The Draft Email AI Should Write, Not the One You Retype
AI in your practice isn't about replacing judgement. It's about killing the repetitive drafting, summarising and chasing that eats your day. Here's where it actually helps.
Ask most practice owners where AI fits in their firm and you'll get one of two answers: a vague hope that it will "do the tax returns" one day, or a firm suspicion that it's a gimmick that can't be trusted near client data. Both miss the point.
The real opportunity isn't lodging returns automatically or letting a chatbot make judgement calls. It's the dozens of small, repetitive drafting and summarising tasks that sit between the work — the ones no one bills for and everyone does badly when they're tired on a Friday afternoon. That's where AI in a practice earns its keep today, and it's worth being specific about what that looks like.
The busywork that lives around the work
A compliance job is rarely held up by the accounting itself. It's held up by the messages, the summaries, the status updates and the chasing. Think about a single individual tax return moving through your firm:
- An email asking the client for their missing documents.
- A polite follow-up when they don't reply.
- A short note to the client explaining what changed from last year.
- A file note summarising a phone call.
- An internal handover comment when the job moves from preparer to reviewer.
None of that is technical work. All of it takes time, and all of it is broadly the same from client to client. Multiply it across a few hundred returns and it becomes a genuine drain — not because any one task is hard, but because staff are writing the same message from scratch, over and over, from memory.
This is exactly the kind of drafting AI is good at right now. Not deciding what to say, but doing the first 80% of writing it so a person can check it, adjust the tone and send.
Where AI genuinely helps in a practice today
Drafting client communications
The best use of AI in an accounting client management setting is the first draft. A request for outstanding items, a follow-up chase, an explanation of a variance, a reminder that a deadline is approaching — these all follow patterns. AI that has context about the client and the job can produce a solid draft in seconds. Your staff member reads it, tweaks it, sends it. The judgement stays human; the typing disappears.
Summarising and file noting
Long email threads, meeting notes, a messy back-and-forth about a client's structure — AI can distil these into a clean summary that actually gets saved to the client record instead of living in someone's inbox. The value here isn't cleverness, it's completeness: notes get made because making them stopped being a chore.
Answering "where is this up to?"
A lot of practice time goes into reconstructing status. What's outstanding on this job? What have we already asked for? When AI sits over your practice data, it can answer those questions from what's actually recorded, rather than someone opening five tabs to piece it together.
Turning notes into structure
A quick voice note or a scribbled list after a client call can become a set of tasks, a checklist, or a draft request. AI is good at converting loose input into structured output — which is precisely the gap between "I know what needs doing" and "it's actually in the system."
What AI should not be doing
It's worth being blunt about the limits, because overselling AI is how firms end up burnt and distrustful of the whole idea.
- It shouldn't make lodgment or advisory decisions. AI drafts and summarises. It doesn't decide what's deductible, what's compliant, or what to tell the ATO. Finye is a practice and client management system — it tracks obligations and runs the practice around the work; it is not a tax-lodgment tool or a ledger, and no AI feature changes that.
- It shouldn't send anything unreviewed. A draft the client never sees before a human checks it is a risk. The workflow has to keep a person in the loop.
- It shouldn't operate on data it can't be trusted with. AI is only as useful as the context it can safely see. This is exactly why it matters where your AI lives.
Why AI works better inside your practice management system
Here's the difference that decides whether AI is useful or just another novelty tab: context.
A general-purpose AI tool sitting in a browser knows nothing about your firm. To get a decent client email out of it, you have to paste in the client's name, the job, what's outstanding, last year's context, and the tone you want. By the time you've done that, you could have written the email yourself. Worse, you're copying client information into a tool with no relationship to your firm — a genuine confidentiality problem.
AI built into your account practice management software already has the context. It knows the client, the job, the outstanding items, the recurring pattern, the deadline. So when it drafts a chase email or summarises where a return is up to, it's working from your actual records — not from whatever you remembered to paste in. That's the whole game: AI is only as good as the data it can see, and the safest, most useful place for it to see that data is inside the system that already holds it.
In Finye, that means the AI drafting a client request is looking at the same client accounting record, work item and portal request your team is already working from. The output lands where the work already lives, instead of in a separate window you copy-paste out of.
Start with one repetitive job
You don't need an AI strategy. You need to pick the single most repetitive piece of writing your firm does and let AI take the first pass at it. For most practices that's the outstanding-items request or the status follow-up — high volume, low variation, and currently written from scratch every time.
Let AI draft it. Have a person check and send. Measure how it feels after a week. Then add the next one — file notes, handover comments, variance explanations.
The firms getting value from AI aren't the ones waiting for it to do the returns. They're the ones quietly removing the retyping, the re-summarising and the from-memory chasing that surrounds the returns — and getting those minutes back on every single client.