Payroll paperwork that reads itself.

The bottleneck

Every timecard got typed. Twice.

Client companies send timecards the way clients always have: photos of handwritten cards, scans, PDFs, all into an email inbox. Staff read each one, matched the worker to the right roster record, keyed the hours into a payroll workbook, then built the client invoice from the same numbers.

Every client company formats its timecards differently, worker names on a handwritten card rarely match the roster spelling exactly, and payroll deadlines do not move. The typing was not just slow: it was slow at exactly the moments the firm could least afford slow.

The work is exact, repetitive, and entirely readable by a machine. It just never had one.

How we did it

1. 1

   Keep the intake. The firm set one requirement from the start: its client companies would keep sending timecards exactly the way they always have, and nothing about the fix could land on them. We built to that. Timecards arrive in the same inbox they always have, in whatever format each client sends, and the system picks them up from there.

2. 2

   Read and match. It reads worker, dates, hours, and overtime from each image or PDF, handwriting included, and matches every worker to the correct roster record even when the name on the card is abbreviated or misspelled. Anything it can't read confidently gets flagged, never guessed: a wrong guess in payroll costs more than a flag ever will.

3. 3

   Verify, don't type. A processor reviews the extracted data against the source image on one screen, corrects what's flagged, and the payroll workbook and client invoice come out in the firm's existing formats. Downstream, nothing changed; upstream, the typing is gone.

Where it stands

Deployed to remove the typing, not the judgment.

The system deployed in September 2026 and is rolling out across the firm's client companies. Every number still passes a human before it reaches payroll; the hours of transcription in between are what go away.

The firm owns the system, and each client company switches over as its timecard formats are verified against real submissions, so accuracy is proven per client before the old way is retired. The staff who used to type now review.

Your version of this bottleneck is fixable too.

Thirty minutes on your company and what you want from AI. If it's not a fit, we'll say so on the call.