Most accountants who tell me they’ve “tried AI” did one thing: they asked ChatGPT a question, got a mediocre answer, and concluded AI isn’t ready.
The problem wasn’t AI. The problem was they picked the wrong task.
AI isn’t good at everything. It’s spectacularly good at specific types of work — repetitive, structured, pattern-based tasks with clear inputs and outputs. Accounting is full of those tasks. You just have to know which ones to target first.
Here are five. Each one can be automated this week — no new software, no setup, just a prompt you paste into ChatGPT or Claude. I’ve included time estimates based on my own tracking.
Time before AI: 2-3 hours per client per month
Time with AI: 30-45 minutes
Weekly savings: 2-3 hours (across all clients)
This is the single highest-ROI automation for any accountant. Not because it’s the most impressive — because it’s the most frequent.
Every client, every month, hundreds of transactions need categories. You know 80% of them on sight. The other 20% require investigation. AI handles the 80% instantly and flags the 20% for your attention.
The prompt:
“You are an experienced senior bookkeeper. Categorize these transactions using this chart of accounts: [PASTE YOUR CATEGORIES]. For each transaction: (1) category, (2) confidence High/Medium/Low, (3) brief reason. Flag anything unusual. Format as a table.
Transactions: [PASTE]”
The key insight: Paste your actual chart of accounts into the prompt. This single change takes accuracy from ~70% to 90%+. Without it, AI guesses at categories. With it, AI maps to your specific structure.
Start today: Export this month’s transactions for one client. Paste them with your chart of accounts. Review the output. Time yourself. You’ll be convinced in 10 minutes.
Time before AI: 10-15 minutes per email
Time with AI: 2-3 minutes per email
Weekly savings: 1.5-2 hours
Document requests. Follow-up nudges. Quarterly summaries. Fee increase notifications. Meeting recaps. You send the same types of emails hundreds of times a year. The content changes; the structure doesn’t.
The prompt:
“I’m going to paste 3 example emails I’ve written. Match my tone, vocabulary, and sentence style exactly. Then draft an email to [CLIENT] about [TOPIC]. Tone: [professional/casual/firm]. Length: [short/medium]. End with a clear next step.”
The key insight: The tone calibration step (pasting your own examples first) is what separates “useful AI email” from “generic robot email.” Without it, AI sounds like a LinkedIn post. With it, AI sounds like you.
Start today: Find three emails you’ve sent that represent your typical tone. Save them in a doc. Next time you need to draft a client email, paste those three examples first, then your request. Compare the output to what you would have written.
Time before AI: 15-20 minutes per meeting
Time with AI: 2 minutes
Weekly savings: 1-1.5 hours (at 4-6 meetings per week)
You leave a client meeting with a page of scribbled notes. Turning those notes into a professional summary with action items, owners, and deadlines takes longer than the meeting itself.
The prompt:
“Convert my rough meeting notes into a professional client summary. Format: (1) Key discussion points, 3-5 bullets. (2) Decisions made. (3) Action items with owner and deadline for each. (4) Next meeting date/agenda items.
Meeting: [CLIENT], [DATE]
Attendees: [NAMES]
My notes: [PASTE RAW NOTES]”
The key insight: The action items with owners and deadlines are the valuable part. Without them, a meeting summary is just a record. With them, it’s an accountability tool. AI structures this consistently every time — even when your notes are barely legible.
Start today: After your next client meeting, paste your notes into this prompt instead of typing up a summary manually. Send the result (after a quick review) within an hour of the meeting. Clients will notice the speed and professionalism.
Time before AI: 30-45 minutes per report
Time with AI: 10 minutes
Weekly savings: 30-60 minutes (during reporting periods)
You have the numbers. Budget vs. actual. Current vs. prior year. The math is done. What takes time is writing the narrative: why revenue was up 12%, why travel expenses doubled, why gross margin tightened.
The prompt:
“Analyze these budget vs. actual results. Flag variances over [5%] or [$X]. For each significant variance: (1) account, (2) budget vs actual, (3) variance $ and %, (4) likely explanation based on context provided, (5) one-time or recurring, (6) recommended action.
Client context: [INDUSTRY, SIZE, ANY KNOWN EVENTS THIS PERIOD]
Data: [PASTE BUDGET VS ACTUAL]”
The key insight: The “client context” line is critical. Without it, AI gives generic explanations (“revenue increased due to higher sales”). With context (“SaaS company, launched new product tier in March”), explanations become specific and useful.
The verification step: AI gets the math right. The explanations are educated guesses. Review each explanation against what you actually know happened. Replace generic ones with specific ones. The final product should read like you wrote it — because the insights came from you, the structure came from AI.
Start today: Next time you’re preparing a monthly report that includes variance analysis, run the numbers through this prompt before writing your narrative. Use AI’s output as a first draft and edit from there instead of starting from a blank page.
Time before AI: 10-15 minutes per workpaper
Time with AI: 3-4 minutes
Weekly savings: 1-2 hours (during close/audit periods)
Workpaper documentation is the task everyone does last, does fast, and does badly. Not because accountants are lazy — because by the time you’re writing workpaper descriptions, you’ve already done the actual work and your brain is done.
The prompt:
“Write a professional workpaper description for:
- Type: [RECONCILIATION / ANALYSIS / SUPPORTING SCHEDULE]
- Account/area: [DESCRIBE]
- Period: [DATE RANGE]
- What I did: [DESCRIBE PROCEDURES IN YOUR OWN WORDS]
- What I found: [RESULTS]
- Conclusion: [YOUR CONCLUSION]
Format as standard workpaper with: Purpose, Source of Data, Procedures Performed, Results, Conclusion.”
The key insight: You’re not asking AI to do the work — you’re asking it to document the work you already did. This is the perfect use case because all the judgment happened during the actual procedure. Documentation is just structuring what you already know.
Start today: Next time you finish a reconciliation or analytical procedure, paste your rough notes into this prompt instead of writing the formal documentation from scratch. The output is clean, consistent, and reviewer-friendly.
Here’s what these five tasks look like together:
| Task | Weekly savings |
| Transaction categorization | 2-3 hours |
| Client email drafting | 1.5-2 hours |
| Meeting notes to action items | 1-1.5 hours |
| Variance analysis narratives | 30-60 min |
| Workpaper documentation | 1-2 hours |
| Total | 6-9 hours/week |
Six to nine hours a week. That’s an entire workday, every week, recovered from tasks that don’t require your professional judgment.
And none of this required buying new software, migrating data, or spending a week on implementation. You open ChatGPT or Claude, paste a prompt, review the output, and move on.
If you’re starting from zero, don’t try all five this week. You’ll get overwhelmed. Here’s the order I recommend:
Week 1: Transaction categorization. Highest savings, most forgiving of errors (you’re reviewing everything anyway).
Week 2: Client email drafting. Second highest savings, and you’ll notice the quality difference immediately.
Week 3: Meeting notes. Quick win that changes how clients perceive your responsiveness.
Week 4: Variance analysis and workpaper documentation. These are period-end tasks, so they’ll naturally come up during your next close.
By the end of month one, all five are part of your workflow. Not because you forced it — because each one proved its value before you added the next.
These five tasks have something in common: they’re structured, repetitive, and the judgment happens before or after the AI step.
What stays manual:
AI handles the production. You handle the profession. That’s the split. It’s a good split.
All five prompts (plus 45 more) are in the free PDF, organized by category with verification checklists for each one. Download it here.
Get 50 copy-paste prompts delivered to your inbox — free.
Get the Free PDF →