How Small Accounting Practices Can Use AI for Client Email Drafts
By Stacey Tallitsch | May 11, 2026
If you run a small accounting practice, your inbox is a second business. Every week brings client questions about deductions, status updates on quarterly filings, and follow-ups on documents you've been waiting two weeks to receive. You answer the same kinds of questions over and over. The work is real, but it's not the work you went into accounting to do.
AI tools can take a meaningful chunk of this off your plate. Not all of it. Not the parts that need your judgment. But the routine drafting work that fills your inbox by Wednesday morning? AI can do most of that, and do it in your voice, in a way that still reads like a human wrote it.
Here is how to set this up without spending a weekend on it, and without putting client information somewhere it shouldn't be.
Why this fits a small practice
A solo CPA or a three-person firm has a specific problem. You have more email volume than you should be handling personally, and you don't have an office manager to triage it. Hiring one costs $50,000 a year minimum. AI doesn't replace the office manager you'd hire eventually. It replaces the office manager you can't afford right now.
The use case is narrow on purpose: drafting the first version of an email. You still read it. You still send it. You still own what goes out. The AI handles the work of pulling the right tone, structure, and content together from a few sentences of context.
A solo bookkeeper on Reddit recently described this as the most boring use of AI that has actually changed her week. That tracks. The boring uses are the ones that compound.
The tools that work for this
You have three reasonable options, all of which work for a small accounting practice.
Claude from Anthropic is what I would start with for professional services work. Claude tends to write in a calmer, more measured tone out of the box, which suits the register an accountant wants to send to clients. The paid plan runs about $20 per month. That gets you enough capacity for a solo practice.
ChatGPT (the paid plan, also around $20 per month) is the most widely used option. The free version will not be enough for this. The paid version gives you the better model and the projects feature you'll want to use.
Microsoft Copilot is the option to consider if your practice already runs on Microsoft 365 (Outlook, Word, Excel). Copilot drafts emails directly inside Outlook, which removes a step. It costs about $30 per user per month on top of your existing Microsoft license.
There is no single right answer. Pick one. The setup is similar across all three.
The setup, in five steps
Step 1: Decide what categories of email you are drafting. Don't try to automate everything at once. Pick three to five common email types. For most small accounting practices, the list looks something like this: requests for missing documents from clients, quarterly filing status updates, responses to common deduction questions, payment reminder follow-ups, and meeting confirmations. Write the list down.
Step 2: Set up a project for each category. In Claude or ChatGPT, both tools let you create a "project" or "custom GPT" that holds standing instructions. This means you don't have to retype the same context every time. For each email category, create a project. The standing instructions should include your practice's tone (formal but warm; clear; never use jargon the client wouldn't know), your typical sign-off, and any phrases you do or don't use.
Step 3: Give it three to five examples of your past emails. Find recent emails you've sent in each category that you were happy with. Paste them into the project as examples. The AI now has a calibrated sense of how you actually write. This step matters more than any other. Without it, the output will sound like a generic AI. With it, the output will sound like you on a good day.
Step 4: Use it for the first draft only. When a new email comes in that matches one of your categories, give the AI the relevant context in two to three sentences ("Client Joan Wexler needs a follow-up on her missing 1099-R from her brokerage; we have been waiting since April 22"), and let it draft. Read what comes back. Edit what you need to edit. Send.
Step 5: Refine the instructions every two weeks. When the AI drafts something you have to substantially rewrite, take 60 seconds to update the project's standing instructions. Tell it what you changed and why. Over a month, the drafts get closer and closer to ready-to-send.
What can go wrong, and how to handle it
The biggest risk in professional services is the AI making up facts that sound right but aren't. This is called a hallucination, and it's the single most important thing to understand about using AI in a regulated field. Never let AI draft anything that contains specific tax figures, citations to specific tax code sections, or claims about the client's specific filings unless you are reading and verifying every number and citation before you send.
A safer rule: use AI for the structure and tone of the email. Type the actual numbers and facts yourself, or have the AI leave placeholders for you to fill in. "Your refund of $[AMOUNT] has been processed; you should expect to receive it by [DATE]" is a perfectly good draft. The AI got you to 95% of the email. You fill in the 5% that has to be exact.
The second risk is putting confidential client information into a tool that retains it for training purposes. For both Claude and ChatGPT paid plans, your data is not used for training by default. Microsoft Copilot in a business plan operates the same way. But your team needs to know this and follow it. If someone on your staff uses a free consumer AI tool to draft a client email and pastes in the client's full SSN or financial details, that data may end up in a training set. Make it a written practice rule: paid business plans only, and no full identifiers in prompts.
The third risk is the email sounding too smooth and losing the friction that signals a real person wrote this. If your AI-drafted emails start reading like marketing copy, dial back the polish. Real emails have small imperfections. The point is not to write the most beautiful possible email. It is to write the email you would have written if you'd had more time.
What to expect
For most small accounting practices, this saves between 30 and 90 minutes a day in the first month, and more in later months as the instructions get refined. The setup itself takes two to three hours total: one hour to decide categories and gather examples, one hour to set up the projects, and 30 minutes of editing in the first week before the rhythm clicks.
The cost is modest. $20 to $30 per month for the AI tool. No new hires. No software integration project. If you decide it isn't working after 60 days, you cancel the subscription and you're out a one-time afternoon and two months of fees. Compare that to the $50,000-a-year hire it postpones, and the math is not subtle.
The biggest change is psychological, not operational. The Wednesday-morning inbox stops feeling like a wall. You handle it in a fraction of the time, and you reclaim the hours that used to disappear into routine drafting. That time goes back into the work that actually needs you: the client conversations, the planning sessions, the strategic advice that the AI can't and shouldn't be doing.
If you are thinking about whether AI fits your practice more broadly, two earlier posts on this site might help. The first is a look at what is really true about the 10x productivity claims. The second is a plain-language explanation of what an AI agent actually is. The honest answer is that AI is a tool, not a savior. Used well in narrow places, it gives you back hours. Used badly in broad ones, it costs you trust.
Start with one email category. Get it working. Add the next.
-- Stacey | The Standalone
About the Author
Stacey Tallitsch runs The Standalone, an AI Implementation Diagnostic practice for small business owners. He has 30 years of experience in technology and has written 21 books on systems thinking and decision-making. More than 30,000 students have learned from his online courses.
- Stacey Tallitsch, The Standalone