News | Articles | Press Releases

Demand Forecasting: How technologies explain the future

Hanger with new clothes in a clothing store


Uncertainty and the subjunctive are the biggest enemies of Merchandise Planning: if spring hadn’t been so rainy, the summer collection would surely have captured record sales. If one had known that this season precisely this look would become the style of the stars, one would certainly have jumped onto the trend early on.

Unfortunately, however, for every hit in fashion retail, more failures occur than a success-oriented buyer would like to admit.

One is always smarter after the fact – unless of course you already have intelligent tools and a smart strategy in place in advance.

Ironically, the most efficient strategy to counter increasingly unpredictable consumer behavior is based on the same factor that actually fueled this rapid development: Data. A lot of data.

Customers leave digital traces everywhere, and successful demand forecasting uses this data to avoid both shortages and surplus inventory. Precise and reliable forecasts of future demand are an invaluable asset when it comes to producing the right items in the right quantities at the right time and making them available exactly where the customer needs them, looks for them and finally purchases them.


However, with each new market segment, each new location and each new channel, the complexity of forecasting systems increases exponentially.

Thus, any corporate and purchasing strategy inevitably very quickly reaches the limits of human judgement and abstraction. Classic forecast methods based on the time series method (for example, exponential smoothing or ARIMA), prove somewhat more resistible, but unfortunately also have their weaknesses. Although they provide relatively detailed information about past developments in demand, they do not allow any conclusions to be drawn about the actual drivers of these developments nor do they provide insight into future developments. Thus, for instance, these traditional methods of analysis either can’t quantify an advertising effect at all, or only through human intervention.


If, however, an intelligent system could automatically identify these influencing factors and take them into consideration for the next forecasting period, forecasting accuracy would increase enormously. This is exactly where Adaptive Merchandise Planning (AMP) comes into play. AMP is based on the principle of machine-based learning and uses intelligent algorithms such as artificial neural networks or deep neural decision forests. In a flood of historical sales data, these algorithms can recognize relevant patterns and interactions, from which clear recommendations for action can be derived for the future. The principle is similar to learning a new language: through application experience and a steadily growing vocabulary, language patterns are recognized, context is understood and thus one’s own communication ability is continuously improved.

In simple terms, Adaptive Merchandise Planning uses learning algorithms to replicate how the human brain functions.

That sound like science fiction? With its new assortment planning at the store level, H&M is demonstrating that this is already the way forward (read more).

Follow us

More Posts

Someone else’s computer: Is cloud computing secure?

Someone else’s computer: Is cloud computing secure?

We see many retail experts who are afraid of “the cloud”. But in the end, it’s just someone else’s computer. In order to process big amounts of information, we need a well-performing computer system that is robust and resilient.

Press Contact

Patrick Brüns - Head of Marketing

Patrick Brüns
Head of Marketing

+49 172 319 8604

Cookie Consent with Real Cookie Banner