Demand & inventory forecasting for Shopify and DTC brands

Why is demand forecasting hard for Shopify brands?

Inventory is usually the largest cash commitment a product brand makes, and it is placed weeks or months before the sales happen. Order too little and you stock out during the period ads are working hardest. Order too much and cash sits in a warehouse until it is discounted away.

Most brands plan from last year’s sales plus a growth assumption. That breaks in predictable ways: last year’s sales were themselves capped by stockouts, a promotion inflated one month, a product launched midway through, or demand moved from one variant to another.

Why can’t you just forecast from past sales?

Because recorded sales are not the same as demand. When a SKU is out of stock, sales drop to zero, but customers still wanted it. Forecasting from raw sales history teaches the model that demand fell, so it recommends ordering less, which causes the next stockout.

The first step of every forecast is therefore to rebuild a clean demand history:

1. Join Shopify orders with inventory levels to find out-of-stock periods.
2. Mark those periods as censored rather than as zero demand.
3. Flag promotion days, price changes and launches so the model can separate them from normal demand.
4. Group variants where customers substitute between them, such as sizes of the same product.

Which forecasting methods work for e-commerce?

There is no single best model, so the process is empirical. We start with a simple baseline, such as the same week last year or a moving average, and only keep a more complex method if it beats that baseline when tested on past periods.

Situation	Typical approach
Stable, high-volume SKUs	Seasonal time series models per SKU
Many low-volume SKUs	Forecast at product or category level, then split down
Strong promotions	Models with explicit promotion and price features
New products	Analogues from similar launches, with wider ranges

Every forecast comes with a range, not just a single number. The range is what makes the inventory decision possible.

How does a forecast turn into a purchase order?

A forecast alone does not tell you how much to buy. The inventory policy does. For each SKU, we combine expected demand over the supplier lead time, the uncertainty in that demand, and the service level you choose (how often you accept running out). The result is a reorder point and a suggested order quantity.

That makes the trade-off explicit. You can choose a higher service level for hero products and a lower one for slow movers, and see what each choice means for stock held.

What data do I need to provide?

Less than most brands expect. The core inputs are:

• Shopify order history with line items, ideally two years or more so seasonality is visible
• Inventory levels over time, or at least restock dates, so stockout periods can be identified
• Supplier lead times and minimum order quantities per product or supplier
• A promotion calendar, even a rough one: sale dates, discount levels, major email sends
• Sales from other channels you fulfill from the same stock, such as wholesale or marketplaces

Where inventory history is missing, stockouts can often be inferred from days when a normally selling SKU sold nothing. The diagnostic phase confirms which of these inputs exist and how much they limit the forecast.

Who is this not for?

If your catalog is print-on-demand, dropshipped, or otherwise held by someone else, there is little inventory risk to plan around. Very young stores with a few months of history also lack the data for a reliable model; a simple reorder rule is enough until seasonality shows up in the data.

When stockouts and overstock are both costing you money and the buying plan still lives in a spreadsheet, bringing in Asabisoft for a forecasting diagnostic is a practical next step.