ChatGPT prompts for inventory management: safety stock, reorder points and excess, worked through

Fictional exampleExample Motors Co., its suppliers and every number in this example are made up for illustration. Don't paste real supplier quotes, NDA pricing, personal data or export-controlled data into an AI tool. Anonymize first and follow your company's AI policy.

ChatGPT is useful for inventory work when you treat it as an analyst who explains its math, not as your planning system. Below are five prompts for jobs planners and material managers do every week, with one fully worked safety stock and reorder point example you can check line by line.

What should you export from your ERP?

For each SKU: 12 months of usage (daily or weekly), lead-time history from receipts (PO date to receipt date), on-hand, on-order, backorders and unit cost. Replace part numbers and descriptions with codes like "SKU A" before you paste, and use your company's approved AI tool.

Prompt 1: Safety stock and reorder point

You are an inventory planner. For each SKU below, calculate safety stock with SS = z × sqrt(LT × σd² + D² × σLT²) at [SERVICE LEVEL] and the reorder point ROP = D × LT + SS. Show each input and each step. Round safety stock up. Then compare inventory position (on-hand + on-order − backorders) to the ROP and say 'order' or 'no order'. List any SKU with missing or suspicious data (zero lead time, negative demand) instead of guessing.

[PASTE ANONYMIZED SKU DATA: SKU, D, σd, LT, σLT, on-hand, on-order, backorders]

Worked example

Example Motors Co. stocks "Bearing 6204-2RS", a generic bearing size used here for illustration.

The formula combines demand variability and lead-time variability:

SS  = z × √(LT × σd² + D² × σLT²)
    = 1.645 × √(9 × 144 + 1,600 × 4)
    = 1.645 × √(1,296 + 6,400)
    = 1.645 × √7,696
    = 1.645 × 87.73 ≈ 144.3 → round up to 145 units

ROP = D × LT + SS = 360 + 145 = 505 units

What good output looks like

The insight: with the demand-only formula (z × σd × √LT = 1.645 × 12 × 3 ≈ 59.2, rounded up to 60) you'd carry less than half the buffer. In this example the lead-time swing drives safety stock, so working with the supplier on lead-time consistency could matter more than tweaking the forecast.

What to check: recompute one SKU by hand or in Excel (z for a 95% service level is =NORM.S.INV(0.95), about 1.645), and confirm that D, σd and LT use the same time unit, all days or all weeks.

Know the limits: this formula assumes demand is roughly normal and lead times vary independently of demand. Intermittent or lumpy demand, such as spare parts that sell a few times a year, needs a different method.

Prompt 2: Which SKUs need ordering now

For each SKU below, calculate inventory position = on-hand + on-order − backorders and compare it to the reorder point I've given. Output a table: SKU, inventory position, ROP, verdict (ORDER / OK), and days of supply at average demand. Sort ORDER rows first, lowest days of supply at the top. Flag any SKU with a missing ROP or zero demand instead of guessing.

[PASTE ANONYMIZED DATA]

What to check: backorders are subtracted, not added, and the on-order quantities are open POs that will actually arrive in time.

Prompt 3: ABC/XYZ segmentation and a policy per cell

Segment the SKUs below. ABC by annual usage value (A = top 80% of value, B = next 15%, C = last 5%). XYZ by coefficient of variation of monthly demand (X < 0.5, Y 0.5–1.0, Z > 1.0). Show the thresholds you used. Output the 9-cell matrix with SKU counts, then suggest a stocking policy for each cell (service level target, review frequency, stock or make to order). Mark suggestions as starting points.

[PASTE ANONYMIZED 12-MONTH USAGE AND UNIT COST]

What to check: the cut-offs are yours to set. Confirm the AI applied them as stated and didn't adjust them silently.

Prompt 4: Excess and obsolete action list

From the data below, list SKUs with more than [N] months of supply or no usage in [M] months. For each: on-hand value, months of supply, last usage date, and a suggested action (stop reorders, return to supplier, transfer, rework, sell, scrap) with a one-line reason. Total the value by action. Do not recommend scrapping anything with open demand or open orders.

[PASTE ANONYMIZED DATA]

What to check: service parts and items with long-tail demand. Write-offs and scrap decisions follow your company's approval process.

Prompt 5: Cycle-count plan

Design a cycle-count plan for [N] SKUs using the ABC classes below: count A items [monthly], B [quarterly], C [twice a year]. Spread the counts across [WORKING DAYS] so each day has a similar workload. Output counts per day, which classes are counted each week, and a tolerance rule for when to recount. Ask me for anything you need.

What to check: the daily workload fits your team, and high-value or high-error locations are counted more often.

How do you check ChatGPT's inventory math?

• Recompute in Excel. Put the AI's inputs into a sheet with the formula and compare.
• Use the data analysis tool if your version has one, so the math runs as code you can inspect instead of being estimated in text.
• Spot-check three SKUs by hand, including the largest and one with odd data.

This isn't excess caution. Researchers have documented that ChatGPT's accuracy on spreadsheet problems breaks down when information is limited or the problem is too complex, and a 2023 paper on ChatGPT and Excel is titled, fittingly, "trust, but verify".

For real company examples, see the inventory optimization and demand forecasting use cases, and Brief coverage tagged #forecasting. Buyers can use ChatGPT prompts for procurement.