Ops & Supply

How to set reorder points and safety stock with AI

Ops & Supply3 AI tools7 steps6 friction points

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

Reorder points and safety stock are the two numbers that determine whether you run out of product or drown in it. Your reorder point tells you when to place the next purchase order; your safety stock is the buffer you hold in case demand spikes or your supplier runs late. Get them wrong in either direction and you're either eating carrying costs or watching a retailer go out-of-stock while you wait on a co-packer.

The math is not complicated — it involves lead time, average daily demand, and some variability factor — but the inputs are a mess. They live in spreadsheets, your 3PL portal, a Shopify dashboard, and whatever your co-packer last emailed you. AI feels like the right tool because the calculation logic is well-documented and an LLM can explain it clearly, run the numbers once you supply the inputs, and help you think through edge cases like promotional periods or seasonal demand.

ChatGPT, Claude, and Gemini can genuinely help here. They know the standard formulas cold. Paste in a table of SKU-level data — units sold, lead time in days, stockout history — and a good model will return reorder points and safety stock levels with reasonable assumptions explained. The output is often better than a back-of-napkin estimate and faster than rebuilding the spreadsheet logic from scratch.

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 sales data from Shopify, your ERP, or wherever orders live — typically a CSV with SKU, units sold by week or month, and date range. The cleaner the export, the better the output.

2 Pull your current lead times from your co-packer or 3PL. This usually means checking an email, a portal, or a spreadsheet you maintain manually. Combine lead times with your sales CSV into a single table.

3 Open ChatGPT or Claude and paste the table directly into the chat. Ask it to calculate average daily demand, demand variability (standard deviation), reorder point, and safety stock for each SKU using a specified service level — 95% is a reasonable default for most CPG operators.

4 Review the model's assumptions. Good LLMs will state what formula they used (typically: Safety Stock = Z × σ_demand × √lead_time; Reorder Point = average demand × lead time + safety stock). Verify these match your actual situation — if your lead time varies, ask the model to account for lead time variability as well.

5 Paste back any corrections — updated lead times, SKUs to exclude, a different service level for slow-moving product — and ask it to recalculate. The model holds context within the session, so iteration is fast.

6 Copy the output table into your spreadsheet or inventory system manually. There is no direct write-back from the LLM to your 3PL or ERP.

7 Flag the numbers that look off and ask the model to explain its reasoning for specific SKUs. This is often where you catch data errors — a missing week of sales, a lead time that's seasonally different, a SKU that launched mid-period skewing the average.

Prompts you can copy

Here is a table of weekly sales by SKU for the last 12 months and lead times in days for each SKU. Calculate the reorder point and safety stock for each SKU assuming a 95% service level. Show your formula and assumptions.

My co-packer lead time varies between 14 and 21 days depending on the season. Recalculate safety stock for SKU-103 accounting for lead time variability, not just demand variability. Use a 95% service level.

We have a promotional event in 6 weeks that historically increases velocity by 40% for 3 weeks. How should I adjust the reorder point and safety stock for that period versus the rest of the year?

Which of these SKUs has the most days of supply risk right now, given current on-hand quantities in column D and the reorder points you just calculated? Rank them and flag anything under 14 days of cover.

Our average demand for SKU-207 is 80 units/day, standard deviation is 15 units/day, and average lead time is 18 days. Calculate safety stock at 90%, 95%, and 99% service levels so I can see the tradeoff in units held.

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 data connection — every run requires you to manually export from Shopify, your 3PL, or your ERP, then paste it in. Stale exports produce stale reorder points.

Nothing persists between sessions. The carefully iterated calculation from last month is gone; next quarter you start over with a blank prompt and hope you remember which assumptions you used.

Context windows cap out at a few hundred rows for complex tables. Large SKU catalogs get truncated or require splitting into multiple sessions, making cross-SKU comparisons unreliable.

Outputs vary between runs even with the same inputs. The model may pick a slightly different formula variant or round differently, so your numbers shift without a clear reason — hard to explain to a buyer or a co-packer.

The model can't write reorder points back to your inventory system. Every number it produces has to be manually copied into whatever tool your team actually uses to trigger purchase orders.

Promotional lifts, seasonal patterns, and co-packer constraints require you to re-describe context every single session. The model has no memory of your Q4 Whole Foods spike or your 21-day holiday lead time extension.

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 — it builds and runs persistent apps connected to your live business data. For reorder points and safety stock, that means an agent builds the calculation engine once, against your actual inventory and sales numbers, and keeps it current as your data changes.

The Inventory Planner app — currently in development, request beta access — is purpose-built for this workflow: reorder automation based on live velocity and lead time, with safety stock calculation built in, across every location your product sits. No more copy-pasting from your 3PL portal.

The Demand Planner app — also in development, beta access available — handles the demand-side inputs automatically: it reads your actual sales signals, factors in promotional lifts and seasonal patterns, and produces the demand forecast that feeds your safety stock calculation, instead of asking you to remember that Whole Foods always spikes in Q4.

Connect Shopify or your ERP from Starch's integration catalog — the agent queries live sales data when your reorder calculation runs, so the numbers reflect this week's velocity, not last month's export.

Describe the model you want in plain English — 'build me a dashboard showing reorder points and safety stock for every active SKU, updated weekly, using a 95% service level and the lead times in this table' — and Starch builds it and keeps it refreshed, instead of requiring you to re-run the prompt chain from scratch each time.

Set up an automation that flags SKUs falling below their reorder point and sends a summary to Slack or email — triggered on a schedule, running against live inventory data, without a manual step in the workflow.

Get closed-beta access →

Toolkit