The Draft Email AI Writes: Where Judgement Still Belongs
AI can draft the chase, summarise the file and sort the inbox. Here's how to let it do the work without letting it make the call.
There's a version of AI in the practice that gets sold as magic: point it at your client base and it runs the firm. There's another version that's real, useful, and already sitting inside good accounting client management software today. It doesn't replace your judgement. It does the typing, the sorting and the first draft — and then it hands the decision back to you.
Getting value out of AI in an accounting or bookkeeping practice is mostly about drawing that line in the right place. Some tasks are pure mechanics with a clear right answer. Some tasks need a human who knows the client, the risk and the relationship. The trick is to hand the first kind over completely and keep the second kind firmly in your hands — while still using AI to make the second kind faster.
The tasks that are safe to automate outright
Start with the work where there is a correct answer and the cost of a small error is low. These are the jobs no one wants to bill for and no one enjoys doing.
- Inbox triage. Every practice inbox is a mix of "my BAS query," "can you resend the invoice," "here are my documents" and cold pitches from software vendors. Sorting that by client, urgency and topic is exactly the kind of pattern-matching AI is good at. It reads the message, matches it to the right client record and routes it to the right person or work item — before anyone has spent ten minutes deciding where it goes.
- Extracting data from documents. When a client sends a trust deed, an ASIC form or an ID document, the ABN, ACN, entity name and key dates all sit inside it. Re-keying that by hand is slow and it's where duplicates and typos are born. AI can read the document and populate the fields, leaving your team to check rather than transcribe.
- Summarising a long thread. A client file that's forty emails deep is a barrier to picking up the work. A short, accurate summary of "where are we up to" turns a fifteen-minute read into a fifteen-second one.
None of this is the AI making a decision about the client. It's clearing the runway so your people spend their time on work that actually needs a brain.
The tasks where AI drafts and you decide
This is the sweet spot, and it's where most of the real time savings live. AI produces a first draft; a human reviews, edits and sends. You get the speed of automation with the safety of a person in the loop.
Chase messages and reminders
Chasing documents, signatures and payments is the tax on every practice — the work that costs you the most and shows up nowhere on an invoice. AI can draft the chase for the outstanding PBC list, the unsigned engagement letter or the overdue invoice, in the right tone, referencing the right client. You read it, adjust anything that needs a human touch, and send. Better still, much of this can run as a rule you set once inside your practice management software, so the routine follow-ups go out on schedule and only the sensitive ones land on your desk.
Client-facing explanations
When a client asks why their BAS is higher this quarter, the numbers come from your ledger — but the plain-English explanation is something you write from scratch every time. AI can draft that explanation from the figures in front of it. What it cannot do is know that this particular client is anxious about cash flow, or that you promised them a call last month. You add that. The draft saves you the blank page; you keep the relationship.
Scoping and proposal wording
Writing up the scope for a new engagement, or wording a service package, is repetitive but not mechanical. AI can turn your bullet points into clean, consistent prose that matches your firm's style. You still decide what's in scope, what it costs and where the boundaries sit.
The tasks that stay human, full stop
Some decisions should never be handed to a model, no matter how confident it sounds.
- Judgement calls on tax positions. AI is not a tax-lodgment tool and shouldn't be treated as one. Whether a deduction stands, how a structure is treated, what's defensible — that's professional judgement, and it's your name on the return.
- Risk and KYC conclusions. AI can gather the identity documents and flag what's missing. It should not be the one that decides a client passes your AML checks. The verification is yours to sign off.
- Anything the client will read as a promise. A commitment on price, timing or outcome carries weight because a person stands behind it. Let AI draft the words; make sure a human owns the promise.
Why the line matters more than the tool
The firms that get burned by AI aren't the ones using it — they're the ones who let it make calls it shouldn't, or who refuse to use it at all and drown in manual work. The winning position is boring and practical: automate the mechanics, draft the judgement calls, and keep the actual decisions with people who know the client.
This is easier when your AI lives inside the system that already holds the work, rather than in a separate chat window. In Finye, the AI sits alongside your client records, boards and portal — so a drafted reply already knows which client it's about, a summary is built from the real file, and a triaged email lands on the right work item. The context is there because the client accounting and practice data are there. You're not copying and pasting between tools and hoping nothing gets lost.
Where to start this week
Pick one repetitive, low-risk task and hand it over completely — inbox sorting or document data entry are good first candidates. Then pick one task where you write the same kind of message over and over — a chase, a status update, a routine explanation — and let AI draft it while you keep the send button. Notice how much of your week those two changes give back.
Do that, and AI stops being a threat to your professional judgement and becomes what it should be: the assistant that does the typing so you can do the thinking.