Demand forecasting is the foundation of profitable e-commerce growth. Without it, you're either constantly out of stock or constantly carrying too much inventory. Neither is sustainable.
Start With Historical Sales Data
The foundation of any forecast is historical sales velocity. Calculate your daily, weekly, and monthly sales by SKU for at least the last 12 months. Identify seasonality patterns — when does each SKU peak and trough? This baseline is your starting point for every future forecast.
Layer in Growth Rate
Historical data tells you what happened. Your growth rate tells you what to expect. If you're growing 20% year-over-year, apply that growth rate to your historical baseline to get your forward forecast. But be careful — apply growth rate by SKU, not at the total business level. Products grow at different rates.
Account for Planned Events
Marketing campaigns, promotional events, and seasonality all affect demand in ways your baseline won't capture. Add a planned events layer to your forecast that accounts for any known drivers of demand change — a Q4 promotion, a product launch, a media feature.
Build in Supplier Lead Time
Your forecast doesn't just tell you what you'll sell — it tells you when to order. Work backwards from your forecast using your supplier lead time to determine your order placement date. Your reorder point should trigger when on-hand inventory plus on-order inventory falls to a level that covers your forecast plus safety stock through the lead time period.
The fix: Build a simple 13-week rolling forecast for your top 10 SKUs. Update it weekly. The discipline of maintaining it will improve your forecasting accuracy more than any tool.
Choose the planning level that matches the decision
Forecast by SKU and channel when those differences affect purchasing or allocation. Aggregate forecasts are easier to build but may hide the exact product that will stock out.
Clean the history before forecasting
Flag stockouts, launches, promotions, one-time wholesale orders and discontinued items. Sales history is not the same as unconstrained demand when the product was unavailable.
Combine statistical history with known events
Use recent demand trends as a baseline, then document adjustments for promotions, new distribution, price changes or supplier constraints. Keep the adjustment reason visible for later review.
Measure forecast error where it matters
Track whether errors lead to excess stock, missed sales or unnecessary expedites. Focus improvement on high-value or volatile items rather than chasing perfect accuracy across every SKU.
For inventory planning support, see e-commerce operations consulting.
Use forecast horizons that match lead time
A four-week forecast is not enough for a supplier that needs four months. Review demand at the horizon where purchase decisions must be made, while using shorter views for allocation and exceptions.
Separate baseline demand from event demand
Store promotions, launches and one-time retail orders as explicit events. This makes it easier to reuse the baseline forecast after the event rather than allowing a temporary spike to inflate future replenishment.
Set ownership for forecast changes
Clarify who can adjust baseline demand for promotions, launches or known events and require a reason for material changes. This keeps overrides traceable.
Connect forecast reviews to purchasing dates
Schedule reviews early enough to change purchase orders or freight decisions. A perfect forecast delivered after the supplier cutoff has little operating value.