Shopify inventory forecasting: your stockouts are a lead-time problem, not a demand problem.

Shopify's inventory reports show stock on hand, sell-through, and a days-remaining estimate. The estimate is a straight-line average of recent sales. It doesn't know your supplier's lead time, because Shopify has never seen a purchase order. It looks at one store at a time, and it counts products, not the size and color that's actually about to run out.

So the report says "18 days remaining." Your supplier takes 35. You were already late when you read it.

Here's the part nobody wants to hear: your demand guess is probably fine. Take a 7-day, a 28-day, and a 90-day sales rate per variant and you'll land within a reasonable margin most of the time. Seasonality and promos move it, but they're visible and you know they're coming.

Lead time is the other half, and it's the half that's never current. It lives in a spreadsheet cell someone typed in January. The supplier has slipped twice since. Nobody updated the cell, because updating the cell isn't anyone's job.

A forecast with a stale lead time isn't wrong by a little. It's wrong by exactly the number of days the supplier slipped.

Off-the-shelf forecasting apps are demand engines. They pull your Shopify sales history, run a model over it, and give you a replenishment suggestion. The demand math is decent. Then they ask you to type in a lead time per supplier, once, and trust it forever.

Three more problems stack on top. Each store is forecast as its own island, so a variant selling in three stores gets three forecasts that don't know about each other. The logic is a black box, so when the number looks wrong you can't see why. And the fee runs monthly for a forecast you don't own and can't change.

They solved the half you already had.

This is where AI earns its place, and it isn't in the forecast math. The math stays simple and readable. AI's job is the inputs and the actions, the parts a human was supposed to do and didn't.

A Tuesday morning, before anyone logs in:

• The supplier's shipping confirmation landed overnight. The agent reads it: ship date is 9 days later than the PO said. Lead time for that supplier gets updated. Not the sheet. The layer.
• Every reorder point tied to that supplier recalculates. Three variants move from "fine" to "order this week."
• One variant was already past the line. The agent flags it: this one stocks out on the 28th regardless. Here are the two stores it affects and the units left in each.
• It drafts the purchase order for the three that can still be saved, and a note to the supplier asking for a partial ship on the fourth.
• Someone approves it over coffee. Nothing was typed into a cell.

That's the whole system. The forecast never got smarter. The inputs stopped being stale.

• Inventory and order webhooks from every store into one layer, so a variant is one row, not three. Shopify Plus expansion stores included.
• Lead time as a living number per supplier, updated from emails and ship confirmations, not typed once.
• Reorder points and safety stock calculated per variant per store, in logic you can read.
• Actions that draft and wait for approval: purchase orders, supplier notes, low-stock alerts, pausing ads on a variant that's about to sell out.

We build this layer for multi-store e-commerce brands running tens of thousands of variants. It's the same inventory layer described on our Shopify inventory automation page. Forecasting is what it does first; the rest of the actions hang off it.