AI for auditors: a practical playbook for Singapore audit teams

By Daryl Aw, CA (Singapore), 3x UiPath MVP · Published 6 July 2026 · Last updated 6 July 2026

The short answer: AI helps auditors most in five places: journal entry testing across entire populations rather than samples, smarter sample selection, cleaning messy client data, first-draft documentation, and analytical procedures. It does not replace professional judgement or scepticism, and it must be used inside confidentiality guardrails. Here is the playbook we teach and deploy at audit firms.

1. Journal entry testing: from sampling to full population

The classic JE test scans for the usual red flags: round amounts, weekend and odd-hour postings, unusual user and account combinations, entries just below approval thresholds. Done by hand, teams test slivers of the population. Automated with AI-assisted tools, the full ledger is screened in minutes and the team's time shifts to investigating the exceptions that matter. This is one of the first robots we ever deployed in practice, and it remains the fastest win for most firms.

2. Sampling and selections

Risk-weighted and stratified selections, generated with documented parameters and reproducible logic, remove both the tedium and the "why these items?" review question. The methodology stays the firm's; the mechanics stop consuming senior time.

3. Client data wrangling

Every auditor knows the ritual: the client sends a GL export with merged cells, subtotals mid-table and three date formats. AI-assisted cleaning turns an afternoon of Excel surgery into minutes, and the audit actually starts on day one. In our classes this single exercise gets more spontaneous applause than anything else we teach.

4. Documentation drafting, with the auditor as author

AI drafts; auditors conclude. Used properly, an AI assistant produces first drafts of testing narratives, walkthrough writeups and fluctuation commentary from your structured inputs, which the auditor then verifies, edits and owns. The standard of documentation, that an experienced auditor with no connection to the engagement can understand what was done and why, does not change. What changes is how long the first draft takes.

5. Analytics and fluctuation analysis

Month-on-month and year-on-year movements, ratio anomalies, and "explain this variance" first passes are natural AI territory, provided every figure is traced back to source before it enters the file.

The guardrails

• Confidentiality first. Client data goes only into enterprise AI services under appropriate terms, or is anonymised, or is processed locally. Consumer chatbots and client ledgers do not mix.
• Verify everything. AI output is an unaudited draft. Treat it with the same scepticism as a client-prepared schedule.
• Document AI use in line with firm methodology and applicable standards, including quality management expectations.
• Train reviewers, not just preparers. Managers must know what AI-assisted work looks like to review it effectively.

Where to start

Pick one engagement, automate the JE screen and the data cleaning, and measure hours saved. Then scale. Our training programmes have audit teams build exactly these tools in class, and our project team deploys them at firms. For funding firm-wide job redesign, see the WDG(JR+) grant.

Frequently asked questions

How are auditors using AI in 2026?

The highest-value uses are journal entry testing and exception identification across full populations, risk-weighted sample selection, cleaning and reconciling client data exports, drafting documentation and testing narratives for reviewer refinement, and analytical procedures such as fluctuation analysis. Judgement, scepticism and conclusions remain the auditor's.

Can auditors put client data into ChatGPT?

Not into consumer tools without safeguards. Audit teams should use enterprise-grade AI services with contractual data protections, follow firm policy and professional confidentiality obligations, and prefer approaches where sensitive data is processed locally or anonymised.