A Real Use Case for AI

A vendor sends monthly invoices as PDFs. Not summaries—full transaction-level detail. All the data you want, hermetically sealed inside a document designed primarily for printing and filing, not analysis. The data is valuable. The data is locked.

 

Hey, why not just demand the vendor send the invoices directly in Excel? Sorry, we’re small. No one even heard our pleas from the bottom of the Grand Canyon. We had to improvise like we always do on a pauper’s budget.

 

So we do what practitioners do. We buy desktop software to convert PDF to Excel. Worth every penny. Later, Excel starts parsing PDFs directly. Incremental progress. Not perfect, but workable. Yes, it takes hours each month correcting obvious misreads across roughly 30 statements. Still a positive ROI.

 

Because once the data is unlocked, it earns its keep.

 

It reveals real money—savings from steering customers from one payment method to another. And better yet, it lets us balance those savings against how fast each method liquidates receivables. This isn’t academic. This is cash flow, timing, and tradeoffs. Operational reality.

 

Then enter ChatGPT.

 

“Hey Chat, can you do this?”

 

Absolutely.

 

Thirty statements at a time?

No.

 

Ten at a time?

Sure.

 

Internal checks for accuracy and completeness?

Of course.

 

So why don’t the line items add up to the monthly statement total?

 

Ah yes. The old joke.

 

“I’m really fast at math.”

 

Okay, what’s 4,259 divided by 7?

 

“60.”

 

I said I was fast at math.

 

And just like that, hours of work shrink to minutes—by occasionally sacrificing accuracy. Which is… suboptimal in accounting. Accountants are persistently anal for a reason. Precision is not a personality trait; it’s the job.

 

So we keep training. Marathon mindset. New shoes

 

Enter Claude.

 

Claude gets it. With the right prompting, Claude checks his own work. Still limited to ten statements at a time, but now the process reliably takes minutes, not hours. A real step change.

 

At this point, many would declare victory. But marathoners don’t stop at “pretty good.”

 

This happens every month. Re-explaining the same task to an AI is not a process. It’s cardio, but it doesn’t scale. The goal isn’t clever conversations—it’s automation.

 

So we push further. Claude builds a desktop app.

 

Voila.

 

We’re down from hours wasted unlocking data to seconds. The assistant who used to spend her days parsing PDFs can now do higher-value work—or enjoy a leisurely coffee klatch. Either way, a win.

 

Lesson learned.

 

Do not blindly trust AI. It makes mistakes. Worse, it confidently affirms incorrect conclusions. Any mistake you can make manually, AI can multiply at machine speed. Buyer beware.

 

But for those persistent marathoners willing to experiment, validate, retrain, and occasionally swap shoes—Adobe, Excel, Excel add-ins, ChatGPT, Claude, Gemini—these tools are genuinely powerful.

 

The new running shoes really do feel like magic. For those willing to put in the miles.

 

Like an Olympic skater making the Triple Lindy look effortless (yes, Rodney Dangerfield—no respect), the hard work disappears only at the end.

 

AI takes effort. Judgment. Controls. Fit-for-purpose selection. And patience.

 

Not a threat.

A godsend.

 

Embrace it.