Demand forecasting is the process of predicting how many units of each SKU will sell over a future period. It's the input that drives every other inventory decision: how much to order, when to order, how much safety stock to hold, and how to allocate working capital across the catalog.

What demand forecasting tries to answer

The forecast translates business intent into operational targets:

How many units of SKU X will sell in the next 30 / 60 / 90 days?
How will that change for hero SKUs versus long-tail SKUs?
What seasonal lift do we expect at peaks (Black Friday, gifting season, summer)?
What does the upcoming product launch cannibalize from existing SKUs?

Without a forecast, replenishment defaults to either reactive ordering (always running short) or cash-driven bulk ordering (always overstocked). Forecasting puts a number on expectations so POs match anticipated demand rather than gut feel.

Common demand forecasting methods

Naive / moving average: the next period equals the last N periods averaged. Simple, useful for stable SKUs, breaks down on seasonal or trend-driven products.
Exponential smoothing: weights recent periods more heavily than older ones. Better at adapting to trend changes than a flat moving average.
Time series with seasonality (e.g., Holt-Winters, ARIMA): separates trend, seasonality, and noise. Standard approach for SKUs with predictable seasonal patterns.
Regression / ML-based: incorporates external signals — promotions, paid spend, weather, holidays — alongside historical sales. The most accurate when those signals are real drivers.
Bottom-up by channel: separate forecasts for Shopify, Amazon, wholesale, and retail, then aggregated. Avoids the false precision of a single blended forecast across channels with different dynamics.

Why demand forecasting matters

The cost of a bad forecast is asymmetric. Forecast too low: stockouts during peak, lost revenue, wasted ad spend on out-of-stock SKUs, customers buying competitors' products. Forecast too high: dead stock, working capital tied up, storage fees, eventual markdowns that compress margin. Most brands underestimate the second cost because it shows up later and feels less urgent — but cumulatively, overstock is often the bigger drain.

Common forecasting pitfalls

One-number forecast for all SKUs: hero SKUs deserve detailed bottom-up forecasts; long-tail SKUs can use simpler methods. Treating both the same overcomplicates one and undercommits to the other.
Ignoring promotional pull-forward: a Black Friday lift isn't free demand — it's often demand that would have happened in December anyway. Forecasts that don't account for pull-forward overpredict the post-promotion period.
Forecasting in revenue rather than units: revenue forecasts hide unit volume. Inventory decisions need unit forecasts.
No forecast accuracy tracking: brands that don't measure forecast error month-over-month can't tell whether their methodology is improving or whether systematic biases (always too high, always too low) are present.

Forecast accuracy tracking, the piece most brands skip, depends on having clean historical sales and inventory data in one place rather than scattered across spreadsheets and channel exports — the kind of consolidated reporting that data and analytics infrastructure is built to support. Once a forecast is trusted, it still has to turn into actual purchase orders and replenishment triggers, which is where forecasting connects to the ERP or inventory system executing on it rather than staying a spreadsheet exercise.

Once a forecast exists, it still has to turn into actual reorder points and safety stock levels inside the store’s inventory system — the operational step this walkthrough of setting up inventory forecasting on Shopify works through.