In the restroom. In the hallway. Once, in the elevator.
Two people talking up ahead, me walking behind them, listening.
They'd glance back and spot me, freeze for a second, then go quiet.
I wasn't hurt, and I wasn't angry.
I'd already thought all the same things myself.
A fresh college graduate, used as someone else's workhorse, and when the dust settled she didn't even get table scraps.
That kind of story was a joke no matter where you told it.
If I'd been the bystander, I probably would've laughed too.
But let them laugh.
Sooner or later, some of those people wouldn't find it so funny.
I started studying the Finance Department's workflow in earnest.
The company's reimbursement process was painfully slow. From the moment an expense form was filled out to the moment the money actually hit someone's account, it averaged two weeks.
In between, a claim had to pass through the department supervisor, the finance audit, the finance director, and the general manager's signature.
Layer after layer. Any link in the chain could stall.
I spent a week compiling every reimbursement record from the past year into one dataset.
Sixty percent of all delays were stuck at the same bottleneck.
Invoice auditing.
Invoice auditing was entirely manual, one receipt at a time. The volume was massive and mistakes were easy to make.
Whenever a non-compliant invoice turned up, it got kicked back for resubmission, and just that round trip ate several days.
I thought: could a program handle this automatically?
Once the idea took hold, I started working on it after hours.
The company's accounting system was an older build that didn't support API integration, but it could export spreadsheets.
I studied the file format inside and out, then wrote a Python script that could automatically read each invoice number and tax code from the spreadsheet, run them against the tax bureau's public verification endpoint, and write the results back into the sheet.
The concept wasn't groundbreaking. Plenty of off-the-shelf accounting software could do the same thing. The company just hadn't purchased any.
So I built one myself. Free of charge.
Three days of coding and the core functionality was running.
I spent a few more days debugging, pushing the recognition rate from eighty percent to over ninety-five.
The remaining five percent were mostly invoices that had been photographed out of focus or partially blocked. Those still needed a human eye.
The program was ready, but I didn't rush to show it off.
Because I already knew.
In a new environment, what you actually did didn't matter. What mattered was what other people believed you did.
I kept going to work as usual, doing my tasks as usual, greeting my coworkers as usual.
Then one day, Marilyn came back lugging a stack of invoices, sweat beading at her temples.
"Overtime again. If I don't get through all of these by the end of the week, the monthly report won't close."
I looked at the stack in her arms. Two hundred receipts at least.
"Marilyn, let me help."
She glanced at me. "You've got your own work. I'll manage."
"I already finished everything on my end. Might as well make myself useful."
I took the invoices from her, sat down at my computer, and pulled up the program.
Marilyn didn't notice what I was doing. She'd already hunched over her own desk, cross-checking receipts by hand.
About fifteen minutes later, she looked up. "Susannah, you haven't gone through a single one yet?"
"Already done."
"What?"
"Two hundred and thirteen invoices. Six flagged with issues. I've marked every one."
Marilyn froze. She got up and walked over to my screen.
An Excel spreadsheet filled the display, every invoice's information laid out row by row, each status column reading either "Valid" or "Anomaly."
The six flagged entries had specific reasons noted in the remarks column.
Invoice number nonexistent. Issue date mismatch. Taxpayer ID incorrect.
Marilyn stared at the screen for a good ten seconds, then turned to me. "How did you do that?"
