It is Tuesday night and a new enquiry has come through while you were still finishing last month’s client work. A new client’s documents have not arrived. An invoice is overdue, and you have been meaning to book that annual review since March.
None of that is fixed by working more hours. It is admin, and admin is the first thing to fall over when the real work gets busy.
This is a plain-English guide to AI for accountants and bookkeepers: five workflows you can hand to an AI agent, what each one actually does, what stays with a human, and the realistic first step. No software overhaul, no rebuild of the practice. One job at a time.
What an AI agent actually does (and what it doesn’t)
Most people have met AI as a chat window: you ask, it answers. An agent is the next step along. It still reads and writes like a chat tool, but it is connected to the things you already use — your inbox, your calendar, your practice software, your invoicing — and it is allowed to take an action. It sends the follow-up. It books the meeting. It attaches the checklist and sends the reminder.
The honest version of the promise: an agent does the legwork and the drafting. You keep the decisions — what to advise, what to charge, who gets an extension, when to say no. If you want the longer explanation, we have written up what AI agents are separately.
Automation vs an AI agent
Ordinary automation follows a script you wrote in advance — “when an invoice is fourteen days overdue, send this exact email”. It falls over the moment reality gets messy: a client replies “can we deal with this after I’m back from leave?” and the script has no idea what to do with that.
An agent handles the messy middle. It can read an enquiry that asks whether a new business should be a company or a sole trader, capture the details and route it to the right person — without attempting to answer the advice question itself. That flexibility is what people mean by agentic AI, and it is why you want a human checking the output while you learn how it handles the odd ones.
AI for accountants starts with the one job nobody wants to do
The practices that get something out of this do not start with a plan to “implement AI”. They start with one task: repetitive, mostly text, and low-risk if it needs a second look before it goes out. One workflow quietly working beats five half-built ones you have stopped trusting.
Three worries come up in nearly every conversation we have about this, and all three are fair.
- “It will sound like a robot to my clients.” It will, if you let it write like one. Feed it your actual past replies so it picks up your wording, keep messages short, and read the first few weeks of output before anything sends by itself.
- “This will turn into a big expensive project.” Only if you scope it like one. A single workflow, plugged into tools you already pay for, earns its keep in the first fortnight or you switch it off.
- “I will lose the relationship with my clients.” The agent takes the parts of the conversation you were not getting to anyway — the chase, the reminder, the confirmation. The moment a conversation needs judgement, or anything that looks like advice, it hands over to you.
If you want the general version of that method, we have covered where to start with AI automation in more detail.
1. Enquiries: answer every lead and flag the sensitive ones
Most enquiries arrive when you cannot answer — evenings, weekends, the last week of tax season. They land on several channels too: the website form, email, Facebook, sometimes the phone.
An enquiry workflow pulls all of that into one queue and answers straight away, capturing the essentials — name, business type, what they need, how urgent. The useful output is not the reply. It is the tidy summary waiting for you in the morning: who got in touch and which ones actually need you.
What to keep human
Any question that asks for advice, tax or otherwise. A client asking whether to set up as a company or a trust needs a conversation with you, not a confident guess from software. Those get flagged and handed over, never answered — same for complaints and anything with a compliance angle you have not signed off on.
First step: turn on automated replies to website enquiries on one channel, and nothing else. You will know inside a week whether it changes anything.
2. Client onboarding: collect the documents without the back-and-forth
Onboarding rarely stalls because the client is difficult. It stalls because nobody chased the missing documents and the client got busy as well.
The workflow: a new client signs on, and the agent sends the checklist — identification, prior returns, access to their accounting software. When something is missing, it sends a polite reminder a few days later, and another a week after that — a short note that names the exact document still outstanding. Each message reads like it was written about that client, because it was.
Two rules keep it safe. Drafts for approval before anything sends on its own — you will edit the first ten, then you will stop wanting to. And it stops the instant the client replies or the file arrives, and respects a “do not chase” list.
What to keep human
Reading the documents, acting on them, and lodging anything. The agent collects and checks for completeness; it never interprets a tax position and never lodges a return, a BAS or an ASIC filing. Client data — tax file numbers, bank details, identification — stays in the systems you already trust, and only the people you choose can see it.
First step: one checklist, one reminder, drafts only. Add the second nudge once you have seen how the first one lands.
3. Scheduling: end the calendar ping-pong
Three emails to agree on a Thursday morning is the most common hidden cost in a small practice, and it doubles in tax season.
A scheduling workflow ends that. The agent offers real windows from your actual calendar, books the one the client picks, confirms, reminds them the day before, and handles the reschedule. The useful bit is the same whatever you call it: the client picks a time without you touching the phone.
Set two guardrails up front. Priority: a late lodgement cannot sit behind a routine catch-up. And a buffer, capping how much of the day is bookable, so the diary cannot fill solid in the last week before a deadline.
What to keep human
Deciding what is urgent. A client who needs you before a deadline should reach you, not the first free slot in a fortnight. You set the rules for what gets fast-tracked; the agent books the routine ones.
First step: make one meeting type bookable — the standard new-client consult or the annual review — and keep everything else on the phone.
4. Invoice chasing: get paid without the awkward call
Chasing money is the job most accountants and bookkeepers put off, and it lands most directly on your cash flow.
The workflow is a reminder ladder with the tone escalating slowly. A friendly note a few days after the due date, with the invoice attached again, because it is often genuinely lost. A firmer one a week or two later. Then it stops and hands you a summary: who still owes, how long it has been, how much. What the agent adds is writing each note so it sounds like a person, and reading the replies — so “paid it yesterday” gets routed to you instead of triggering another chase.
What to keep human
Everything about the money itself. The agent never changes an amount, never offers a discount, never agrees to a payment plan and never comments on a client’s tax position. It chases politely; it never negotiates.
First step: the first friendly reminder only, sent to yourself for a fortnight, so you can read exactly what your clients would have received.
5. Client follow-ups: reviews, reminders and the touch points you forget
Reviews and referrals do a lot of the work in whether a new client picks you or the practice down the road, and most of the reviews you never got were simply never asked for. So are the annual reviews you meant to book and the BAS deadlines you meant to flag.
The workflow: an engagement closes, and a short review request goes out the same day with a direct link to your Google profile, while the work is still fresh. The agent can also draft replies to reviews as they come in for you to approve and post. Separately, it keeps the recurring reminders on schedule — the annual review, the next lodgement, the check-in you promised.
What to keep human
The actual review of a client’s financial position, and any advice that comes out of it. An agent can remind a client their review is due and book it; it cannot tell them what to do with their money. One compliance point: ask every client, not just the happy ones, and never offer a discount or prize for a review — Google’s policies prohibit incentivised reviews.
First step: automate the review request for one engagement type — the completed return or the signed-off BAS — and leave the recurring reminders for the next round.
How the five fit together (a Melbourne example)
Picture a two-partner accounting practice in Hawthorn with a part-time bookkeeper.
Tuesday, 7.40pm: a new enquiry arrives through the website. “I’ve just started a business and I’m not sure if I should be a company or a sole trader, and my BAS is late.” The agent replies within a minute, captures the name and the business — and, because the message asks for advice, flags it for a partner to call back. It does not try to answer the structure question.
Wednesday morning the summary is waiting. You call the client, answer the question yourself, and book a consult. The agent sends the onboarding checklist and nudges the client when the ID and prior returns are slow to arrive. The consult happens, the engagement letter is signed, the invoice goes out. Day five, a friendly reminder lands and the payment follows. The same day, a review request goes out with a direct link, and a “next BAS due” reminder is scheduled for the following quarter.
That is one thread, not five products. Five tools that do not talk to each other is the situation most practices are already in — which is exactly why the admin still lands back on you.
What to watch out for before you switch anything on
- Keep a human in the loop for the first fortnight. Everything drafts-for-approval before anything sends by itself.
- Write down what the agent must never do. Give advice, quote a fee you have not priced, change a figure, or lodge a return or filing. Advice and lodgement stay with a human, always.
- Know where your client data goes. Which tool holds it, which country it sits in, who at your end can see it. Ask before you connect anything to your inbox or practice software.
- Leave an obvious way to reach a person. A client who wants a human should get one within a message or two.
- Follow the Australian messaging rules. The Spam Act 2003 applies to commercial electronic messages regardless of who or what pressed send. You need consent, a working unsubscribe, and unsubscribe requests actioned within five business days. A booking confirmation is different from marketing, so keep the two separate.
- Make it sound like you. Read the drafts out loud. If a sentence is not one you would say to a client, change the instructions until it is.
Where to start this week
- Keep a list for one week. Every time you do a repetitive admin task, write it down. The list is the audit.
- Pick the single most annoying one. Not the biggest — the one that makes you sigh. Momentum matters more than scope.
- Run it as drafts for a fortnight. Let it write, you press send. When you stop editing, let it send on its own.
That is genuinely enough for a first project. Most practices do not need a platform; they need one annoying job handled properly.
If you would rather not work out which task to start with on your own, that is the sort of thing a short call sorts out in half an hour. Web Intelligenz builds AI agents and workflow automation for Melbourne accountants and bookkeepers, and you are welcome to book a free strategy call to talk through which single workflow is worth doing first.
Frequently asked questions
What is AI automation?
AI automation is handing a task that used to need a person reading, writing or deciding to an AI model instead — answering an enquiry, drafting a follow-up, sorting a message to the right place. The difference from older automation is that it copes with input it has not seen in exactly that form before.
What is AI workflow automation?
A workflow is a sequence: the enquiry arrives, gets answered, gets captured, gets followed up. AI workflow automation means an agent carries that sequence from step to step instead of you shepherding it along. The value is in the handovers, not in any single message.
Can an AI agent give tax or financial advice?
It should not, and a well-designed one will not. Advice is a human’s job — the agent can collect the documents, book the meeting and draft the notes, but the recommendation, the lodgement and the sign-off stay with you. Draw that boundary before you build anything.
Is my clients’ data safe with AI automation?
That depends on the tool and how it is set up, not on the word “AI”. Before you connect anything, check which provider holds the data, which country it sits in, who at your end can see it, and what happens when you stop using it. Client financial data — tax file numbers, bank details, identification — deserves a proper answer to each of those questions first.
Do I need to be technical to set this up?
No, but somebody does need to think clearly about the process. Most of these tools are configuration rather than code now. The hard part is not the software; it is deciding what the agent is allowed to do on its own and what it must always pass to you.
Sources: Google Business Profile Help, “Prohibited & restricted content” — fake engagement and incentivised reviews (accessed August 2026). Australian Communications and Media Authority, “Avoid sending spam” and the Spam Act 2003 consent and unsubscribe rules (accessed August 2026). No statistics or survey figures are quoted in this article.
