Article · · 6 min read
ChatGPT prompts for warehouse managers — 12 that actually work on WMS exports
You've been told to use AI. Nobody said which prompt, on which export, or what to do when it makes a number up.
Here are twelve. Each takes an export you already pull, or notes you already type, and hands back one page. Every number in this article comes from a fake warehouse, not a real one.
The one rule
Your company's AI and data policy wins. If it says no, the answer is no. If it says nothing, that is not a yes. Ask.
Two questions for IT: which tool is approved, and does an anonymized operational export fall inside the data-loss rules? That means IDs, timestamps, quantities, and costs. No names. Write the answers down before you paste anything. The ten rules cover the rest.
Five habits that keep you out of trouble
- Never names, wages, or confidential files. Delete the column before upload. IDs and hours only.
- The tool only reads exports. It never touches the WMS. It never sends anything.
- Gaps stay blank. If a number cannot be computed from the file, it stays blank with a reason. No estimates.
- Verify one number by hand, every time. Thirty seconds in Excel.
- Everything it writes is a draft. You sign it.
Attach the export or paste the table. Paste the prompt. Read the page. If your tool can run code, add one line: "Compute using code and show every formula in a footnote." Arithmetic done in prose is where wrong totals come from.
Every morning: two prompts in full
1. The Morning Read. Input: the shipments export and the labor-hours export, last two weeks, IDs only.
Prompt 1 · paste as writtenUsing only the two tables I paste below (shipments; labor hours), build a one-page morning brief for the latest ship date in the data — never today's date. Compare each metric to the same weekday one week earlier. Metrics: orders shipped; median hours from order to ship (clock hours); % of orders shipped within 24 hours; ship cost per order; hours worked; OT % overall and by department (top 3). Show formulas in footnotes. If a metric cannot be computed from these tables, leave it blank and list it under "Gaps" with the reason — never estimate. Do not mention or rank any individual; use IDs only. End with up to five questions for the floor, each tagged [FLOOR CHECK].
2. Backlog and aging read. Input: the open-orders export. Decide what "aged" means in your building first.
Prompt 2 · paste as writtenFrom the pasted open-orders table only, using the latest timestamp as the report day, show total open orders and units, the past-due count and share, an age-bucket table (0–1, 2–3, 4–7, 8+ days from release), a by-type table with aged and past-due counts, and the ten oldest orders. Aged = over 3 days. Formulas in footnotes; anything not computable goes under "Gaps." No names.
Every week: two more in full
5. OT drift. Input: eight to thirteen weeks of labor hours, IDs only.
Prompt 5 · paste as writtenFrom the pasted labor table, compute OT % (overtime ÷ total hours) by week and by month, OT % and share of overtime by function, and whether OT is rising or falling (first four weeks vs. last four). Report the share of overtime carried by the top 20% of employee IDs as a percentage only — do not list IDs. Formulas in footnotes.
6. Adjustment reason-code Pareto. Input: the inventory-adjustments export, plus units shipped by month if you have it.
Prompt 6 · paste as writtenFrom the pasted inventory-adjustments table, compute total adjusted units (absolute value) by month and by reason code, the share of each reason code, and the top ten SKUs or locations by adjusted units. If I paste units shipped for the same months, compute the adjustment rate (adjusted units ÷ units shipped). Say what changed between months in one sentence, marked [UNVERIFIED]. No names; user IDs only, never ranked.
Notice what the four share. One named file. The report day comes from the data. What to leave blank is spelled out. No names. That is what stops the tool from inventing.
The other eight
| # | Prompt | Input file | What you get back | The trap |
|---|---|---|---|---|
| 3 | Carrier-cutoff risk | Open orders, plus your cutoff times | At-risk and missed orders by carrier and service, by minutes left | It doesn't know your cutoffs. Give them or it guesses. |
| 4 | Labor plan vs. actual | Labor hours, plus your planned hours by function | Variance to plan, OT % by function, units per hour where you gave units | No units column, no units per hour. If one shows up, it was invented. |
| 7 | Weekly review draft | This week's and last week's figures | Three-line summary, KPI table, three largest changes with formulas | Anything you didn't paste should read [GAP]. A filled blank is invented. |
| 8 | Five-why from incident notes | Your notes, not the WMS | Timeline, five-why chain, inferences tagged [UNVERIFIED], two facts to confirm | If the output reads like a write-up of a person, the input was. Start over. |
| 9 | Shift handoff | Whatever you typed on shift | Done / carrying over / watch, plus a carry-over table with owners blank | It adds things you didn't write. The prompt forbids it; check anyway. |
| 10 | SOP from a walkthrough | Notes or a transcript from walking the process once | One-page SOP, numbered steps, two common errors, a "done when" line | An unclear step gets a guess unless you demand [FLOOR CHECK]. |
| 11 | Email in your voice | Three emails you sent, plus the facts for this one | A draft under 150 words, one ask, no filler | A fact you didn't give it is a fact it made up. You send it, never the tool. |
| 12 | Vendor demo questions | The vendor's pitch | Ten claims, one checkable question each, five things the pitch left out | Never ask it whether to buy. It has no opinion worth having. |
What prompt #1 returns on the practice warehouse
The free files are one month of Copperline Fulfillment Co., a fictional 3PL generated by script: 17,467 shipments and 890 person-days of labor, IDs only. Run prompt #1 and a working tool returns this.
Report day: Tuesday 2026-03-31, the latest ship timestamp in the file. Not today. Not the filename. Same weekday: Tuesday 2026-03-24.
| Metric | Report day | Same weekday | Delta |
|---|---|---|---|
| Orders shipped | 907 | 976 | −69 |
| Median click-to-ship (clock hrs) | 25.4 | 25.2 | +0.2 |
| Shipped within 24 clock hrs | 46.3% | 45.8% | +0.5 pt |
| Ship cost per order | $19.10 | $18.62 | +$0.48 |
| Hours worked | 385 | 409 | −24 |
| OT % (building) | 12.8% | 17.7% | −4.9 pt |
| OT % by function (top 3) | PACK 27.5 · PICK 25.6 | PUTAWAY_REPLEN 39.3 · PACK 35.0 · RECEIVE 24.5 |
The two blanks
Read the table like you are at the huddle. Fewer orders than last Tuesday, and PACK still ran 27.5% overtime. Cost per order up 48 cents on a similar service mix. Click-to-ship flat at 25.4 hours. Stable, not good.
And two blanks, both correct: units per labor hour (no units column) and anything about inventory (no adjustments file). If your tool gave you a units-per-hour number, it invented one. If it compared Tuesday to Monday, it ignored the weekday rule. Fix the prompt, not the tone.
Never
- Never names, wages, or confidential files. Delete the column first.
- Never let the tool pick the report day. The latest timestamp in the file, every time.
- Never accept a number without its formula. If it won't show one, it doesn't have one.
- Never let it rank, name, or single out a person. The question itself is the problem.
- Never fill a gap. A blank you can trace beats a number you can't.
The close
Twelve prompts, one habit. Run them on the fake building first. Match the answer key. Then run them on yours. The prompt that gives you 907 orders and two honest blanks on the practice files is the one that is safe on your export.
The free PDF below has all twelve, with the two practice files. The ten-minute daily brief walks prompt #1 end to end, and if your only tool is a paste box, start with the Copilot workflow.
Your company's AI and data policy wins. Never paste names, wages, or confidential files.
— Floor Read
Sources: morning-read/shipments.csv, morning-read/labor.csv — Copperline practice warehouse, fictional.
Written by a sitting director of operations at a large distribution center — eight years on fulfillment floors, from frontline supervisor to running a multi-hundred-person building's P&L. No software vendor pays for a mention here. The author keeps the day job and the brand separate, which is why there's no headshot; the sample chapter is free so you can judge the work instead.