Walk through any South African accounting firm and you will find ChatGPT open in a browser tab. Staff use it to draft client emails, explain regulations, summarise documents, and untangle Excel formulas. Most of that is genuinely useful. Some of it is quietly dangerous — and almost none of it is building capability the firm owns.
This guide covers both halves: the guardrails your firm needs today, and the step beyond DIY AI when the volume justifies it.
Where ChatGPT genuinely helps accountants
Used well, general AI tools are strong at:
- Drafting client emails, engagement letters, and plain-English explanations
- Summarising long documents and correspondence threads
- Explaining unfamiliar concepts or software as a starting point for research
- Spreadsheet help — formulas, structure, cleanup approaches
The pattern: language work where a qualified human reviews the output before it matters.
Where it gets dangerous
Client data leaves your control. Pasting a client’s financials, ID numbers, or payroll details into a consumer AI tool sends personal information to systems outside your firm’s mandate. Under POPIA, your firm is responsible for where that data goes. This is the single biggest risk, and most staff have already done it without thinking.
Confident nonsense. General AI tools answer tax and regulation questions fluently and are sometimes wrong. Fluent-but-wrong is more dangerous than obviously wrong, especially under deadline pressure.
No memory, no ownership. Every good prompt, every correction, every lesson learned lives in one person’s chat history. When they leave, it leaves. The firm gets faster individuals and no smarter as a business.
The guardrails every firm should set this month
You do not need a 40-page policy. You need one page, enforced:
- Approved tools list. Which AI tools may be used for work, on which accounts.
- Forbidden data. No client identifiers, financials, or payroll data in consumer AI tools. Anonymise or don’t paste.
- Mandatory review. Nothing AI-drafted reaches a client or SARS without qualified human review.
- No advice delegation. AI never answers a tax or advisory question directly to a client.
- A named owner. One partner or manager owns the policy and the questions.
That covers the risk. It does not capture the opportunity.
Beyond DIY: when the firm should own the system
If your team uses ChatGPT daily, the demand signal is clear: there is repetitive language work everywhere in your practice. The question is whether it stays as scattered individual use or becomes a managed capability.
A managed AI employee is the grown-up version: one defined job (say, document collection or client reporting prep), approved knowledge sources, POPIA-conscious data boundaries agreed upfront, human sign-off on anything sensitive, and monthly review — so the system improves and the improvement belongs to the firm.
The difference in one line: staff using ChatGPT makes people faster; a managed AI employee makes the business smarter.
Why BizSage starts with an AI Opportunity Audit
Before building anything, it pays to know which workflow is actually worth automating and what the data boundaries need to be. The BizSage AI Opportunity Audit maps your firm’s repetitive work, quantifies the hours, designs the guardrails, and scopes the Company Brain and scopes the Company Brain and identifies the first AI employee worth piloting — so your firm moves from quiet DIY risk to owned, controlled capability.
FAQs
Is it safe for accountants to use ChatGPT with client data?
Not by default. Pasting client financials into a consumer AI tool can put personal information into systems your firm does not control, which raises POPIA obligations. Firms need clear rules: what may be shared, in what form, with which tools, and who reviews outputs.
Should firms ban ChatGPT?
Bans rarely work — staff use it quietly anyway. A better approach is a short AI usage policy: approved tools, forbidden data, mandatory review, and named ownership. Then move the high-volume workflows into a managed system with proper controls.
What is the difference between staff using ChatGPT and an AI employee?
Individual ChatGPT use makes one person faster for one task, and the learning evaporates. A managed AI employee is a controlled system: approved knowledge, defined jobs, POPIA-conscious data boundaries, human sign-off, and improvement that stays with the firm.