Hong Kong Accounting Firm AI: What to Automate

Which work can be automated, and which cannot

Hong Kong accounting and finance practitioners are among the fastest adopters of generative AI across the professions. 2025 survey data shows that 88% of Hong Kong accounting and finance respondents use generative AI tools at work, a figure that reflects broad adoption in the industry, and not a wait-and-see stance. But there is a gap between broad adoption and a clear compliance framework: the Hong Kong Institute of Certified Public Accountants (HKICPA) has so far not issued formal AI guidance equivalent to the Law Society’s position paper, and the industry is working out its boundaries by itself in a relatively loose guidance environment.

HKICPA’s Current Position

In November 2024 the HKICPA held an IT Conference themed on transforming accounting practice with AI, emphasising that technology adoption and risk mitigation go together. The Institute’s overall orientation favours promoting adoption over issuing restrictive guidance. In the absence of a formal AI policy document, the industry relies on the general principles of the HKICPA Code of Ethics: the requirement for professional scepticism is not reduced by the use of AI tools, and the certified public accountant (practising) who signs the audit report bears full legal responsibility for the audit opinion.

For the corresponding situation in the legal profession, see Hong Kong Law Firm AI Review: Limits and Liability.

Workflows That Can Be Automated

The following workflows can technically be automated with AI, and the industry already has practical applications:

Data entry and reconciliation: Financial data extracted from source documents is entered into the system automatically, and cross-account, cross-document reconciliation checks can greatly compress manual time.

Transaction classification: Large numbers of transactions are given a first-pass classification against preset account codes, with accountants then reviewing the exception items.

Compliance checks: Client financial documents are given a first-pass check against a fixed rule list, identifying missing or non-conforming items.

Expense and payroll management: Periodic payroll calculation and expense reimbursement in a uniform format, where the rules are clear and the output can be verified programmatically.

For methods and costs of extracting data from source documents, see AI Document Processing Automation: Turning Unstructured Documents into Structured Data.

Judgement Areas That Cannot Be Automated

Forming the audit opinion: Under the Hong Kong Companies Ordinance (Cap. 622), the auditor’s responsibility is not transferred by the use of AI tools. The final interpretation of account balances, the evaluation of client representations and the judgement on the overall fairness of the financial statements must all be borne by the licensed person and cannot be delegated to AI tools.

Applying professional scepticism: AI can identify statistical anomalies in data, but judging whether an anomaly represents fraud, error or a legitimate business matter is a professional judgement. Assessing edge cases, unusual transactions and management representations is the hardest part of audit work to standardise.

Client communication: The interpretation of Hong Kong tax ordinances (profits tax, salaries tax) in edge cases, and tax advice to clients on complex structures, both need a licensed person’s judgement, and AI tool output on such tasks is for reference only.

Differentiated Data Security Risk

The large firms (the Big Four) are usually equipped with enterprise-grade AI usage agreements, handling client financial data in a secure environment. Smaller local firms that use public AI services directly (such as the free tier of ChatGPT or Claude.ai) to process client data face a significantly higher PDPO compliance risk: client financial data is highly sensitive personal and commercial information, and the data handling terms of public AI platforms must be assessed with care.

For the privacy questions to answer before handing documents to an AI system, see Enterprise Data and AI: Privacy Questions to Ask Before Handing Documents to a System.

Summary

The current state of AI adoption in Hong Kong accounting is high usage with little formal guidance. Until a formal framework becomes clear, the practical principle is clear: AI is used for data processing and initial screening, while the licensed person handles judgemental work and final responsibility. On data security, the difference between enterprise deployment and public platforms must be clearly distinguished, and convenience of use is not a reason to overlook it.