eCommerce Inventory Forecasting in Switzerland: Methods and Formulas for SMEs (2026)

An inventory forecast is useless if it’s based on data that doesn’t reflect the reality of the warehouse. This is the starting point of this guide.

Three concepts are often confused. Demand forecasting estimates the market’s appetite for a product, regardless of your ability to sell it. Sales forecasting translates this demand into expected revenue, channel by channel. Inventory forecasting determines the physical quantities to hold based on current levels, supplier lead times, and storage locations.

This distinction matters for a multichannel small business. The same product SKU might sell quickly on Galaxus but slowly in your own store. Forecasting a single overall figure masks these discrepancies and leads to stockouts.

One term comes up repeatedly throughout this guide: WMS. A WMS (Warehouse Management System) is software that organizes and tracks inventory operations, from receiving to shipping. Emaloja relies on a WMS that enables real-time tracking of inventory and orders.​

The guiding principle can be summed up in one sentence: a formula is only reliable if its inventory data reflects the physical situation and actual movements. An elegant forecast based on incorrect inventory levels leads to incorrect orders.

Before choosing software or a formula, verify the data you have available. A sophisticated method applied to incomplete data will inevitably produce false results.

The recommended baseline is at least two years of transactional sales history by SKU and by channel, including returns and cancellations. An SKU (Stock Keeping Unit) is a unique product identifier. The analysis must account for the sales velocity and life cycle of each SKU: launch, growth, maturity, and decline.

Certain data must come from the WMS in real time: physical inventory, available inventory, inventory reserved for orders, inventory in transit, receiving transactions, actual outbound shipments, storage locations, and open orders. Emaloja’s WMS automatically updates online store inventory in real time.​

Before incorporating historical data into a trend, clean it up. Filter out outliers—such as a one-time inventory clearance—to prevent a single event from skewing the average. For each SKU, you should be able to identify the source, update frequency, and owner of each data point.

Sales and demand data

Sales by SKU and by channel form the foundation. Add external variables that may influence demand: promotions, price history, marketing spend, web traffic, weather, and search trends. These signals explain why a given SKU experiences an uptick or downturn without any underlying change.

Include returns and cancellations in the history. A canceled or returned sale does not count as a unit consumed by the market. Net demand is the only actionable metric for determining inventory levels.

Inventory and supply data

Four levels of inventory must never be confused. Physical inventory is what is counted in the warehouse. Available inventory is what remains for sale after reservations. Reserved inventory has already been allocated to orders currently being prepared. Inventory in transit is expected but not yet received.

Confusing these four values leads to “invisible” stockouts: the store displays inventory that the warehouse has already allocated elsewhere. Supplier lead times—both average and variable—are used to calculate safety stock and the reorder point.

Events that distort or alter demand

Sales events, Black Friday, and the holiday season can distort a trend if you treat them as a permanent increase. Document each campaign: dates, discounts, marketing expenses, and substitution effects between SKUs.

A massive discount on one product can cannibalize sales of a related product, which your forecast must account for.

How can I forecast my ecommerce sales to avoid stockouts? The answer lies in a methodical approach: analyze historical data by SKU, distinguish between channels, and adjust for seasonality. Start by understanding the past to anticipate the future.

Choose a method based on demand stability

For products with stable demand, a 30-, 60-, or 90-day moving average smooths out fluctuations and provides a predictable baseline [1].When demand changes, a weighted moving average gives greater weight to recent sales and better tracks the current trend.

Exponential smoothing is another basic method for stable products, used in conjunction with the moving average. It adjusts the forecast with each new observation, applying a weight that decreases for older data.

Start with simple, easy-to-understand methods for regular SKUs. Reserve more advanced models for product lines where the volume, volatility, and data quality justify their use.

Adjusting for trends, seasonality, and events

Seasonal decomposition breaks down data into three components: the underlying trend, seasonality, and random variation [2].It is suitable for products with seasonal peaks that are predictable from one year to the next.

Incorporate the Swiss and European calendars: sales events, Black Friday, and the end-of-year holidays. These dates recur and can be planned for.

When you want to model the effect of marketing spending, prices, or market variables on sales, regression analysis becomes useful [3].It links several explanatory factors to expected sales volume.

Measuring the gap between forecast and actual

MAPE (Mean Absolute Percentage Error) is the average percentage difference between the forecast and the actual result. Track it by SKU and by channel—never just at the aggregate level.

Before approving a new model, require at least a 20% improvement over the existing method. Without a measurable net gain, a more complex model adds risk without providing any benefit.

Finally, set a forecasting horizon that covers at least each supplier’s replenishment lead time. Forecasting over a period shorter than the supplier’s lead time is equivalent to placing orders too late.

Safety stock is the buffer designed to absorb discrepancies between the forecast and actual demand, particularly demand spikes and supply delays [4].​

Define each variable before any formula:

• SS: safety stock, in units
• Z: service level coefficient, dimensionless
• σD: standard deviation of daily demand, in units per day
• D̄: average daily demand, in units per day
• L: stable supplier lead time, in days
• L̄: average lead time, in days
• σL: standard deviation of lead time, in days

Time periods must be consistent. Use either daily demand with lead times in days, or weekly demand with lead times in weeks. Never mix the two.

Case 1: variable demand and stable supplier lead time

When the supplier lead time is reliable and only demand varies, the formula is SS = Z × σD × √L, where L is the stable supplier lead time expressed in days [5].You measure the standard deviation of daily sales, choose a Z value based on the target service level, and then multiply by the square root of the lead time.

Fictitious illustrative example. A product sold 14, 22, 18, 25, 16, 20, and 25 units over seven days. The average D̄ is 20 units per day. To obtain the standard deviation σD, first calculate the deviations from the mean, square them, take their average, and then take the square root: the squared deviations are 36, 4, 4, 25, 16, 0, and 25, for a sum of 110 and a variance of 110 / 7 ≈ 15.7. The square root gives σD ≈ 4 units per day.

To stay close to typical figures, let’s use σD = 6 for a more volatile inventory. The supplier lead time is stable at 9 days (L = 9). For a 95% service level, Z is 1.65. The safety stock is 1.65 × 6 × √9 = 1.65 × 6 × 3 = 29.7, rounded to 30 units. You therefore hold 30 units of buffer stock in addition to the coverage for average demand.

Case 2: variable demand and supplier lead time

When both demand and lead time vary, the formula combines the two uncertainties quadratically: SS = Z × √((σL × D̄)² + (L̄ × σD)²)​ [6].​

Selecting the target service level

The service level determines the coefficient Z. Common benchmarks are: 90% corresponds to Z ≈ 1.28; 95% to Z ≈ 1.65; 97.5% to Z ≈ 1.96; 99% to Z ≈ 2.33 [7].​

The trade-off is straightforward. Raising the service level reduces the risk of stockouts but increases the safety stock. Reserve the highest levels for high-margin or strategic SKUs.

When statistical data is lacking, the fallback “average-maximum” method remains useful: safety stock = maximum sales during the maximum lead time − average sales during the average lead time [8].​

A general rule of thumb of 20 to 30% buffer stock can serve as a starting point for highly variable products, to be adjusted based on supplier lead time and actual data. Two words of caution to conclude: do not apply a statistical formula to a SKU without sufficient historical data, nor to data contaminated by uncorrected stockouts.

The reorder point is the inventory threshold that triggers automatic replenishment before a stockout occurs. It links the forecast to the safety stock calculated earlier.

The Reorder Point formula

The main formula is: reorder point = average daily demand × average replenishment lead time + safety stock [9].​

In shorthand notation, ROP = D̄ × L̄ + safety stock, where D̄ and L̄ are exactly the variables from the previous section [10].The reorder point therefore covers the expected demand during the lead time, plus a buffer for uncertainty.

The Reorder Rule by SKU

Calculate the threshold by SKU, never for the entire catalog. Sales rates, lead times, margins, and variability differ from one SKU to another.

Distinguish between three levels of automation. A restocking alert indicates that a threshold has been reached. A supplier order awaiting approval prepares the purchase but requires a human decision.

An automatic order is submitted without human intervention. Not all suppliers accept the same level of automation, so define the approval process for each supplier.

Order quantity and checks before approval

The order quantity must take into account demand over the selected coverage horizon, available inventory, inventory in transit, orders already placed, and, if necessary, supplier minimum order quantities. Placing an order without checking inventory in transit can lead to duplicate orders.

Optionally, the economic order quantity (EOQ) balances ordering costs and storage costs: EOQ = √(2DS/H), where D is annual demand, S is the ordering cost, and H is the annual storage cost per unit [11].​

Test any automation on a limited group of SKUs before rolling it out to the entire catalog. A poorly calibrated rule applied to the entire catalog multiplies errors.

The forecast is only valid if projected sales align with the inventory actually available for sale. Do not confuse store data with physical quantities in the warehouse.

The data flow to be synchronized

The WMS orchestrates a precise sequence:

1. An ecommerce order is placed on the online store or marketplace.
2. The corresponding inventory is reserved.
3. The actual shipment is prepared, packed, and labeled.
4. Available inventory is updated.
5. Stock coverage and reorder points are recalculated.
6. An alert or a supplier order proposal is triggered if the threshold is reached.

Emaloja uses a​ WMS for precise inventory tracking, fast order fulfillment, and compliant packaging and labeling.Here inventory and order tracking are done in real time.

Managing multi-warehouse and cross-border inventory

In a multi-warehouse environment, aggregate quantities by storage location while maintaining the actual availability at each site, reserved inventory, transfers, and inventory in transit. A total figure masks the fact that an item available in Geneva may not necessarily be available at the warehouse responsible for shipping.

For Switzerland–France–European Union shipments, do not consolidate inventory and lead times without considering the shipping warehouse, the supplier, and the actual lead time associated with each shipment. Emaloja has​warehouses located in Switzerland, France, and other European countries,and the flow of orders between these warehouses is automated by the WMS.​

Why real-time inventory improves decision-making

Real-time, up-to-date inventory prevents sales from promising items that the warehouse has already allocated elsewhere.Integrating a WMS is beneficial for companies with more than 100 products, multiple sales channels, or increasing logistical complexity.

Before investing in a more complex forecasting solution, ask yourself this diagnostic question: Does your WMS report real-time inventory levels to the forecasting tool? Without this data feed, no formula will produce reliable decisions.

Formulas produce false signals when the data does not reflect actual demand. This section sets the limits to be observed.

Promotions and sales peaks

Treat promotions, sales, Black Friday, and holidays as events separate from baseline demand— never as a permanent upward trend.Otherwise, the forecast will consistently overestimate demand after each peak.

Document campaign dates, assumptions,discounts, marketing expenses, and substitution effects between SKUs. This log makes it easier to adjust the forecast and analyze subsequent campaigns.

New products with no history

A new product with no sales history requires documented assumptions, given the lack of transactional data. This is a limitation of the method: no statistical formula applies to a product that has not yet been sold. The detailed method is described in the FAQ.

Historical out-of-stock periods, returns, and intermittent sales

Past out-of-stock periods underestimate observed demand. Sales recorded during an out-of-stock period do not reflect the demand the market would have generated if the product had remained available. Adjust for these periods before calculating a trend.

Include returns and cancellations in the sales history used for forecasting. For intermittent sales or very inactive SKUs, opt for a manual review of assumptions and thresholds rather than uncontrolled automation. Keep a log of manual corrections made to the forecast and their rationale.

A method is managed using metrics tracked by SKU and by channel, not at the aggregate level.

Key metrics to track by SKU and channel

Track the out-of-stock rate, inventory coverage in days, forecast accuracy by SKU and channel, service level, bias, inventory turnover, and the frequency of manual adjustments. Inventory turnover is the ratio of outflows over a given period to the average inventory over the same period [12].​

Forecast bias measures the systematic tendency to overestimate or underestimate demand. It differs from mean absolute percentage error (MAPE): two opposing errors can cancel each other out in bias, whereas they inflate the MAPE. Tracking both together reveals whether your deviations are random or biased.

Implementation checklist

It takes just 6 steps to implement the method:

• Ensure the reliability of sales and inventory data.
• Segment SKUs by volume and variability.
• Choose a forecasting method for each segment.
• Calculate the appropriate safety stock for each SKU.
• Set the reorder point and alert rules.
• Test on a pilot group, while also monitoring cross-border flows (supplier lead times, transfers, and inventory in transit between Switzerland, France, and other European destinations), then review the variances.

Regular inventory counts remain necessary to verify consistency between system inventory and physical inventory [13].​

Review schedule and responsibilities

Establish a clear schedule: weekly monitoring of stockouts and inventory coverage, monthly review of MAPE, bias, and service levels. Emaloja’s WMS is designed to optimize warehouse operations, strengthen inventory control, and support ecommerce growth​

Appoint two managers: one responsible for data quality and one responsible for validating restocking exceptions. Without designated owners, alerts pile up without any decisions being made.

The key principles can be summarized in a short list:

• Forecast by SKU and by channel, not at the aggregate level.
• Clean the data before any calculations.
• Adjust the safety stock to account for variability in demand and lead time.
• Automate the reorder point with a defined approval workflow.
• Synchronize the forecast with the WMS in real time.
• Track KPIs by SKU and correct variances.

Start by evaluating three things in your catalog: the quality of inventory reporting, the reliability of supplier lead times, and your replenishment rules.

For a discussion on WMS integration, real-time inventory tracking, and the organization of ecommerce fulfillment and order picking between Switzerland and the European Union,Emaloja can assist you.