The mathematics of Christmas demand peaks – How companies can prepare for forecast uncertainty

Demand forecasting has never been a straightforward exercise. While organizations have long tried to uncover patterns in historical data, the inherent uncertainty of demand has always posed natural limits. For decades, companies have attempted to balance business intuition, statistical techniques, and market signals—yet every season brings a new layer of complexity. The Christmas period illustrates this particularly well: demand not only surges but becomes more volatile, influenced by external factors that traditional forecasting approaches are often too rigid or too simplistic to capture effectively.

International research highlights just how critical demand forecasting has become. A survey of more than 300 professionals found that 61 percent consider improving forecast accuracy to be “very important” or “critical,” while only 2 percent are satisfied with their current performance. For fast-moving products, typical item-level forecast error ranges between 20 and 40 percent—already a significant gap, especially during the year-end season when volatility tends to intensify. For slower-moving items, the variation can be even higher, which makes it essential for forecasting methods to distinguish between different product behaviours and treat each category with the appropriate level of granularity.

Source: https://www.aimms.com/story/latest-news-new-aimms-survey-shows-that-organizations-are-betting-on-new-technologies-to-improve-forecast-accuracy/

The research also shows that most organizations still rely on Excel-based forecasting, a method that becomes extremely labor-intensive with large product portfolios and offers limited capability to detect subtle seasonal shifts or one-off demand spikes. It is therefore not surprising that more companies are turning to statistical models, machine learning techniques, and demand sensing solutions. These approaches are far better suited to uncover the underlying structure of demand and to manage uncertainty in a systematic, scalable way.

From a mathematical perspective, the holiday season represents a highly complex time-series challenge. What appears to be a simple surge in demand is, in reality, the combined effect of three components: trend, seasonality, and event-driven factors such as promotions or price campaigns. If the forecasting process cannot properly disentangle these elements, the holiday peak can easily distort future projections. Overestimating demand leads to excess inventory in January, while underestimating it results in empty shelves as early as mid-December.

A shortage in the days leading up to Christmas is typically a signal that the forecasting process has failed to separate trend, seasonality and exceptional demand spikes. Many companies still rely on manual, Excel-based forecasting methods that struggle to account for seasonal fluctuations and promotional effects. As a result, certain products sell out within days, and availability can vary widely across stores—not because the market is inherently unpredictable, but because the underlying methodology is not sensitive enough to the structure of demand.

Overestimation, by contrast, becomes visible only after the season ends. When the forecasting process embeds an exaggerated seasonal effect, companies enter January and February with inventory levels that no longer reflect actual demand. Slow-moving stock ties up capital, increases storage and financing costs, and often can be sold only through heavy markdowns. This is not merely a commercial issue but a financial one: post-season clearance sales are a direct consequence of overestimation, frequently forcing sales below profitable levels.

After the holiday period, excess inventory quickly becomes a financial burden. As demand drops and unsold stock accumulates, companies are often pushed into markdowns, frequently at margins below profitability. Clearance activity is a clear indication of what happens when the holiday peak is overgeneralized in the forecasting process and carried forward into future expectations.

Modern forecasting systems are therefore not designed to produce a single “most accurate” number. What matters far more is the ability of the methodology to manage variance: how demand may evolve across different probability ranges, how widely data can disperse around the seasonal peak, and how uncertainty can be embedded into planning in a way that keeps business decisions stable and well-grounded.

Machine learning–based forecasting, hierarchical models and demand sensing are all approaches that help separate trend and seasonal components more clearly and prevent exceptional values from distorting future plans. In this sense, the Christmas season provides a valuable datapoint each year: it reveals how the system performs under extreme conditions and highlights where risk buffers, adjustments or methodological changes may be required.

The holiday demand peak is therefore not an isolated event, but a true stress test for the entire forecasting process. Companies that systematically analyse the lessons of the season become more resilient, more predictable and more data-driven over time. In this sense, improved forecasting is not about chasing a single accurate number, but about making uncertainty manageable, keeping inventory practices sustainable, and ensuring that decisions rest on solid foundations. At Optasoft, we support this mindset with modern, data-driven forecasting and optimization solutions designed to strengthen decision-making in an increasingly unpredictable market environment.