Over/Short/Damage (OS&D) Discrepancy Reporter

The problem this kills

Your dock crew already sees the problems. A pallet arrives crushed. The packing slip says 50 but only 40 made it onto the truck. Someone ships you the wrong SKU. The receiver shrugs, scribbles a note, and moves on to the next truck - because there's no fast, formal way to turn "this is wrong" into "we're getting our money back."

So the credit never gets requested. The replacement never gets ordered. The damaged goods get quietly written off. Multiply that by every receiving door, every week, and you're leaking real money to suppliers who never hear a peep of complaint - which means they have zero reason to ship better next time.

The fix isn't a meeting. It's a dead-simple tool that turns a 30-second log entry plus a phone photo into a tracked claim that a buyer approves and a supplier can't ignore.

What you'll build

A small, login-protected web app your receiving team and buyers actually use:

• A dock-friendly logging screen. The clerk picks a PO and item, says what's wrong (over, short, damaged, wrong item), enters expected vs. received quantity, and snaps photos. Done in under a minute.
• Automatic case creation. Each discrepancy becomes a case, linked to its PO and receipt, with the estimated dollar impact calculated for you.
• A buyer review queue. The buyer opens the case, sees the photos and the math, and chooses the resolution path: claim, credit, return, or replacement.
• A drafted claim email - that never auto-sends. The tool writes the supplier email for the chosen path. The buyer reads it, edits if needed, and clicks send. Only then does Resend deliver it.
• Tracking to closure. Cases move from Open to In Progress to Closed, with follow-up reminders so nothing goes cold.
• Supplier discrepancy rates. Roll-ups that show which suppliers cause the most problems - the start of a real scorecard.
• A CSV export of open and closed claims, in clean columns you can hand to finance or drop into your ERP.

What's inside the Implementation Plan

The plan is a single file you paste into Claude Code. It walks the AI agent through the entire build, step by step, with a ready-to-copy prompt at the end of each step.

The most important part comes first: the plan opens by interviewing you about your business. Before a single line of code, the agent asks how your receiving process actually works, what your POs and SKUs look like, who your buyers are, your typical and peak volumes, and your messiest edge cases. It reads back a short tailored spec, waits for your thumbs-up, and only then builds - so you get a tool shaped around your dock, not a generic template you have to bend yourself into.

Inside you'll find:

• The full discovery interview (the questions that make it fit you).
• A data model for POs, receipts, discrepancy cases, photos, and an audit log.
• The dock logging screen, the buyer review queue, and the approval gate.
• The claim-email drafting and Resend send flow.
• Financial-impact math, supplier discrepancy rates, and follow-up reminders.
• The "No API yet?" path: import POs and receipts from a Sheet or CSV and export clean claim data back out.

The governance it includes (this is the point)

This isn't a toy. The plan bakes in the controls that make a tool safe to put in front of suppliers and finance:

• Login so only your team can use it.
• Row-level security so each organization only ever sees its own cases and POs.
• A complete audit trail - who logged what, who approved which resolution, and exactly when.
• A hard human-in-the-loop gate: the AI drafts the claim, but a buyer must review and approve before any email goes to a supplier. Nothing auto-sends.
• Duplicate guards keyed on PO line + delivery reference + discrepancy type, so the same problem can't spawn two competing claims.

Who it's for

Receiving clerks and dock supervisors who keep spotting short and damaged deliveries that never get chased - and the buyers who own the supplier relationship and the money. If "we definitely paid for stuff we never got, and nobody followed up" sounds familiar, this is your tool.

You've got this - paste the first prompt and let the agent interview you.