Home › Operations & Admin

Reconcile physical stock counts against your inventory system

DifficultyMedium
SetupNo-code
Best forOperations
ToolsGoogle Sheets or Excel + ChatGPT or Claude, NetSuite, Cin7, inFlow Inventory, Zapier Agents, Make or n8n + an LLM
The problem

After a cycle count or a year-end stocktake, someone has to line up the counted quantities against what the inventory system says, find the mismatches, work out why they happened and post adjustments. It is slow spreadsheet work, small variances get waved through, large ones get investigated late, and the adjustments often go in without a clear record of why.

How the agent works

The agent takes two inputs: the count file (from a scanner, a count sheet or a spreadsheet) and an export of system quantities for the same locations and date. It matches rows by SKU and location, calculates the variance in units and value, and sorts each line against tolerance thresholds you set: within tolerance, worth a recount, or needs investigation. For lines over the threshold it looks for likely explanations in the records you give it, such as goods received but not booked in, returns not yet processed, unit-of-measure mix-ups or a transfer logged to the wrong location, and labels each one as a hypothesis, not a finding. It then drafts an adjustment batch with the variance, the suggested reason code and the evidence, and puts it in a review queue. Nothing posts to the inventory system until a person approves it, and every approved line keeps its evidence as an audit trail.

Setup steps
  1. Pick one location or product group and agree a count date so counted and system quantities refer to the same moment
  2. Standardize the inputs: SKU, location, unit of measure and quantity in both the count file and the system export
  3. Set tolerance thresholds in units and in value, and stricter ones for high-value or high-risk items
  4. List the records the agent may read for explanations: receiving logs, open purchase orders, returns, transfers and recent sales
  5. Have the agent produce a variance report and a draft adjustment batch with a reason code and evidence per line
  6. Send lines over the threshold to a recount before anyone investigates
  7. Have the stock or finance owner approve the batch, then post it through your system's normal import or adjustment screen
  8. Review recurring causes each month and fix the process that produces them, such as receiving steps or unit-of-measure setup
Tools that can do this
  • Google Sheets or Excel + ChatGPT or Claude
  • NetSuite
  • Cin7
  • inFlow Inventory
  • Zapier Agents
  • Make or n8n + an LLM

Not ranked. Tools marked Featured pay for a highlighted listing; that never decides which tools we include. Check each vendor's current features and pricing.

Make one of these tools? Feature your tool on every playbook where it is already listed.

Human in the loop

A person approves every stock adjustment before it posts, and recounts are always physical. The agent never writes directly to the inventory or accounting system. Write-offs, suspected theft and anything over your value threshold go to the finance or operations owner. Changing a tolerance threshold is a human decision.

Risks and guardrails

Mismatched units (each versus case) and counts taken at a different time from the export produce false variances, so check both before trusting the report. The agent's explanations are guesses from the records it can see; treat them as leads for a person to confirm. Auto-posting adjustments can hide real loss, so keep the approval step and the evidence. Stock adjustments usually affect the books, so involve whoever owns your inventory accounting and follow your auditor's requirements for count records.

ROI notes

Measure against your own baseline: hours spent reconciling each count, how many variance lines needed a recount, how long it takes from count to approved adjustments, and how often the same cause comes back. The value is faster, documented reconciliations and earlier attention on real problems, not fewer counts.

Also useful: reconcile bank transactions the same way; analyze the variance spreadsheet.

All AI agent examples, by function · More Operations & Admin playbooks