top of page
Image by Jess Bailey

Get finance and accounting tips delivered right to your inbox

ChatGPT for Accounting: Can I Trust It?

15 minutes ago
4 min read
ChatGPT for Accounting: Can I Trust It?

Try ChatGPT for Accounting on Your Next Month-End Close


Try ChatGPT for Accounting on Your Next Month-End Close

ChatGPT for Accounting prepares invoice entries, bank reconciliations, and a first draft of your monthly report. It flags anything unclear for you to review.


Download ChatGPT for desktop, install the Accounting plugin, and try an accounting task on your next close.


OpenAI was nice enough to sponsor this email. #OpenAIPartner #ad


OpenAI recently released ChatGPT for Accounting. They say it means less manual bookkeeping.


I'll admit… I've been hesitant to try it. Obviously I'm biased. I make a living doing this work for clients, so what happens to me when they can do it themselves?


At the same time, I knew I couldn't sit on the sidelines. I had to try it out. So I ran ChatGPT through the four parts of a month-end close, and I tracked what it got right and what I still had to check.


The first part of any close is getting transactions into the right accounts. I gave ChatGPT a sample company's October bank feed, 45 transactions with no categories, plus its chart of accounts and accounting policies.


About four minutes later, it had categorized 32 of them, and every one was right. It got the tricky ones too. An annual software plan went to prepaid expenses, and a customer deposit went to deferred revenue.


It split one loan payment into principal and interest, and flagged the other 12 for my review instead of guessing.


45 transactions sorted, with 12 flagged for review
45 transactions sorted, with 12 flagged for review

Only one flag wasn't needed. It asked whether a $2,400 laptop is a fixed asset… when the company's policy says anything $1,000 or more is. But asking is better than guessing.


Next comes the bank reconciliation. I gave ChatGPT the bank statement and the cash ledger, then asked it to flag anything missing, duplicated or different.


The sample files had seven differences planted in them, and it found all seven. One was a check booked at $1,540 that cleared at $1,450. And ChatGPT didn't change a thing in the books. It told me to verify that check before correcting the ledger.


The seven differences ChatGPT found in the bank rec
The seven differences ChatGPT found in the bank rec

I added one number myself, the corrected cash balance. ChatGPT tied the books to the bank's $297,150. But once the books are corrected, cash is $303,300, and that's the number the report needs.


Now, the third part is probably the toughest. Journal entries are the heart and soul of a month end close. This is where you decide how a cost gets split across months and accounts.


I gave ChatGPT the bookkeeper's month end notes, a list of eight open items. Seven needed an entry, and it got all seven right in about a minute. The eighth needed no entry at all, and I was really impressed that it caught that.


The payroll accrual, split by working days
The payroll accrual, split by working days

Then I checked its summary of what those entries did to the month. It said operating expenses went up $27,550. The right number is $20,950.


The entries were right. The math in the summary wasn't. It had left out a $6,600 credit from its own first entry.


Finally, the analysis. This is the one I expected ChatGPT to do best on. I gave it September's management report, plus October's P&L and balance sheet, and typed one prompt.


"Prepare this month's report following last month's report and records as an example. Cite your sources."


The prompt, with the Accounting plugin and three files attached
The prompt, with the Accounting plugin and three files attached

It worked for 1 minute and 47 seconds. What came back was October's report, laid out like September's, with the source cell listed next to every figure.


I traced one figure back. Marketing spend rose $9,500 to $31,900, and row 15 of the P&L said exactly that.


The draft report, with a source next to each figure
The draft report, with a source next to each figure

The written commentary is the part that still needs me. EBITDA fell $4,000 even though revenue grew 6.5%… and that line sits inside the bullet about professional fees. It belongs at the top.


And next month, I'll run the same prompt on November's records, with October's report as the example.

Looking back, what ChatGPT gave me is a first draft that still needs my review. But it does impress me.


In about ten minutes of working time, ChatGPT did much of the preparer's work in a close. The reviewer's work is still mine.


So what does that mean for my clients? Will they use ChatGPT directly, or will they keep working with me?


My take is that trust is a big part of everything we do. My clients now have the same tools I have. What they don't necessarily have is the expertise and judgment to check the work.


But that's just for today.


This email is sponsored by OpenAI. #OpenAIPartner #ad


There's another side to AI, though. This week I was in New York for FATE, the finance and accounting expo. It was my first time hosting four sessions at a conference, and my very first time hosting a dinner for fractional CFOs.


My first dinner for fractional CFOs, this week in New York
My first dinner for fractional CFOs, this week in New York

It reinforced one idea for me. The better these tools get, the more valuable it is to meet people in person and build real relationships.


But enough about me… What do you think? Would you trust ChatGPT with your month end close?


Hit reply and let me know.


Have a great weekend.


Josh (Your CFO Guy)

Thanks for reading my article. My goal is to help you grow in your career with proper Finance and Accounting.


If we're not already connected on Social media, let's get connected and send me a message to say hi. I promise to respond

 
 
bottom of page