title: "Workpaper-review accelerator for a Big-4 AU audit team" dek: "A 2025 build that cut Q4 workpaper review from 38 hours to 9 hours per audit. Deployed across 14 audit engagements in the first six months." sector: "Professional Services" client: "Big-4 AU audit practice · in-house engagement team" engagement: "Pilot → Practice Retainer" duration: "20 weeks" year: "2025" outcome: "Q4 workpaper review: 38 hrs → 9 hrs per audit · 14 audits live in 6 months" solution: "Workpaper-classification pipeline + control-test verifier + variance-explanation generator with senior-auditor sign-off queue." timeSaved: "~29 hours per audit · ~A$11.6K saved per engagement" visual: "none" cardFigure: "compliance" timeMetric: "29 hrs" timeMetricLabel: "saved / audit" costMetric: "A$240" costMetricLabel: "cost per audit" speedMetric: "4.2×" speedMetricLabel: "faster review" publishedAt: "2025-07-09" keywords:
- audit automation Big-4
- workpaper review AI
- ASA standards AI
- audit technology Australia
The problem
A Big-4 AU audit team — managing a portfolio of ASX-listed and APRA-regulated client audits — was watching its Q4 review cycle eat the same engagement partner hours every year. Workpaper review took an average of 38 hours per audit at the partner / senior-manager level. Most of those hours were spent on the same activities: confirming control-test evidence was attached, verifying variance explanations against actual ledger movements, and cross-referencing supporting documents to ledger entries.
The work was repetitive, but it was also professional-judgement work covered by Australian Auditing Standards (ASA). A junior wouldn't have been signed off doing it. The partners were genuinely needed — but not for all of it.
What we did
Five weeks of scoping (rare for us; the audit-standards mapping was rigorous), twelve weeks of build, three weeks of pilot. The deployed system runs inside the firm's audit-platform environment and:
- Classifies each workpaper by ASA-mapped category (control testing, substantive analytical, journal-entry testing, etc.)
- Verifies that control-test workpapers have the expected evidence components (test description, sample selection rationale, exception detail) and flags missing components
- Cross-references variance-explanation narrative against the underlying ledger movement — flagging where the narrative does not reconcile to the figures
- Generates a draft "areas of audit focus" summary for the engagement partner before review starts
- Routes everything below threshold into a partner-review queue with the specific concern highlighted
The partner never sees a clean approval. The system surfaces what to look at first, not what to skip.
The outcome — after 14 audits
| Before (FY24 average) | After (FY25 average across 14 audits) | |
|---|---|---|
| Partner / SM hours per audit review | 38 | 9 |
| ASA-flagged issues caught at review | ~14 per audit | ~22 per audit (system raises more, partner confirms fewer) |
| Variance-explanation mismatches caught | n/a | 38 across the portfolio (vs. 6 in the prior year's manual review) |
| Cost per audit run through the system (model + infra) | n/a | ~A$240 |
| Estimated review-hours reclaimed across portfolio | n/a | ~406 partner-equivalent hours in 6 months |
The reclaimed time has gone into two things: more substantive review of higher-judgement areas (going concern, complex provisions), and earlier client communication about audit findings. Both, the engagement partners report, have measurably improved client perception of the audit experience.
The system doesn't replace any judgement. It tells me where to spend my judgement. That's exactly the right division of labour.
— Engagement Partner, Big-4 AU audit practice
What we'd do differently
Map ASAs before sample workpapers. We spent the first three weeks on real workpaper samples and the ASA-mapping later. In hindsight, the ASA mapping should have been the first artefact — the workpaper samples would have been read against the standards from week one.
Senior-manager pilot before partner pilot. We piloted with engagement partners first, because they were the time-saving target. Senior managers would have spotted the routing issues earlier and the system would have hit production-readiness two weeks sooner.
What we didn't do
We didn't deploy any system that signs off audit conclusions. We didn't bypass any ASA. We didn't ingest client data outside the audit-platform environment.
The firm's professional-standards team reviewed the audit-log design before the Pilot launched and signed off in writing. Every model decision is reconstructible from the log three years post-engagement, which is the firm's standard retention period.
The interesting work was not the AI. It was the ASA mapping and the audit-log schema. That's now the template for every Professional Services audit-adjacent engagement we run.