Ops & Supply

How to audit inventory shrinkage with AI

Ops & Supply3 AI tools7 steps6 friction points

Want an agent to handle this on your live data? See the Starch version →

Inventory shrinkage — the gap between what your records say you have and what's actually on the shelf — is one of those problems every product-based operator knows about and almost nobody has a clean system for. It shows up in your COGS, your margins, and your end-of-quarter reconciliation, and the causes range from spoilage and theft to receiving errors and data entry mistakes. Finding out where the gap lives requires pulling data from multiple places and comparing numbers that were never designed to line up neatly.

AI feels like a natural fit here because the workflow is fundamentally analytical: you have two sets of numbers, they don't match, and you need to systematically identify where the discrepancy came from. The comparison logic isn't complicated — it's the volume and messiness of the underlying data that makes it tedious. An LLM can categorize discrepancy types, write formulas, generate audit checklists, and help you structure a reconciliation process that would otherwise take hours of manual cross-referencing.

ChatGPT, Claude, and Gemini can all contribute meaningfully to this workflow today — primarily as thinking partners and structured-output generators. They can help you design an audit methodology, interpret patterns in the data you paste in, draft supplier dispute templates, and convert raw shrinkage figures into a categorized loss report. The constraint is everything that requires your actual live data: they can't pull from your systems, they don't persist between sessions, and you're doing all the data assembly yourself.

Skip the manual work → See the Starch version

Ops & Supply3 AI tools7 steps6 friction points

AI walkthrough

How to do it with AI today

A practical walkthrough using ChatGPT, Claude, and other off-the-shelf LLMs — what they're good at, what you'll have to do by hand.

Tools that work for this

ChatGPTClaudeGemini

Step-by-step

1 Export your theoretical inventory (from your ERP, spreadsheet, or POS) and your physical count results. Clean the file down to SKU, location, expected quantity, actual quantity, and unit cost, then paste the data directly into Claude or ChatGPT — keep the dataset under 500 rows to stay within reliable context.

2 Ask the LLM to calculate shrinkage by SKU and location, sort by dollar impact, and flag any line where the variance exceeds a threshold you specify (e.g., more than 5% or more than $200). Ask it to output a clean table you can paste back into a spreadsheet.

3 Paste the ranked discrepancy list back into a fresh prompt and ask the LLM to categorize each variance into likely cause buckets — receiving error, spoilage, counting error, undocumented transfer, or theft. It will give you educated guesses based on patterns, not definitive answers.

4 Use Claude or ChatGPT to generate a root-cause investigation checklist tailored to your top 10 discrepant SKUs. Describe your operation (e.g., 'we use a co-packer and a 3PL, and product moves between two warehouses') so the checklist reflects your actual supply chain steps.

5 Ask the LLM to draft a supplier-facing discrepancy notice for any shortages you suspect originated at receiving — include fields for lot number, PO reference, quantity expected vs. received, and a response deadline. Copy the template into your own document.

6 Run a summary prompt: paste your categorized shrinkage table and ask the LLM to write a one-page audit memo summarizing total shrinkage, breakdown by cause category, estimated dollar impact, and recommended corrective actions for each category. Use this for internal reporting or retailer conversations.

7 Repeat this process next quarter — manually. Nothing from this session carries forward automatically; you'll reconstruct the workflow from scratch each time you want to run it.

Prompts you can copy

Here is my inventory count vs. expected quantities by SKU and location [paste data]. Calculate the variance in units and dollars, sort by dollar impact descending, and flag any SKU where variance exceeds 5% or $150.

Based on the shrinkage variances below [paste ranked list], categorize each into the most likely cause: receiving error, spoilage or expiration, counting error, undocumented internal transfer, or theft. Add a confidence column (high/medium/low) and a one-sentence rationale.

I run a CPG brand. Product moves from a co-packer in Ohio to a 3PL in New Jersey, and then to two regional warehouse locations. Generate a step-by-step audit checklist for investigating a 12% shrinkage variance on SKU #1047 across those locations.

Draft a supplier discrepancy notice for a receiving shortage. Include fields for PO number, lot number, expected quantity, received quantity, unit cost, total dollar discrepancy, and a 10-business-day response deadline. Keep the tone firm but professional.

Here is my categorized shrinkage data for Q2 [paste table]. Write a one-page internal audit memo summarizing total shrinkage by dollar value, breakdown by cause category, and three recommended corrective actions prioritized by potential dollar recovery.

Reality check

Where this gets hard

The walkthrough above works — until your numbers change, the LLM hallucinates, or you have to re-paste everything next month.

No live connection to your inventory system, 3PL portal, or ERP — you manually export and paste data every single time you want to run this audit.

Dataset size limits what you can actually analyze in one session; if you have more than a few hundred SKUs across multiple locations, you're either truncating or splitting the audit into multiple disconnected prompts.

Nothing persists between sessions — the categorization logic you refined last quarter, the threshold rules you landed on, the memo format that worked — all gone. You rebuild from scratch next audit cycle.

Outputs are inconsistent across runs — the same data pasted into the same tool on different days can produce different category labels, different column structures, or different variance calculations depending on model version and context.

The LLM cannot cross-reference your actual lot records, receiving logs, or transaction history — it can only reason about whatever you paste in, which means root-cause categorization is educated guessing, not a data-backed finding.

No alerts, no scheduling, no automation — inventory shrinkage audits only happen when you manually decide to do one, which in practice means they happen less often than they should.

Tired of the friction?

Starch runs the whole workflow on live data — no copy-paste, no hallucinated numbers, no re-prompting next month.

See the Starch version →

Starch alternative

The same workflow on Starch

Starch is an agentic operating system — agents build and run the software your work depends on, connected to your live data. For inventory shrinkage audits, that means a persistent app that compares actual versus expected quantities on a schedule, flags variances automatically, and keeps a running record of every audit cycle — not a prompt you re-run manually each quarter.

Starch connects to 3,000+ apps through its integration catalog. Connect your ERP, spreadsheet, or 3PL portal once — the agent queries live data when your shrinkage dashboard runs, so the numbers reflect what's actually in your system today, not last month's export.

Starch's Transaction Insights app syncs your Plaid-connected bank accounts on a schedule, automatically flagging vendor charges that look anomalous — a co-packer invoice that's 40% higher than last month's gets surfaced before it becomes a shrinkage mystery downstream.

Lot Tracker — currently in development, request beta access — will give you full lot-level traceability from supplier to shelf, FSMA 204-compliant chain of custody, and mock recall capability in minutes. When an auditor asks for every customer who received a specific lot, you pull it up immediately instead of assembling it from spreadsheets.

Inventory Planner — also in development, request beta access — will give you a single view of stock across your co-packer, 3PL, and warehouse locations with shelf-life tracking and first-expired-first-out rotation built in, so shrinkage from expiration becomes visible before it happens.

Describe the audit app you want in plain English and Starch builds it: 'Every week, compare my inventory system quantities against my 3PL's reported stock, flag any SKU with more than 5% variance, categorize by likely cause, and Slack me a summary.' The agent builds that workflow — it runs automatically, not only when you remember to prompt it.

Audit history persists. Every shrinkage run is stored, so you can compare this quarter's variance patterns against last quarter's, track whether corrective actions are actually working, and walk into a retailer or SQF audit with a documented record instead of a one-time export.

Get closed-beta access →

Toolkit