Inventory Forecasting for Online Stores
Before You Start
Effective forecasting requires historical sales data, and the more data you have, the better your predictions will be. A minimum of 6 months of sales history per product gives you enough to calculate meaningful averages. A full 12 months lets you see seasonal patterns. Two or more years of data lets you distinguish true seasonal trends from one-time events. If you are launching a new product with no sales history, you will need to estimate demand using comparable products in your catalog, competitor sales data (tools like Jungle Scout and Helium 10 provide estimates for Amazon products), and pre-launch interest signals like email signups and social media engagement.
Before building forecasts, clean your data to account for periods where your inventory count hit zero. If a product was out of stock for 10 days in March, your March sales data understates actual demand for that product. You need to estimate what those 10 days would have sold based on the daily run rate before the stockout and add that estimate back into your historical data. Without this adjustment, your forecast for future March sales will be too low, creating a self-reinforcing stockout cycle: you under-order because last March was low, which causes another stockout, which makes the next forecast even lower.
Step-by-Step Forecasting Process
Export sales data from your ecommerce platform at the daily or weekly level for every SKU you want to forecast. Shopify exports are available under Analytics, then Reports, then Sales by Product Variant. Amazon provides the Business Reports detail page sales data. Pull the data into a spreadsheet or your inventory management software's forecasting module. For each product, note any periods where stock was at zero and estimate the lost sales during those periods. Also flag any anomalous sales spikes caused by one-time events (a viral social media post, a competitor going out of stock, a pricing error) that should not be included in your baseline demand calculation.
The simplest and most practical baseline is a moving average. A 30-day moving average adds up the last 30 days of sales and divides by 30 to get average daily demand. A 90-day moving average smooths out more noise and is better for products with variable daily sales. For a product that sold 450 units in the last 90 days, the 90-day average daily demand is 5 units per day, or roughly 150 units per month. Moving averages work well for products with steady demand and no strong seasonal pattern. If you have products with trending demand (growing or declining over time), a weighted moving average that gives more importance to recent data captures the direction of the trend better than a simple average.
Compare monthly sales across years to see which months consistently run above or below your annual average. Calculate a seasonal index for each month by dividing that month's average sales by the overall monthly average across all months. If your annual average monthly sales are 300 units and January averages 210 units, January's seasonal index is 0.70 (30% below average). If November averages 540 units, November's seasonal index is 1.80 (80% above average). To create a seasonal forecast, multiply your baseline monthly demand by the seasonal index for each future month. If your current baseline is 350 units per month and November's index is 1.80, your November forecast is 630 units. This method requires at least two years of data to calculate reliable seasonal indices.
If your business has been growing 5% month over month, your forecast should reflect that growth trajectory. Apply the growth rate to your baseline before applying seasonal adjustments. A 350-unit monthly baseline growing at 5% per month becomes 368 units next month, 386 the month after, and so on. Then adjust for specific planned events: a product launch campaign that you expect to double the normal run rate for two weeks, a 20% off promotion that historically increases sales by 40% during the sale period, or a planned price increase that may temporarily reduce demand. These event adjustments are inherently estimated, but even rough adjustments are better than ignoring planned events entirely. Document your assumptions so you can review what happened versus what you predicted and improve future event estimates.
Forecasting is an ongoing practice, not a one-time exercise. At the end of each month, compare your forecast to actual sales for every product you forecasted. Calculate forecast accuracy as a percentage: if you predicted 400 units and sold 380, accuracy is 95%. If you predicted 400 and sold 280, accuracy is 70%, and you need to investigate why. Common reasons for large forecast misses include unexpected competitor actions, supply chain disruptions that caused stockouts, marketing campaigns that performed above or below expectations, and external events like economic shifts or weather patterns. Track accuracy over time and set a target of 80% or higher for your A items (top sellers) and 70% or higher across all products. Identify which products consistently miss and examine whether the forecasting method is appropriate for that product's demand pattern.
Forecasting for Products With Overseas Lead Times
Products sourced from overseas manufacturers typically have 60 to 120 day lead times, which means your purchasing decisions today are based on demand forecasts 2 to 4 months into the future. The further out you forecast, the less accurate the prediction, which creates a fundamental tension: you need to commit to purchase quantities months before the inventory arrives and demand materializes. This is why safety stock becomes more important for products with long lead times, and why many sellers order slightly more than their forecast suggests for critical products where a stockout would be more costly than carrying extra inventory.
For seasonal products ordered from overseas, you typically need to place your holiday season order in July or August to receive goods in October or November. That means your holiday forecast is created 4 to 5 months before the selling season begins, based on last year's performance plus adjustments for growth and any known changes. Our seasonal inventory planning guide covers the specific timeline and calculations for planning seasonal orders with long lead times, and our supplier lead time guide explains how to track and reduce lead time variability.
Forecasting Tools and Software
A spreadsheet is the most accessible forecasting tool and works well for businesses with under 200 SKUs. Build a template with columns for historical monthly sales, moving averages, seasonal indices, growth rates, and event adjustments. The formulas are straightforward arithmetic, and having everything in a spreadsheet gives you full visibility into how each forecast is calculated. The downside is manual data entry and the risk of formula errors as the spreadsheet grows in complexity.
Most dedicated inventory management platforms include built-in forecasting modules. Cin7, Extensiv, and Zoho Inventory all generate demand forecasts based on historical sales data, apply seasonal adjustments, and convert forecasts into suggested purchase order quantities. These automated forecasts are a good starting point, but they should be reviewed and adjusted by someone who understands the business context, because the software cannot know about planned promotions, competitor moves, or supplier changes that will affect future demand.
For businesses with the data and technical capability, more sophisticated forecasting methods like exponential smoothing, ARIMA models, and machine learning-based prediction tools can improve accuracy by 5% to 15% over simple moving averages. However, the improvement depends heavily on data quality and volume. If you have clean sales data for 50+ SKUs over 2+ years, advanced methods can meaningfully outperform spreadsheet-based approaches. For newer businesses with limited data, the sophistication of the model matters less than the quality of the human judgment applied to the forecast.