AI Inventory Forecasting for One Shopify SKU
Calculate and verify a defensible reorder point for one Shopify SKU using sales, lead time, sellable stock, and supplier-delay risk.
For one Shopify variant, AI inventory forecasting should show whether usable stock has reached a defensible reorder point using records you can check. A precise-looking number isn't enough.
I've run two Shopify stores, and "reorder when it looks low" fails once sales outrun new stock. This workflow replaces that guess with a reorder point you can calculate this week.
Key takeaways
What is AI inventory forecasting?
AI inventory forecasting learns from a SKU's past sales and predicts demand. Stock tools then add supplier timing and available units to suggest a reorder. Demand prediction and reorder math are separate jobs. If any source record is wrong, the reorder point will be wrong.
Use this workflow when you hold stock yourself, use a warehouse, or have a supplier reserve units for you. When the supplier owns every unit, estimate demand first and then apply its available stock in a demand-before-inventory workflow.
This ownership check separates stock planning from the wider field of AI dropshipping. It tells you whether you can place the next order or must ask the supplier to replenish.
Build the reorder workflow, input by input
Use clean sales history, a lead-time range, and a checked stock count to calculate the reorder point.
Use the four parts in this order,
- Sales history: Measure demand for the exact variant.
- Lead-time range: Record normal and late supplier delivery times.
- Usable stock: Count available and incoming units.
- Reorder point: Add expected lead-time demand and safety stock.
Work through them in sequence. If any of the three source records is wrong, the reorder point will be wrong too.
1. Clean per-variant sales history
Use sales from the exact variant you plan to reorder, and exclude periods when it couldn't sell. Export daily variant sales, then label days affected by stockouts, promotions, or bulk orders. A zero during a stockout isn't zero demand, while a promotion or bulk order may not represent the next normal week.
I'd start a fast-selling SKU with eight clean weeks of sales because that captures several weekly cycles. That's my review starting point rather than a universal minimum. Keep each excluded day visible so another person can reproduce the choice.
Demand research used for finding winning products can guide a first buy. The SKU's own history becomes the forecasting input once sales exist.
2. Use an honest supplier lead-time range
Record the supplier's normal and late delivery times because the gap sets your delay buffer. A single "usually ten days" estimate hides the risk that safety stock should cover. Ask for recent dates from order to receipt, then use the average for expected demand and the slower end for the buffer.
Safety stock covers supply delays while new stock is in transit. Build the range from recent receipts rather than a supplier's best-case promise.
That delivery evidence also belongs in your process for choosing a supplier. A vendor who shares only one estimate keeps the reorder in Verify.
3. Reconcile current and in-transit stock
Use the latest counted stock. Subtract open-order units, then add only shipments the supplier has confirmed. A purchase order doesn't count as incoming until the supplier confirms the shipment.
Compare that result with Shopify and the open-order report. When the systems disagree, use a new count for stock on hand and the order report for commitments. Use the supplier's shipment confirmation for incoming units. Resolve the difference before using the reorder trigger.
4. Calculate a lead-time-aware reorder point
Your reorder point is expected demand during average lead time plus safety stock for supplier delay. The average covers a normal delivery, while the buffer covers the extra days between average and late delivery.
Suppose one variant sells six units a day and the supplier averages ten days, while late deliveries take fourteen. Expected lead-time demand is 60 units. A simple buffer adds 24 units for the four-day gap, bringing the reorder point to 84 units.
Compare usable stock with 84. Wait at 90 units, and trigger the reorder at 84 or below.
The threshold tells you when to review an order. It doesn't set the order quantity.
This buffer covers late supply only. If sales don't stay near six units a day, add a separate demand-variation method before automating the trigger.
When to trust, verify, or override the forecast
Act, Verify, or Override tells you what to do with a forecast before money moves. Together, those states form the Confidence-Weighted Reorder Line.
Use the three positions in this order,
- Act: All three inputs are current and complete.
- Verify: One input is thin but still useful.
- Override: Act by hand when key data is missing.
Move a SKU only when the missing record has been replaced with checked evidence.
1. Act on a confirmed fast mover
Act when checked sales and a measured lead-time range support the reorder point. Checked usable stock tells you whether the SKU has reached it. A fast mover with complete records can use an automated trigger.
Review the order quantity separately when it would tie up more cash than the SKU's normal cycle.
2. Verify a seasonal SKU
Verify when the trend looks sound while the history leaves a pattern uncertain. One prior season shows when demand rose. Keep the SKU in Verify until a second comparable season exists. A person may also approve the order when current sales closely follow the earlier curve.
3. Override a new SKU
Use Override for a new SKU with three weeks of sales and no known delivery range. Those unknowns can still produce a neat number even though the proof is thin. Divide available buying cash by landed unit cost, then round down to get the cash cap in units. Compare it with storage capacity and use the lower cap. More sales and one completed delivery can move the SKU to Verify.
What a wrong reorder call actually costs
A wrong reorder costs lost gross profit from a stockout or carrying costs and markdowns from overstock. A replaceable, low-margin item may justify less safety stock than a high-margin hero product. Extra stock ties up cash and may need storage or a markdown.
The market figures provide context,
- IHL Group's 2026 study puts global inventory distortion at $1.7 trillion, or 6.2% of retail sales, across more than 4,500 large retail and hospitality companies.
- A companion analysis assigns 65.6% of that total to out-of-stocks.
Those figures show the scale of the problem. They don't predict one Shopify store's losses.
Estimate stockout gross profit at risk as average units sold per day multiplied by late days and gross profit per unit. Compare that result with the cash tied up in the added units plus known storage costs. Use profit margin rather than revenue when you size the risk. A high price can look urgent even when product and shipping costs take most of it.
What replaced Shopify's native forecasting tool
Shopify replaced Stocky with its own stock tools and Sidekick demand planning.
A third-party app may fit when you need visible reorder math, custom alerts, multi-site controls, or a different order workflow. Use one only when a missing control affects a real order.
Shopify's Stocky transition guide says Stocky shut down after 31 August 2026, while users can still read and export data for at least 90 days. Past purchase orders and stocktakes must be exported or rebuilt because they don't move on their own.
The same migration guide says Sidekick reads sales data, suggests when to restock, and can draft a purchase order in Shopify admin.
The current choices differ in fit, visible inputs, and limitations:
I'd begin a small catalog with Shopify's own tools. Switch when hidden math blocks review or multi-location stock can't be controlled. A late alert or blocked purchase order also gives you a measurable reason to switch. That keeps automated dropshipping tied to a known task instead of adding another dashboard first.
What AI actually adds to a demand forecast
Keep AI only when it predicts held-back sales data better than a simple moving average. A model can learn patterns that a flat average smooths away, including weekly cycles, promotion effects, and slow seasonal changes. Stable demand leaves less room to improve.
Test the gain on your own SKU. Build a four-week baseline by using mean units from the prior four full weeks to predict next week. Absolute unit error is the gap between predicted and sold units. Give AI and the baseline the same history.
I'd test one steady SKU over several held-back weeks before paying for a wider rollout. Keep AI when its error stays lower and you can inspect its inputs. Otherwise, keep the baseline. A lower-error forecast still needs a current stock count and real lead-time range.
FAQ
Does Shopify include inventory forecasting?
Shopify's Sidekick can use sales data to suggest restocks and draft purchase orders. Sellers who need clear formulas or custom controls may still prefer an inventory app.
How much sales history does a forecast need?
The needed history depends on the pattern you want to learn. Several clean weekly cycles can support a basic fast-mover review, while seasonal demand needs more than one comparable season.
What if my supplier gives one lead-time estimate?
Keep it in Verify until two delivered orders show a range. Don't automate the delay buffer before then.
Can AI forecast demand for a new product?
Use similar-product or category data for the first estimate. Wait for the new SKU's own sales before automating reorders.
What if I reorder based on gut feel?
You may order too late after a sales spike or too early after a brief slowdown. A recorded reorder point shows your assumption, so you can fix it after the next cycle.
