What AI actually does for an accounting practice — and what it doesn’t

AI in an accounting practice does the repetitive reading, sorting and drafting — extracting data from documents, triaging an inbox, suggesting categorisations and surfacing what needs attention. It does not exercise professional judgement, sign off work, or file with SARS for you. The useful framing is simple: AI handles the volume so the practitioner handles the decisions.

There is a lot of noise around AI for accountants right now. Some of it is real and already working; some of it is a demo that falls over on the second client. This is an honest line between the two, and a note on the dignity-first way we think a practice should adopt it.

What AI genuinely does today

• Reads documents into data. AI extracts transactions from bank statements and details from SARS letters, so you review structured data instead of typing it.
• Triages the inbox. AI sorts incoming email and SARS correspondence, flags what matters, and turns a message into a tracked task.
• Suggests categorisations. It proposes how to code transactions; you confirm or correct.
• Surfaces what needs attention. It points partners at overdue work, ageing WIP and compliance risk across the book.
• Drafts. It produces a first draft of routine correspondence for you to edit, not send blind.

What AI does not do

The honest list matters as much as the capability list — because over-promising is how firms get burned.

• It does not exercise judgement. The categorisation, the tax position, the advice — those stay with the qualified human.
• It does not file for you. A tool can track every SARS deadline and surface SARS correspondence, but you still submit through SARS eFiling.
• It does not run payroll. It can track payroll-tax obligations and deadlines; it does not produce payslips or calculate PAYE.
• It is not infallible. Extraction and suggestions need review. The review step is a feature, not a failure.

The dignity-first way to adopt it

How a firm adopts AI matters as much as which AI it adopts. We build to a frame we call Emergent Intelligence (EI) — the dignity-first way we think about what is more commonly called AI: agency over automation, transparency over black boxes, and a human in the loop on every decision that carries professional weight.

In practice that means three commitments. AI proposes; the practitioner decides. Every automated action is auditable, so you can answer for it. And the AI removes the drudgery rather than the judgement — it gives you back the hours, and leaves you the work that actually needs a person.