Count Variance Reconciler & Adjustment Approval
Compare counted vs system on-hand, see the dollar impact instantly, require a reason code on every difference, and route large adjustments for sign-off — so no inventory adjustment posts without being deliberate, explained, and auditable.
A web tool where you import count results and system on-hand, and it computes the unit and dollar variance per SKU, requires a reason code on every difference, and routes adjustments by dollar value — a manager approves smaller ones, finance signs off on the big ones. Approved adjustments export as a clean CSV ready for your ERP, recurring-variance SKUs get flagged for recount, and every action lands in an audit log.
Before you start
- A Supabase account (free)
- A Vercel account (free)
- A count-results export and a system on-hand export (CSV is fine)
- A unit-cost list and your adjustment-approval thresholds
- Claude Code or any AI coding agent
The problem this kills
The count is done. Now someone has to figure out what it means. You've got a stack of counted quantities and a system on-hand report that disagree, and the question every line raises is the same: do we trust the count, or do we trust the system? Get it wrong and you've either hidden a real loss or invented a phantom one — and either way the general ledger now believes something that isn't true.
So the variance work happens in a spreadsheet at the end of a long day. Counted minus system, eyeball the big ones, maybe multiply by a cost to see the damage, and then the part that actually matters gets skipped: why. Why is bin A-12 short 40 units? Nobody wrote it down. The adjustment posts to the ERP with no reason code, no second look, and no record of who decided. Next quarter the same SKU is off by the same amount, and there's no way to know it's a pattern because last quarter's reasoning evaporated.
This is exactly the kind of high-stakes, rules-based reconciliation a small internal tool does far better than a spreadsheet — and you don't need to be a developer to build it.
What you'll build
A simple internal web tool for your cycle counts and physical inventories. You import two files — your count results and your system on-hand — plus a unit-cost list. The tool matches them by SKU and location, computes the variance in units and in dollars, and shows you the whole thing on one screen sorted by impact. Every line with a difference requires a reason code before it can move forward — damage, miscount, mis-pick, receiving error, shrink, unit-of-measure mix-up — so the why is captured while it's still fresh.
Then it routes by money. Small adjustments go to a warehouse manager to approve; anything over your dollar threshold also needs finance to sign off. Nobody can post an on-hand change without that approval. On approval the tool produces a clean adjustment CSV in the exact columns your ERP imports, flags any SKU that keeps showing variance so you can schedule a recount, and writes every step to an audit log.
What's inside the Implementation Plan
The plan is a single file you paste into an AI coding agent. It opens by interviewing you about your business — how you count today, what your ERP and SKU/location conventions look like, the exact columns in your count and on-hand exports, your reason-code list, your approval dollar thresholds, and your nastiest edge cases — and then tailors the data model, the variance math, the validations, and every later step to your answers. This is a build shaped around your warehouse, not a generic template.
From there it walks the agent through the database schema, the CSV imports with their duplicate guards, the variance engine that values every difference at unit cost, the reason-code requirement, the tiered manager/finance approval gate, and the export of a GL-ready adjustment file plus a recount watchlist. Every step ends with a ready-to-copy prompt. Because the whole thing runs on CSV in and ERP-ready CSV out, you can build and use it this weekend even if you have no direct connection to your system of record.
The governance it includes (this is the point)
This touches on-hand quantities and the general ledger, so it's built like it matters: login so only your team can use it, row-level security so you only ever see your own organization's data, and a complete audit trail of every import, reason code, edit, approval, and export — who did what, and when. Nothing adjusts on-hand automatically: each adjustment is a draft until the right person approves it, and approval is the hard human-in-the-loop gate — a manager for smaller values, finance above your dollar threshold. Duplicate guards on the count line and the adjustment mean the same count can't be processed twice and the same adjustment can't be posted twice.
Who it's for
Inventory control specialists, warehouse managers, and finance controllers who own stock accuracy and want every adjustment to be deliberate, explained, and defensible — the same way every count. If you can explain how you decide whether to trust a count, you can build this.
You've got this — open the plan, paste the first prompt, and let it interview you about your counts.