The Draft Email AI Wrote While You Read the File
AI in your practice isn't about replacing judgement — it's about removing the typing between you and the answer. Here's where it actually earns its place.
There's a particular kind of task that fills an accounting practice day and never shows up on a timesheet. Reading a client's file to remember where things stand. Drafting the email that summarises it. Summarising a long back-and-forth into a status note. Pulling the three numbers you need from a document someone sent last week. None of it is hard. All of it takes time. And across a practice of even five people, it adds up to hours a day that no one chose to spend.
This is the gap AI is genuinely good at filling. Not the dramatic version — AI doing your tax planning or signing off on advice — but the ordinary version: writing the draft while you read the file, so the work between the thinking and the doing disappears.
What AI should actually do in a practice
It helps to be specific, because "AI" has become a word that means everything and nothing. In a well-run practice, the useful applications are narrow and concrete:
- Drafting the routine email. The follow-up on outstanding documents, the "your return is lodged" note, the reply to a client's quick question — these follow patterns. AI that can read the relevant client record and draft the message in your firm's tone saves the staring-at-a-blank-reply time.
- Summarising the file. When a job moves between staff, or a client rings and you haven't touched their work in three weeks, you need the state of play fast. AI can read the history — notes, emails, work items — and give you the gist in a paragraph instead of making you scroll through six screens.
- Pulling data from documents. The figures sitting in a bank statement, a trust distribution minute, a prior-year return — AI can read the document you already have and lift the data out, rather than you keying it in field by field.
- Answering questions from the file. "What did we charge this client last year?" "When was the engagement letter signed?" "Has the BAS been lodged?" These answers already exist in your system. AI that can read it answers in seconds.
Notice what these have in common. The information already exists. The judgement has already been made, or doesn't need to be. The only thing standing between the question and the answer is retrieval and typing. That's exactly the work to hand off.
Why the AI has to live where your data lives
Here's where a lot of firms go wrong. They adopt a general-purpose AI tool, open it in a separate tab, and paste in context by hand. It drafts a nice email — but you had to tell it who the client is, what the job is, and what's outstanding, because the tool can't see any of that. You've moved the work, not removed it.
AI only saves real time when it can read the records you already keep. A draft reply is useful because it knows this client has two overdue documents and a BAS due Friday. A file summary is useful because it's built from the actual notes and work items, not a prompt you had to write. A data-entry suggestion is useful because it's reading the document attached to the right client.
This is the practical case for keeping client records, work items, deadlines and correspondence in one system rather than scattered across a tax return software login, a spreadsheet, an inbox and a shared drive. It's not just tidier. It's the difference between AI that genuinely works for you and AI that's a slightly fancier blank page. In Finye, the built-in AI reads the same client records, boards and compliance deadlines your team already uses — so the draft it writes or the summary it gives you is grounded in your actual practice, not in whatever you remembered to paste in.
The line AI shouldn't cross
Being honest about what AI is good for also means being honest about what it isn't. AI drafts; you decide. It can write the client email, but you read it before it sends — because tone, nuance and the occasional sensitive situation are yours to judge. It can pull figures from a document, but you confirm them before they matter. It can summarise a thread, but the summary is a starting point, not a substitute for knowing your client.
The firms that get value from AI treat it like a capable junior who's fast and tireless but needs a review. The ones who get burned treat it like an oracle. In a compliance environment, where a wrong number or a misjudged message has real consequences, the review step is not optional — but it's also fast, because checking a good draft takes far less time than writing from scratch.
Where to start
You don't need to reinvent your practice to get value here. Pick one repetitive, low-judgement task and let AI take the first pass:
- Start with the email you write most. Document-chasing, status updates, lodgment confirmations. Let AI draft from the client record; you edit and send. Measure how it feels over a week.
- Use it for handovers. Before a job moves between staff, generate a file summary. It forces the context into words and saves the incoming person a morning of reading.
- Lean on it for document data. Rather than keying figures from a statement or a prior return, let AI read the file and propose the entries. Confirm, don't retype.
None of this replaces the skill that makes your firm worth paying for. The advice, the judgement, the relationship — those stay human. What changes is how much of your day you spend on the typing and retrieving that sits around that skill. That's the work AI should be doing, from the documents and records you already have.
The draft email you were about to write? The AI can have it ready by the time you've finished reading the file. You still decide whether it's right. You just don't have to start from nothing.