Pre-season stock planning is one of the decisions that most affects profit margin in e-commerce, yet is managed the least systematically. Entering a season understocked means lost opportunity, while entering it overstocked means discount pressure and warehouse cost. Demand forecasting is the discipline that balances these two risks and turns intuition into a data-driven process.
Why demand forecasting is a necessity, not a "nice to have"
For seasonal products (summer wear, winter wear, holidays, back-to-school, and the like), lead time typically begins weeks, sometimes months, in advance. This means that reacting to sales data as you see it is often too late: noticing mid-season that "this product is out of stock" means missing the rest of the season when you try to restock.
This is why demand forecasting is a proactive, not reactive, process: analyzing historical data and turning it into next season's supply decisions must be completed weeks before the season starts, not once it has already begun.
Extracting seasonality from historical data
The foundation of an accurate forecast is at least the last two to three seasons of sales data. A forecast built on a single season's data risks mistaking that season's random fluctuations — an unexpected weather event or a competitor's stock shortage, for instance — for a general rule.
When examining historical data, separate it into three layers: base demand (independent of season, the product's normal sales pace), seasonal uplift (a recurring, predictable rise in certain months), and one-off effects (a campaign or external event specific to last season). Only the first two layers should carry forward into next season's planning; one-off effects should be filtered out. For example, if a sudden spike in demand for a particular product last year was driven by an influencer post, treating that spike as a baseline this year would be misleading.
Segment products by sales velocity
Applying the same forecasting method to every product is inefficient. Using ABC analysis logic, divide your products into three groups:
- Group A (high-volume, critical products): A small number of products that make up the bulk of revenue. For this group, it's worth doing detailed, weekly historical data analysis.
- Group B (mid-volume products): A simpler, monthly comparison against last season's data is enough to produce an adequate forecast.
- Group C (low-volume, long-tail products): Detailed analysis on these products is a waste of time; simple rules (for example, a fixed percentage above last season's average sales) are sufficient.
This segmentation lets you focus your limited planning time on the products that will have the most impact. The comparison below can be a starting point for deciding which method to apply to which product group:
| Method | Accuracy | Data Needed | When to Use |
|---|---|---|---|
| Simple average / percentage increase | Low | Single season of data | Group C, long-tail products |
| Moving average | Medium | 2-3 seasons of monthly data | Group B, mid-volume products |
| Seasonal index | Medium-high | At least 2-3 seasons, weekly breakdown | Group A with clear seasonality |
| Trend + regression analysis | High | Multiple seasons, plus external factor data | Critical, high-volume Group A products |
"Demand forecasting isn't about producing a perfect number — it's about reducing uncertainty to a manageable range."
Accounting for external factors
Historical sales data alone isn't enough; external factors that will affect the season must also be built into the forecast. Chief among these are your planned campaign calendar (your own promotions as well as major marketplace campaign days), changes in the holiday calendar, price changes relative to the previous season, and newly opened sales channels.
For example, if a new marketplace channel that wasn't active last season is live this season, a forecast based purely on historical data won't reflect the additional demand that channel will bring — that difference must be added manually. Similarly, a planned price increase will directly affect past sales velocity, and the forecast should be adjusted accordingly.
Spreading risk with phased ordering
Ordering all the stock needed for a season in one go passes every forecasting error straight through to your bottom line. Instead, wherever possible, follow a phased ordering strategy:
- Place your first order based on a conservative lower bound of your forecast.
- Track the actual sales velocity daily during the first 1-2 weeks after the season starts.
- If actual velocity deviates from the forecast, adjust the quantity of the second order up or down accordingly.
- Repeat this cycle up to mid-season, as long as lead time allows.
This strategy works particularly well for products with short supplier lead times (the time between order and delivery). For products with long lead times (overseas manufacturing, for example), the option for phased ordering is limited; for these products, it becomes more critical to get the forecast right the first time — so it makes sense to invest more forecasting effort in long-lead-time products. Whether you run your supply chain through your own warehouse, a 3PL partner, or dropshipping also determines how much flexibility phased ordering can offer; we covered these models in our fulfillment models comparison.
Balancing the cost of stockouts against overstock
Both mistakes are costly, but in different ways. Stockouts cost you lost sales as well as customer trust; a customer who keeps seeing "out of stock" will eventually turn to another store at the start of the next season. Overstock, on the other hand, creates pressure to discount or liquidate at the end of the season, and it also eats into the warehouse space needed for the next season's products. We covered how to use warehouse space efficiently by product class in detail in our warehouse layout and picking efficiency guide.
Which mistake is more costly depends on the product: a bit of extra stock on a high-margin, easily reorderable product is relatively harmless, while overstock on a low-margin product with a fast fashion cycle can turn into a serious loss. That's why your safety buffer shouldn't be a fixed percentage across the board — set it based on each product's margin and fashion cycle.
Monitoring and adjusting during the season
Forecasting isn't finished the moment the season starts. Compare actual sales velocity in the first week against your planned forecast; if the deviation exceeds a certain threshold (say, more than 20%), update your remaining supply decisions and campaign planning with early intervention. Waiting until mid-season can eat up the lead time you'd need to make a correction. You can find how to tighten this monitoring cycle during high-intensity campaign periods like Black Friday in our Black Friday preparation calendar article.
Quick checklist
- Have at least two to three seasons of historical sales data been analyzed?
- Have one-off effects (campaigns, external events) been filtered out of the forecast?
- Have products been segmented via ABC analysis for differing levels of forecasting effort?
- Has the campaign calendar and any new channels been factored into the forecast?
- Is there a phased ordering plan for products with short lead times?
- Has the safety buffer been set per product based on margin and fashion cycle?
- Is actual in-season sales performance being compared against the forecast weekly?
Running this analysis by hand in scattered spreadsheets quickly becomes unmanageable for a large catalog. Şimşek Software's e-commerce platform reports your historical order data by product and category, making seasonal sales velocity visible; we can go through your existing sales history together to ground your pre-season stock decisions in this data.