A bookkeeper messaged me this morning. She has a new client in the US, and she was honest about what she needed. In her words: upload about 100 PDF invoices, turn them into a CSV in a few seconds, do the same with bank statements, and save herself the hours it normally eats. Right now she does it by hand through ChatGPT, and she is in the middle of building a whole system in Excel so the client can dump everything in one place.
I want to walk through her exact situation, because it is the single most common request I get, and because how you solve it is a good test of whether you actually understand AI or are just poking at it. This is the work every bookkeeper dreads, and it is also the work AI is best at. Let me show you why.
First, give her credit
Her instinct is right. She looked at a repetitive, soul-draining task and thought "a machine should do this." She is already using ChatGPT for it, which is more than most. And she is trying to build a clean intake system instead of accepting the chaos. That is exactly the right direction.
The problem is the tool and the method, not the thinking. Doing 100 invoices by hand through a chat window is using a screwdriver as a hammer. It works, barely, and it is exhausting. There is a better way, and it is not more effort. It is less.
The work is really two jobs, not one
When you break her request down, there are two distinct jobs hiding inside it.
The first job is documents to data. One hundred PDF invoices need to become one clean, structured file: vendor, date, invoice number, amount, maybe a category. That is an extraction job. The skill is reading messy, inconsistent documents and pulling the same fields out of every one, reliably, even when invoice number forty seven is laid out completely differently from invoice number three.
The second job is bank statements to a reconciled position. A bank statement, or an export, needs to be matched against the books so she knows what cleared, what is outstanding, and what does not belong. That is a reconciliation job, and it is a different skill entirely. It is not just reading a document, it is comparing two sources and explaining the differences.
The reason her ChatGPT approach feels heavy is that she is doing both jobs manually, one prompt at a time, and re-explaining the format on every batch. A purpose-built agent already knows the job. You do not teach it the fields every time. You hand it the stack and it runs.
What the invoice job looks like with an agent
Instead of pasting invoices into a chat and asking nicely, you point a receipt and invoice capture agent at the batch. It reads each PDF, finds the same fields across all of them despite the layout differences, and gives you back one structured file you can drop straight into the books or her Excel system. The hundredth invoice takes the same effort as the first, which is to say almost none.
The afternoon she described becomes a few minutes of review. Not zero minutes, a few. Which brings me to the part most people skip.
The part nobody tells you: you stay in control
Here is the honest version, and it is the version that separates people who lead with AI from people who get burned by it. An agent is fast and consistent, but it is not infallible, and accounting is not a field where "mostly right" is acceptable. So the workflow is never "let the robot do my books." The workflow is "let the agent do ninety five percent of the keystrokes, then you review the exceptions."
The agent flags what it is unsure about. You check the handful of odd ones instead of keying all hundred. You sign off. That is the difference between a bookkeeper who has been replaced and a bookkeeper who now does in twenty minutes what used to take a day, with her judgment still on top of it. The judgment is the job. The typing was never the job.
That is also the answer to the fear underneath every one of these messages. AI does not remove the bookkeeper. It removes the part of the bookkeeping that was never worth her time.
The bigger lesson for the firm
So here is what I told her, and what I would tell you. Stop building the whole system by hand, and stop running production work through a general chat window one batch at a time. The extraction and the reconciliation are solved problems. Use an agent built for each, keep your review step, and put your energy into the Excel intake she is designing, which is the genuinely valuable part because it is specific to her client.
And notice the move she is one step away from making. She is doing this for a US client and saving herself hours. The firm that figures this out does not just save time, it turns the saved time into a service. "We process your documents with AI and hand you clean books" is a real offer her clients would pay for. That is how a back-office cost becomes a front-office product.
Try it on your own stack
If you have a pile of invoices or a bank statement that is eating your week, do not take my word for it. Watch it work on your own files.
- Claim your first AI agent free and point it at a real task: https://aiaccounting.legacysbc.com/claim-agent
- Join our free AI for Accounting community, where bookkeepers and firm owners trade what is actually working: https://aiaccounting.legacysbc.com/community
- Or give me fifteen minutes and I will set one up live on your workflow, the invoices and the bank statements, so you walk away with it running, not just downloaded: https://calendly.com/simpleacademy/steve-s-private-calendar-15-clone
The work you dread is the work AI does best. The only question is whether you keep doing it by hand, or let the machine do the typing while you keep the judgment.
Yvonne
No comments:
Post a Comment