Claude for Auditors: Workpaper Review Prompts (ACRA-Aligned)
Six Claude prompts for audit workpaper review, sample testing, and exception summaries. ACRA-aligned. Confidentiality-first prompt patterns for Singapore audit firms.
Audit work breaks into a few distinct tasks: planning, risk assessment, sample selection, substantive testing, analytical procedures, workpaper documentation, and conclusion forming. Claude's leverage varies sharply across these.
High leverage: workpaper review (consistency, completeness), variance commentary on analytical procedures, sample-summary writing, exception summaries, draft management letters, audit-committee report drafting.
Medium leverage: risk assessment (Claude can ideate; humans must verify), sample selection (rationale-writing helps; selection itself is professional judgement).
Low leverage: anything requiring real-time client interaction, judgement on materiality, or final conclusions on going concern / impairment.
Apply Claude where the work is structured, repetitive, and the senior associate's time is currently the bottleneck.
Six prompts that work
1. Workpaper completeness review.
*"Below is a draft workpaper for [audit area, e.g. revenue testing]. ACRA expectations: clear objective, procedures performed, evidence obtained, conclusion. Review the workpaper and identify: (a) missing required elements, (b) unclear or unsupported conclusions, (c) procedures that don't connect to the conclusion, (d) potential ACRA inspection risk areas. Output as a structured review note for the manager."*
2. Analytical procedures commentary.
*"Below are current-year and prior-year financial figures for [account]. Compute: variance, variance %, trend over 3 years if data provided. Write analytical commentary (60–120 words) covering plausible explanations for the variance, follow-up procedures recommended, and any red flags. Tone: professional skepticism."*
3. Sample summary.
*"Below are results of sample testing for [account]. For each item: what was tested, result (pass/fail), exceptions noted, supporting documentation. Summarise: total sampled, total exceptions, exception rate, projected misstatement (if applicable), recommended sample expansion (if applicable). Output as a clean one-page summary."*
Three more for the back end of the engagement
4. Exception summary for partner review.
*"Below are all exceptions identified across the audit. Group by area, summarise the nature of each, indicate whether resolved or open, and rate severity (low / medium / high / clearly material). Output as a one-page summary for the engagement partner."*
5. Management letter draft.
*"Below are observations from this year's audit (control weaknesses, process improvement opportunities, regulatory observations). Draft a management letter to [client name]: cover note, observations grouped by theme, each with description / risk / recommendation / management response (placeholder). Tone: constructive, specific, no jargon."*
6. Audit-committee report draft.
*"Below are: audit scope, key risks identified, key procedures performed, key findings, areas of management judgement reviewed, audit conclusions, key matters for AC discussion. Draft a 2-page AC report. Singapore commercial English. Tone: factual, not promotional."*
Confidentiality and ACRA expectations
Three rules.
Rule 1: enterprise-tier only. Free / consumer Claude tiers may train on input. ACRA-regulated audit data must not. Switch your firm's Claude accounts to Pro / Team / Enterprise.
Rule 2: redact client identifiers in prompts where the analysis doesn't need them. Replace client name with [CLIENT_A], specific financial figures with [AMOUNT_X] for analytical procedures, etc. The audit logic is the same.
Rule 3: document AI use in the audit file. A note in the workpapers: *"[date]: Used Claude Sonnet for first-pass review of [workpaper]. Output verified by [associate name]."* This is the audit trail an ACRA inspector will look for.
Reference: ACRA Audit Regulatory Report 2023 and ACRA's later guidance on technology use in audit.
What Claude gets wrong (and how to catch it)
Quantitative errors. Claude can sometimes miscompute totals or percentages. Always recompute totals manually or via Excel.
Hallucinated standards references. Claude may cite a non-existent SFRS or ISA paragraph. Always verify against ACRA's published standards and the SAICA / ISCA published standards.
Plausible-but-wrong commentary. Claude's variance commentary can sound credible while missing the actual driver. Use it as a draft, not as a conclusion. Always cross-check against the year's actual events (acquisitions, restructurings, market conditions).
For broader prompting safety rules across professional services, see PDPA-Safe Claude Prompts: A Lawyer's Checklist — most patterns translate.
Is it ACRA-compliant for Singapore auditors to use Claude?
Use of generative AI as an audit tool is consistent with ACRA expectations provided the auditor maintains professional skepticism and remains responsible for conclusions. ACRA's [Audit Regulatory Report](https://www.acra.gov.sg/docs/default-source/default-document-library/training-and-resources/publications/reports/research-and-reports-on-audit-quality/practice-monitoring-programme-public-reports/audit-regulatory-report-2023.pdf) emphasises documentation and judgement — both still apply when AI assists.
What about audit-client confidentiality?
Use enterprise-tier Claude (no training on input). Redact client identifiers. Document AI use in the audit file. For highly sensitive audits, consider on-device deployment.
Can Claude replace senior associates?
It compresses first-pass extraction and analytical work — typically 40–60% of senior-associate time. Final professional judgement on materiality, sample selection rationale, and audit conclusions remains a human responsibility.
How does this affect audit fees?
Most SG audit firms using AI internally maintain fees and capture the time saving as margin or as capacity for additional work. Some boutique firms have started passing efficiency through to clients as a competitive lever.
Is there a workpaper-specific Claude tool?
Big 4 firms have proprietary tools. For mid-tier and boutique SG firms, OTG can build custom workpaper-review tools as [Functional App Sprints](/pricing) or [Process Automation engagements](/services). Claude is the underlying model, the workflow is bespoke to the firm.
