the same person in earlier years, and compares this year with last.
Drafts plain-words questions to preparers, chases on the reply time and reads each reply against the support.
Drafts the conclusion memo, the audit committee summary and the evidence pack, citing every number.
Closes every entry; the audit manager decides escalations. The agents propose; a named person decides.
Every screen has an ask box. Ask about an entry, a criterion or the population — or tell the agents what to do. Every answer cites its source, and can take you there.
What should I do first?
Start with JE 900481: it is the highest score of the year (93) and the strongest case for escalation. Then approve the three drafted questions, because preparers need three working days to reply and the evidence is due Monday. The three sample checks take about ten minutes and unlock closing the 318 explained entries.
Is the population complete?
Yes, since 2 July. 186,412 entries and 1,912,604 lines from 41 entities; the roll-forward ties for all 2,318 accounts. It did not at first: KV-NL03 was out by $1,284,220.00 because 212 consolidation adjustments never reach the ledger. I pulled them in; the 41 skipped numbers in Brazil are explained by an approved change.
Why not ask Graham first?
He prepared the entry and is a senior finance officer, and the same entry happened last year. Asking him first could warn him; AS 2401 expects inquiry of others in the reporting process too. Tomasz decides who is asked, and in what order.
Which entries moved revenue between years?
Fourteen entries were reversed in full within ten days of the year end, by the people who posted them. The largest is JE 900481, $249,500 into deferred revenue and back on 6 July, the same as JE 771204 a year earlier. I have highlighted all fourteen on the ribbon.
13 screens from the working solution, on its sample data. Pick one to see it large.
The thresholds every entry is scored and tested against are settings. A change is a proposal: the audit manager approves it, and it applies from the next scoring run.
| Setting | Default | Choose from |
|---|---|---|
| High band starts at | 50 points | Every High entry is tested |
| Medium band starts at | 30 points | Medium entries are tested by a random draw |
| Controller and CFO approval limits | $250,000 · $1,000,000 | From your group approval policy |
| “Just below a limit” window | 2% | How far under a limit counts as just below |
| Random draw | 40 Medium + 20 Low | For unpredictability; a new seed every period |
| Quality sample | 8%, at least 25 | Share of agent-explained entries a person re-checks |
| Agents may explain an entry on their own from | 85% confidence | Below it, the reasonableness agent always hands to a person |
| Reply time for a question to a preparer | 3 working days | One reminder on day 2; overdue goes to the audit manager |
Every figure is labelled: a target the solution is built to, an estimate, a typical published result, or a proven one.
“demo” = seen in the working solution, on its sample data · “target” = the design goal, measured in the live solution · “estimated” = our estimate · “typical” = published figures (ACFE, Occupational Fraud 2024: A Report to the Nations — financial statement fraud 5% of cases, median loss $766,000; owners and executives 19% of cases, median $500,000). Standards named: PCAOB AS 2401.58–.62 and PCAOB staff, Audit Focus: Journal Entries (January 2025); IAASB ISA 240 (Revised, 2025). People, companies, entries and amounts named on this page are fictional — characters and sample data in the working solution.
Journal entry testing is the audit procedure that tests journal entries and other adjustments for management override of controls. External auditors must perform it under AS 2401 and ISA 240, and internal audit performs it too. Journal Entry Testing proves the ledger complete, risk-scores every entry against the audit team's own criteria and checks the risky ones against their support — seven agents do the work, and the internal auditor decides and signs.
Every entry is scored, with each point explained. Every entry that meets the High band is tested against its support, plus a random draw of Medium and Low entries for unpredictability — in the sample year, all 312 High plus 40 Medium and 20 Low, 372 in all.
Before scoring, the population agent extracts the year from every ledger and the consolidation system, rolls forward each account from opening to closing balance, and checks debits against credits, number gaps and duplicates. Adjustments made only in the consolidation system are pulled in, and any difference goes to a person.
For each selected entry, the support agent fetches the backup from ledger attachments, the close library, billing and contracts. The reasonableness agent judges whether it explains the account pairing, the amount and the timing, citing the passage for each, and the pattern agent shows any reversal after year end and the same entry last year.
The inquiry agent does not draft a question to him. It proposes escalation, and the audit manager decides who is asked and whether the external auditor and the audit committee are told.
Yes. Only the auditor closes an entry, with a reason code and an e-signature; the audit manager approves criteria and decides escalations; no question goes out unseen. A person re-checks a random sample of the entries the agents explained, and below 85% confidence the agents always hand to a person.
That decision is the external auditor's; the solution gives them what they need to make it. They get read-only access to the completeness proof, the criteria with their written rationale, every decision and a chained audit trail, and an evidence pack with the population proof, the tested entries and their support.
The Agentic Solution Engine builds and deploys it from your requirements and documents — your methodology, approval policy and a past year of your ledger — and it goes live once every quality gate has passed. Setup tries your criteria on last year first, so you see what they would have selected.
Every sample cited, re-performed and signed before your auditor asks.
1-min demoAudit Evidence CollectionEvidence checked before the auditor sees it, and the request list closed before fieldwork.
1-min demoContinuous Controls MonitoringControl failures found in a day, not at year-end, and every fix proved by a re-test.
1-min demoWe'll run Journal Entry Testing on a past year of your own general ledger and its support.