The overbilling was $315,097.55. No one could catch it by hand.

A founder-led marine manufacturer that builds 20,000–30,000 parts a month.

They design and build equipment and accessories for OEM boat builders, dealers, and direct-to-consumer — out of a large in-house injection molding, CNC, welding and fab operation. To make all of that, they buy constantly: metals, actuators, and specialty materials, from a long list of vendors, on a purchasing operation that runs every single day. Every one of those invoices used to be checked by hand.

Checking invoices by hand doesn't scale — and it misses things.

At their volume, a 3% overcharge buried in one line of one invoice, three vendors deep, is invisible. Not because anyone's careless — because no person can line-match thousands of invoices against thousands of quotes every cycle. Here's what that cost them:

Invoice Intelligence — a system that reads every vendor invoice to the cent.

Custom-built for their actual vendors and actual formats, and deployed onto their own isolated infrastructure — not a SaaS they log into. It reads their whole supply chain: metals and industrial supply, aluminum extrusion, cast marine hardware, fasteners, paint and specialty materials, and actuator components. The vendor list has changed since launch — suppliers were added and one was retired — and the system absorbed that without a rebuild.

• Reads every format — digital, text-layer, and scanned (OCR) invoices, with no templates to maintain.
• Reconciles each invoice to the cent — every line extracted and checked against the invoice's own printed total.
• Checks each line against the quote — and flags overbilling before the invoice is paid, not after.
• Tracks per-SKU price history — month over month and as a year-to-date range, so price creep becomes visible.
• Keeps a human in the loop — the operator approves derived prices; nothing auto-commits without a check.
• Fails closed — per-line reconciliation and PII redaction on scanned documents, by default.

I sat with the finance team and built around the work they actually do.

The multiple reading paths exist because their real vendors demanded them. One vendor was quietly dropping line items until I caught the parser doing it and fixed it. Another sends copier scans with no text layer, so I built an OCR path from scratch. Every parser reconciles to the cent against the printed total before anything is shown — and the genuine edge cases are held for a person to review, not guessed at. The result is delivered as a branded, secure client portal with role-based access. The code and the data are theirs.

Real money surfaced, and still counting.

On top of the money: hours of manual reconciliation come off the finance team's plate every cycle, and for the first time they can see their spend down to the smallest part.

“A huge congratulations to Ayden Oswell and the team at PhosOps for developing such an impressive vendor spending and pricing analytics platform. The tool transforms complex purchasing data into clear, actionable insights, making it easy to identify pricing trends, monitor vendor performance, and uncover cost-saving opportunities in real time. It's intuitive, visually well-designed, and provides the intelligence needed for better procurement decisions. An outstanding example of how great software can simplify complex business challenges.”

John Runske · COO, Director of Product Design & Engineering

I didn't hand it off and vanish.

Vendors change, formats change, new parts get added. I run the system, monitor it, and keep improving it on a flat monthly, so it keeps up instead of going stale the way a one-time build would. They get the outcome without hiring an AI team to maintain it — and the code and data stay theirs the whole time.