If you run a small accounting or bookkeeping office, you have probably tried ChatGPT or Claude. Maybe it rewrote an awkward email or explained a spreadsheet formula. Useful, but it didn't give you your Friday back.
You're in good company. In a Goldman Sachs survey of small business owners, only 14% said they had built AI into how the business actually runs. And in Thryv's 2026 survey, 70% of small businesses said they need more training to get more from it.
The gap is rarely the tool. It's that nobody has sat down and said: this job, every week, goes to the AI first, and this person checks it.
Here are five jobs worth handing over in a small practice, and three that should stay with a person. The examples are illustrations of the kind of work we mean, not client stories.
What makes a job a good one to hand over
Three things, and all three matter:
- It repeats. Every week, every month-end, every new client.
- It runs on information you already hold. Emails, your client list, your ledger, your own templates.
- A person stays responsible. The AI prepares; someone qualified approves.
If a job fails any of the three, leave it alone for now.
Five jobs to hand over
1. The "what's still missing" list
Before every month-end or filing deadline, someone works out which clients haven't sent which documents. It's slow, and it's the same every time.
Handed over, it looks like this: the AI compares what has arrived against each client's checklist and produces one list of who is missing what. A person reads the list and decides who to chase.
2. Payment reminders
Most firms chase their own unpaid invoices late, because nobody enjoys it. The AI can draft a polite first reminder, a firmer second one and a final one, each using the client's name, the invoice number and the amount from your records.
Keep a person on the send button, at least at first. Some clients need a phone call, not a third email, and only you know which.
3. First drafts of replies to routine questions
"When is my return due?" "Which receipts do you need?" "Can you resend last year's statement?" The same questions come in every week.
Give the AI your own standard answers and let it draft the reply for a person to approve. This is about wording you have already agreed, not advice. If a question needs judgement, it goes to a person untouched.
4. The first draft of the monthly summary
Many owners want a short note with their numbers: what changed, what looks unusual, what needs a decision. Writing it from scratch for every client is why it often doesn't get written.
The AI can prepare a first draft from figures you have already checked. The accountant corrects it, adds the judgement, and sends it. The figures themselves still come from your ledger, not from the AI's memory.
5. Meeting notes into a task list
After a client call, the notes sit in someone's notebook. Handed over, the notes become a short summary and a list of tasks with a name next to each, ready to paste into whatever you use to track work.
It's the smallest job on this list and often the first one a team keeps using.
Three jobs to keep with a person
Tax positions and advice. AI tools can sound confident and be wrong. Anything a client will rely on as professional advice needs a qualified person who has read the source documents.
Final figures and filings. Let the AI prepare, sort and draft. Don't let it be the last check on a number that goes to a tax authority or a bank.
Anything where you can't say where the data goes. Owners are right to worry about this. In Bluevine's survey, data safety was small business owners' top concern about AI, named by 33%, with accuracy close behind. If you can't answer "which account is this in, and who can see it?", don't put client data there.
The tools are closer than you think
You may not need new software. Anthropic launched Claude for Small Business in May 2026, with ready-made workflows and connections to tools such as QuickBooks, Google Workspace and Microsoft 365. OpenAI added workspace agents to ChatGPT's business plans in April 2026.
What these tools don't do is decide which of your jobs to hand over, connect them to the way your office really works, or teach your team to trust and check the result. That part is still yours.
How to start this month
- Pick one job from the list. The one that annoys you most is usually the right one.
- Write down how you do it today. Where the information comes from, what the result looks like, who checks it.
- Write one page of rules. What may go into the AI, what may not, and who approves anything that leaves the office. Your professional body's guidance and data protection rules come first; this is not legal advice.
- Run it for a month next to the old way. Compare the results before you rely on it.
- Only then add a second job.
If you'd rather not work it out alone
This is what our AI Check-up is for: a 60-minute call and a one-page plan naming the three jobs in your business worth handing over first. If you then book a build with us within 30 days, the fee comes off the price. Everything we set up stays in your own accounts.
Not sure where AI would save your business time?
A 60-minute call and a one-page plan naming the 3 jobs in your business worth automating first. Taken off the price of any build booked within 30 days.
Related Articles
Half of Greek Firms Say They Use AI. Few Use It Well.
47% of Greek firms say they use AI, but official figures are far lower and most users stay at the basics. What the numbers show, what the rules really require, and a free 10-question self-check.
Multi-Agent AI: The Enterprise Shift From Chatbots to Operations
Multi-agent AI systems deliver 3x faster task completion for enterprises. Learn how to move beyond single chatbots to coordinated AI operations.
Context Window Bloat: Why Feeding More Data to LLMs Is Ruining Your AI Agent's ROI
Stuffing huge payloads into LLM context windows makes agents slower, more expensive, and less accurate. Here is what context bloat costs and how to build a lean context architecture.