The Review Step AI Can't Replace (And Shouldn't)
AI can draft, extract and summarise — but someone still signs off. Here's where automation belongs in your practice, and where human judgement stays.
There's a quiet anxiety running through a lot of accounting firms right now. If AI can read a document, draft an email, summarise a file and pre-fill a workpaper, what's left for the people who used to do that work? And for practice owners weighing up new tools, the opposite question nags just as hard: if I let software touch client data and client communication, where does my professional judgement actually sit?
Both questions share an answer. The work AI does well and the work that must stay human aren't in competition — they're two halves of the same job. The firms getting this right aren't the ones that automate the most. They're the ones that know exactly where the review step belongs and never skip it.
What AI is genuinely good at in a practice
Strip away the hype and the useful applications are narrow, repetitive and surprisingly dull — which is precisely why they're valuable. In a typical accounting or bookkeeping practice, AI earns its keep on tasks like:
- Pulling data from documents you already have. Bank statements, invoices, trust deeds, ID documents — reading structured information out of a PDF and dropping it into the right field, rather than someone retyping it.
- Drafting the first version of client emails. The chase for an outstanding signature, the request for missing PBC items, the status update — a sensible draft produced in seconds that a person then edits and sends.
- Summarising a file before a call. What's outstanding, what changed, what the last few interactions were, so you walk in informed instead of scrolling through six screens.
- Answering factual questions from a client's own records. "When did we lodge last year's return?" or "What's our current engagement scope?" — questions where the answer already exists in your system.
Notice what these have in common. They're all about retrieval and first drafts, not decisions. AI is excellent at producing something close to right, fast. It is not accountable for whether that something is actually right.
The review step is the whole job
Here's the part that gets lost in product demos: the value a firm provides was never the typing. It was the judgement. A client doesn't pay you to retype their bank statement — they pay you because you know when something on it looks wrong.
So when AI handles the typing, the drafting and the extracting, it doesn't remove your value. It relocates it. The work shifts from producing to reviewing, and review is where experience, context and professional scepticism live. The AI can tell you the invoice says $4,200. Only a person knows that this client never has a supplier invoice that size, and it's worth a second look.
This is also where professional obligations don't move. Your name goes on the lodgment. Your AML conclusion stands up to scrutiny or it doesn't. An AI draft of a client's position is a starting point, never a sign-off. The tool can shorten the path to a decision; it can't make the decision.
A practical rule for every automated task
Before you let software do something unattended, ask one question: if this goes wrong, who notices, and when?
- If an AI-drafted email sends the wrong tone, the client notices immediately, and the cost is minor. That can run with light review.
- If AI-extracted figures feed a workpaper that nobody checks, no one notices until it's wrong in something you've lodged. That needs a hard review gate, every time.
The lower the visibility of an error and the higher its cost, the more human sign-off you build in. The more obvious and cheap the error, the more you can let automation run. Sort every automated step by that test and you'll know where to put your attention.
Where this fits in client accounting software
The trap with AI is treating it as a bolt-on — a separate chatbot in a separate tab, disconnected from your client records, your jobs and your deadlines. When the AI can't see the file, it can only give generic answers, and a generic answer in a professional context is worse than no answer.
The point of building AI into your account practice management software rather than alongside it is that the tool works from your actual data. In Finye, the AI drafts emails from the real client record, summarises the real job history, and answers questions from the information already in the system — then puts the output in front of a person to approve, edit or send. The draft arrives; the human decides. That's the shape every AI feature in the practice should take.
This matters more for a growing firm than a small one. When you've got a handful of clients, you carry the context in your head and the review happens naturally. At two hundred clients across recurring jobs, compliance deadlines and a client portal full of requests, the volume of first drafts and data entry is exactly what buries a team — and exactly what AI should absorb, so your experienced people spend their hours reviewing instead of typing.
What staff actually end up doing
A fair worry is what happens to junior staff who used to learn the trade by doing the repetitive work. There's a real tension here, and pretending otherwise helps no one. But the shift isn't from "skilled work" to "no work" — it's from "do the data entry" to "check the data entry and understand why it matters."
Reviewing an AI draft is a faster way to learn judgement than producing a hundred of them yourself, provided someone teaches the review. The firms that handle this well are explicit about it: here's what the AI gave us, here's what we changed and why, here's the thing it missed. That's training. It's just training at the level that actually earns fees.
The honest summary
AI in your practice isn't a replacement for client accounting work and it isn't a magic lodgment button — no tool is a tax return software that thinks for you. It's a way to get from a blank page to a reviewable draft in seconds, across the boring, high-volume tasks that never deserved an expert's full hour.
Keep the review step. Build it in deliberately, size it to the cost of getting it wrong, and make sure the AI works from your real client data rather than a separate silo. Do that, and automation stops being a threat to your professional value and becomes the thing that frees you up to apply it.