Cycle Counting Program: Continuous Inventory Accuracy, No Annual Shutdown
Generate daily count tasks by ABC class, capture counted quantities, compute the variance against your system, and route every over-threshold adjustment through a controller's approval before it's exported or posted.
An internal web tool that generates daily cycle-count tasks by ABC rotation, lets counters record actual quantities (blind optional), computes variance against the system snapshot, routes every over-threshold variance through a controller's approval, and exports a clean inventory-adjustment CSV plus a running count-accuracy metric.
Before you start
- A Supabase account (free)
- A Vercel account (free)
- A stock snapshot export you already have: item, location, system quantity, and (ideally) ABC class — CSV or Google Sheet
- Your variance threshold for requiring approval
- Claude Code or any AI coding agent
The problem this kills
Once a year you shut the place down, hand out clipboards, and count everything at once. It's expensive, it's exhausting, and the numbers are stale the moment you reopen. Worse, between counts your system quantities quietly drift — a pick that wasn't logged, a putaway to the wrong bin, a part that got consumed off the books — and you don't find out until a line goes down because the screen said you had stock you didn't.
Cycle counting is the fix every operations textbook recommends and almost nobody runs well: instead of one big count, you count a small slice of your inventory every day, rotating through your items so the fast-moving, high-value A items get counted often and the slow C items get counted occasionally. Accuracy climbs continuously, you never shut down, and you catch drift while it's small. But running it by hand means juggling spreadsheets to decide what to count today, chasing down who counted what, doing variance math line by line, and arguing about which adjustments are big enough to bother posting.
You don't need a warehouse-management system to do this right. You can build the whole program yourself, this weekend.
What you'll build
An internal web tool your stockroom team logs into. You import your stock snapshot — item, location, system quantity, and ABC class — and the tool generates today's count tasks on an ABC rotation: A items come up frequently, B less often, C rarely. Counters open their task list, walk the floor, and record what they actually find (with an optional blind count mode that hides the system number so nobody just confirms the screen). The tool computes the variance for each count — units off, percent off, and dollar impact — and sorts the results.
Anything inside your tolerance is recorded and closed automatically. Anything over your variance threshold lands in front of a controller, who reviews the count, can request a recount, and approves the inventory adjustment before it goes anywhere. On approval you export a clean adjustment CSV in the exact columns your ERP or WMS import wants — and the tool tracks your count accuracy % over time so you can prove the program is working.
What's inside the Implementation Plan
The downloadable plan is a single markdown file you paste into an AI coding agent. It opens by interviewing you about your own stockroom — how you identify items and locations, your ABC cutoffs and count frequencies, your real variance threshold, whether you count blind, who's allowed to approve an adjustment, and your messiest edge cases (parts in two bins, unit-of-measure mismatches, in-transit stock). Then it reads a short spec back for your thumbs-up before it builds anything. That's the difference between a tool shaped to your floor and a generic template you have to fight.
From there it walks the agent through the data model (items, locations, the snapshot, count tasks, counts, and adjustments), the snapshot importer, the ABC rotation that generates daily tasks, the count-capture screen with blind mode, the variance engine, the controller approval gate, the adjustment export, and the accuracy metric over time. Every step ends with a ready-to-copy prompt. There's a full "No API yet?" path: import your stock CSV, export a clean adjustment CSV, and you never have to touch your ERP's API to ship.
The governance it includes (this is the point)
Inventory adjustments move money on your balance sheet, so the controls are the product. The plan builds them in: a login so only your team can use it; row-level security so each site or organization only ever sees its own inventory; a complete audit trail of who counted, who approved, and when; a hard human-approval gate so no adjustment is exported or posted until a controller signs off; and a duplicate guard keyed on the count task ID (item + location + count date) so the same count can't be processed twice. Over-threshold variances physically cannot be exported until someone with authority approves them. That's the audit story your controller and your auditors want.
Who it's for
Inventory controllers, stockroom leads, and material handlers who own inventory accuracy and are tired of the annual fire drill. If you can explain to a new hire which items get counted most often and what makes a variance worth posting, you can build this — no developer required.
You've got this — open the plan, paste the first prompt, answer a few questions about how your stockroom actually runs, and you'll watch today's count list generate itself.