Use case #32 · CPA
CPA team finds AI flags variances but gets three of twelve explanations wrong
- Review/QC
- Beginner
- AI assistant (flux-analysis model not specified)
- Source checked · practitioner follow-up
The short version
What they did
Krogue followed his team’s AI-assisted variance work through four closes. The team logged explanation errors and changed review to require evidence for the claimed cause, including records outside the ledger. The article references an earlier ChatGPT-in-Excel search but does not identify the model for every flux-analysis run.
Result (as claimed)
Krogue reports twelve detected variances across four closes: “Then it explained three of them wrong.” Two explanations were plausible enough that Krogue says they would have survived normal review. This is the author’s small sample, not a general accuracy rate.
Prompt
I’m an accountant reviewing changes between [current period] and [prior period]. I’ll attach [both trial balances or income statements] and [available contracts, invoices or notes]. Flag accounts that meet [my dollar or percentage threshold].
1. Calculate each flagged change and show the source rows, both balances, dollar change and percentage change.
2. Separate the measured change from any proposed explanation. Cite the document and passage supporting each claimed cause.
3. Mark explanations without support as unresolved and say what evidence I should seek.Return a table followed by a short reviewer checklist in plain English. Don’t treat a plausible story as evidence or infer a cause from the ledger alone. Show how you handle zero or negative prior balances instead of reporting a misleading percentage. List files or rows you couldn’t read.
Do it yourself
These steps are my suggested way in, not Jacob Krogue’s setup.
What you need
- A chat assistant that reads spreadsheets, such as Claude or ChatGPT.
- Two periods of a trial balance or income statement.
- The contracts, invoices or notes that explain big changes.
- Use a business plan of the AI tool, or sample data, for anything with client details in it.
Start simple
- Use a past close where you already know why each account moved.
- Paste the Prompt with both periods and your threshold for a variance.
- Compare its list of variances and causes with what really happened.
- Mark each explanation right, wrong or unsupported.
Take it further
- Attach the evidence, such as contracts or amendments, and require it to cite a source for every cause.
- Keep a log of wrong explanations across closes, as Krogue’s team did.
- Turn its output into a reviewer checklist for each close.
Check your work
- Treat every explanation as unproven until you see the evidence behind it.
- Watch for plausible wording. Two of Krogue’s wrong answers would have passed normal review.
- Recompute the variance amounts yourself for the biggest items.
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.