The $70k Save
Rusty Williamson
A 4-person manual process became a zero-touch pipeline. Here's the exact stack and what broke along the way.
I just finished an automation build for a roofing company that was running their entire inventory and procurement operation by hand.
Two full-time data entry people. Invoices processed line by line. Pick tickets created from scratch every morning. Reorder decisions based on someone remembering to check stock levels.
The kind of operation where errors don't just slip through. They compound.
Here's what I built:
The system runs on Python, Airtable, OpenAI, and Claude. Five interlocking pieces:
AI agents scan incoming vendor emails, extract line items, quantities, and pricing, then push everything into inventory. No human touches it.
Automatic verification that received counts match what was invoiced. Stock levels update in real time.
Every evening, the system generates the next day's worklist and pick tickets. Ready before the warehouse crew clocks in.
Once product ships, an invoice fires to accounts receivable automatically. No handoffs.
A watcher agent monitors par levels across every product. When stock hits a reorder point, it generates a PO and routes it to the manager's phone for approval. One tap, and it goes straight to the vendor.
What broke along the way:
The hardest part wasn't the automation logic. It was mapping the actual process. The way people described their workflow and the way they actually did it were two different things. I spent more time in the warehouse watching people work than I did writing code.
The invoice extraction was messy too. Vendor invoice formats are wildly inconsistent. Getting AI agents to reliably parse 15+ different vendor templates took more iteration than I expected.
The results:
Eliminated 2 full-time positions, saving $70k/year
Reduced inventory loss by 7% in year one
Zero manual invoice processing
Reorders happen automatically (no more stockouts because someone forgot)
Pick tickets generated overnight, every night
What used to take 3 people and a prayer is now a single manager reviewing purchase orders on their phone.
If your business still runs on "someone remembers to check," that's not a process. That's a liability.
Like this project
Posted Jun 14, 2026
A roofing company was burning $70k/year on manual inventory ops. I replaced the entire workflow with a closed-loop automation system using Python, Airtable, and AI agents. Here's the full stack, what broke, and the results.
Likes
0
Views
0
