Shopify Inventory Forecasting: How to Get Started with Demand Prediction

Demand forecasting sounds like enterprise jargon, but it is something every ecommerce founder does intuitively. When you look at your bestselling product and think "I should reorder before we run out," you are forecasting demand. The difference between intuitive forecasting and systematic forecasting is precision and consistency. Your gut feeling might correctly predict that you need more stock, but it cannot tell you whether you need 200 units or 400 units, or whether you should order now or wait three weeks. Systematic forecasting uses your actual sales data to answer these questions with numbers rather than intuition. It does not need to be perfect. Even a basic forecast will usually beat gut feeling on a catalogue of 200+ SKUs, where you simply cannot hold detailed intuitions about every product.

The foundation of any forecast is historical data. Shopify stores this natively, and you can export your order data or sales reports from Shopify admin. For demand forecasting, you need at minimum 90 days of sales data. Ideally, 12 months gives you seasonal patterns and 24 months lets you verify that seasonal patterns are consistent. Export your data at the SKU level (not just product level) including the date of each sale and the quantity sold. If you sell "Navy Tartan Collar" in Small, Medium, and Large, you need sales data for each size separately because they sell at different rates. A common mistake is forecasting at the product level and then splitting the forecast across variants evenly. A Medium might sell 3x more than an XS, equal splitting guarantees you will stockout on Medium while sitting on excess XS.

The simplest demand calculation: take the total units sold of a SKU in the last 90 days and divide by 90 to get the average daily sales rate. Then multiply by 7 to get the weekly rate. For example, if you sold 180 units of "Navy Tartan Medium" in the last 90 days: 180 / 90 = 2 units per day, or 14 units per week.

This is your baseline demand rate. It tells you that at current velocity, you sell roughly 14 of these per week. If you have 112 units in stock, you have 112 / 14 = 8 weeks of cover.

The limitation of this method is that it assumes the future will look like the recent past. If you are approaching a seasonal peak, 90-day averages will understate demand. If you just finished a sale that inflated recent numbers, it will overstate demand. More sophisticated forecasting addresses these issues, but this baseline calculation is where everyone should start.

Beyond the baseline average, look for three types of patterns in your sales data:

Growth or decline trends: Is the SKU selling more each month or less? A product selling 100 units in month 1, 120 in month 2, and 140 in month 3 has a clear upward trend. Forecasting based on the average (120/month) will underestimate future demand. You should project the trend forward: expect ~160 in month 4.

Seasonal patterns: Does demand spike or dip in certain months? This requires at least 12 months of data to identify. Compare the same month across years, if November is consistently 2x your average, that is a seasonal pattern you can plan for.

Promotional effects: Did past sales spikes correspond to promotions, marketing campaigns, or external events? If you ran a 25% off sale in February that doubled sales for that week, those inflated numbers will distort your baseline calculation unless you adjust for them. The easiest adjustment: remove promotional periods from your baseline calculation and treat promotional demand as a separate, planned uplift.

This is where many brands get forecasting fatally wrong. Your forecast is not just "how many will I sell?", it is "how many will I sell between now and when my next order arrives?" If your lead time is 14 days (domestic UK supplier), you need to forecast 2 weeks of demand. If your lead time is 190 days (China sourcing with sea freight), you need to forecast more than 6 months of demand.

The formula: Reorder Quantity = (Average Daily Sales x Lead Time in Days) + Safety Stock - Current Stock on Hand. To find the stock level that should trigger that order, use a reorder point calculator.

For the "Navy Tartan Medium" example selling 2 units per day with a 190-day lead time: (2 x 190) + safety stock = 380 + safety stock. With 4 weeks of safety stock (56 units), the reorder quantity is 436 units minus whatever they currently have on hand.

The longer your lead time, the more critical your forecast accuracy becomes. A 10% forecast error on a 14-day lead time means you are off by 1.4 days of stock. The same 10% error on a 190-day lead time means you are off by 19 days of stock, nearly 3 weeks.

For a brand importing from China, the forecasting challenge is extreme: predicting demand 190 days ahead for thousands of SKUs. At that horizon, no forecast is highly accurate, but even a moderately accurate forecast beats guesswork. A practical approach combines three inputs:

First, the baseline: a trailing 90-day average sales rate per SKU, adjusted for known seasonal patterns from previous years.

Second, the growth factor: your year-over-year growth rate applied to the baseline, discounted to avoid over-ordering on an optimistic projection.

Third, the pipeline: stock already on order but not yet received. If you placed a PO 60 days ago that has not arrived yet, those units are coming and need to be subtracted from the reorder calculation.

This three-input model gives a reorder suggestion for every SKU. It will not be perfect, but it prevents the worst stockouts and overstock that come from ordering on gut feeling. The SKUs where it goes wrong are typically new products (no history to forecast from) or products hit by unpredictable viral social media exposure.

Spreadsheet-based forecasting works well up to about 200-300 SKUs. Beyond that, the maintenance burden becomes unsustainable, you are spending hours updating formulas, adjusting for promotions, and manually checking results. Dedicated tools automate the process by analysing your sales data and generating reorder suggestions as new sales come in. Some use statistical or AI models to detect seasonal patterns and trends that are hard to spot by hand.

When evaluating any forecasting or reorder tool, look for: explainability (can you see why the system is suggesting a particular quantity?), override capability (can you adjust for information the tool does not have, like a planned promotion?), and accuracy tracking. Canopy Inventory does not use AI. It works out weeks of cover from your recent Shopify sales over a window you choose (7 to 90 days), and its suggested buy quantity comes with a confidence badge that explains itself.