She typically spends hours each week processing vendor invoices. She reviews the job details, chases down discrepancies, confirms the numbers, and logs them for payment. She is highly skilled in this area. She's performed the work a thousand times. She is no longer gaining anything from the repetition. She is ready for an AI-powered tool to free her from this tedious repetitive work while she remains in control and retains ownership of the outcomes.
The way she teaches it is simple. She records her screen while she works through a bill and talks about what she's doing. We get four useful things out of that single recording, and she never has to fill out a form or explain her job in a meeting.
What one recording gives us
It gives us a time. Her first recording ran just under four minutes for one bill.
It gives us her reasoning. The AI transcribes her narration, so we know why she took each step, not only which screens she opened.
It gives us the evidence. The AI pulls stills from the key moments, so we can see what was actually printed on the bill.
And it gives us rules. After the AI watched her first recording, it came back with a few questions her narration didn't answer, like what she does when a client hasn't paid yet. She answered them in a short Slack reply, and those answers became rules too.
The exceptions become the rules
One bill came in and the job number matched nothing when she searched QuickBooks and Asana. The bill also named the client and described the product, so she searched for those and found the job.
The vendor had swapped two digits. The wrong number was in both the PO field and description line. It's standard practice for the AI to compare those two places and flag the bill when they disagree. On this bill they agreed, so that check would have caught nothing.
The AI rule was created. Going forward it will check both places on the bill, and when the number matches nothing, it will search by client and product to locate the matching job.
We put on our data scientist hat
The point of timing her is the case study. We want to know how much time this tool actually saves her, and that only means something if we measure the work before the AI was involved.
Training is simple. She presses record. She works at her normal pace, narrates lightly, and the recordings will capture everything we need.
When she begins using the AI tool, we'll compare it against the baseline and measure the lift. The tool will perform the research in advance and show her the results. She simply reviews and approves. If something doesn't make sense, it's another training opportunity. The human stays in the loop and the AI system learns from every correction.
This is part one. Once she's working with the tool, we'll come back with the before-and-after numbers and what she's doing with the hours she gets back. The goal is to empower her to spend that time on work she's still growing from. In the meantime, think about who on your team has done the same task a thousand times.