Use case #25 · CPA
Firm runs an AI review agent over year-end financials and returns before a human reviewer sees them
- Review/QC
- Intermediate
- In-house AI review agent (the firm builds its systems on Claude), Xero
- Source checked · practitioner article
The short version
What they did
The firm’s year-end review agent takes the financial statements and tax returns for a family group and runs a full review: it suggests journals to fix errors in the Xero file, flags issues such as missed Division 7A interest, and finds where a preparer’s numbers do not tie back to the workpapers. Preparers run jobs through it before submitting for review. The firm also ran it across hundreds of year-end jobs to find patterns in review points.
Result (as claimed)
“What used to be a twenty-minute exchange, where the preparer has to get their head back into the job, becomes a direct question the reviewer can ask the agent while the file is open.” Across hundreds of jobs: “Systemic errors surfaced. That pointed us straight to focused training.”
Prompt
I’m preparing a year-end job for a manager’s review. I’ll attach [financial statements], [tax returns] and [workpapers] for [client group], covering [year and jurisdiction]. Use [our review checklist and supplied tax notes] if included.
1. Trace statement and return figures to the workpapers. List amounts that don’t agree or have no traceable support.
2. Flag possible errors and missed adjustments, including related-party loan interest where the documents support checking it.
3. Propose correcting journals only where the evidence supports the accounts and amounts.Return a findings table with file and page or cell references, both amounts, the issue and the proposed next step. Finish with a short list of unresolved questions. Don’t guess tax rules or missing figures, post journals or change files. Clearly distinguish suspected errors from confirmed differences.
Do it yourself
These steps are my suggested way in, not Ben Walker’s setup.
Level: Intermediate for running it across many jobs. Reviewing one job in a chat is Beginner.
What you need
- A chat assistant that reads PDFs and spreadsheets, such as Claude.
- A year-end job already reviewed: statements, return, workpapers and the review notes.
- Use a business plan of the AI tool, or sample data, for anything with client details in it.
Start simple
- Pick a finished job that had review points.
- Paste the Prompt and attach the statements, return and workpapers.
- Compare its findings with the reviewer’s notes.
- Add anything it missed to the prompt as something to check.
Take it further
- Have preparers run it on each job before they submit it for review.
- Save the prompt and your firm’s review checklist in a project.
- Collect its findings across many jobs to spot repeat errors and plan training.
Check your work
- Check each proposed journal entry against the workpapers before you post it.
- Confirm every “doesn’t tie” finding yourself. It can misread a schedule.
- It comes before the reviewer, not instead of one. A person still signs off.
Before you rely on it
- This is general information about how accountants use AI, not professional guidance for your engagements.
- You stay responsible for the work under your professional standards, including review and sign-off. An AI tool’s output is a draft, not evidence.
- Before you send client data to an AI tool, read the tool’s data and training terms, and get any client consent you need. For tax return information, that includes IRC section 7216.
Source checked. The link, author, date and quote were confirmed against the source. The result itself has not been reproduced. Prompt adapted for this library, inspired by the case. Business size: Not stated. Results are quoted as the source claimed them and never improved. Spot an error or a broken link? Email charlie@cbarmorecpa.com.