Demand sensing is a concept for capturing and modelling various big data demand signals into intelligence to foresee short-term sales demand patterns and demand changes to proactively adapt your supply chain planning. Reduction of scrap, increased customer service levels, lower inventories, improved replenishment decisions and reduced time spent by the planners are the business benefits of a more accurate short-term sales forecast.
In today’s markets, customers are demanding shorter lead time, more availability and more variants and newer products every day, putting a costly pressure on the companies providing the services and goods. Achieving a high customer service level potentially comes with an increase in inventories and a high risk of increasing scrapping cost – especially within fast-moving consumer goods. Mitigating the risks associated with high inventory and scrapping requires better short-term decision-making based on more nuanced forecasts.
It is obvious that the sales of tomorrow of a given product is influenced by many different elements, from market trends, social media exposure, product substitution, customer order patterns, week day patterns, POS data, cannibalisation, competitor activity, etc. The challenge is, despite having all the information available, that it is difficult to paint a picture without an in-depth analysis. With hundreds or thousands of product combinations, and with new input coming in every day, it is simply not a viable option to manually adjust the forecast week on week. However, it is vital to be able to interpret this information on an ongoing basis to make good short-term planning decisions in order to balance a high service level with an ambitious inventory target.
How can demand sensing help to comprehend the demand signals and extrapolate it into a short-term forecast in order to increase the short-term forecast accuracy and enable better plans with less scrap and improved customer service level?
Demand sensing is a method for creating a short-term forecast based on a range of recent inputs to the forecast. In traditional statistical forecast methods, the forecast is in most cases based on one input, the sales history, which is smoothed or averaged into a future forecast. On the contrary, demand sensing incorporates a broad span of recent demand signal inputs in order to project the detailed forecast for the nearest forecast horizon. Based on advanced regression models and pattern recognition of the past history of the demand signals, the demand sensing algorithms are able to make a qualified guess on the short-term forecast.
Naturally, the worldwide trend in increasing computer performance and data storage has enabled the possibilities for performing such magnitudes of analytics with rapid response time. A typical demand sensing flow consists of the following steps:
In our experience, demand sensing works very well in industries where a lot of shortterm decisions are made due to demand variations and are rapidly changing market characteristics. The demand variations are typically influenced by weather, campaigns and cannibalisation, competitor activities in the market place and microtrends. The short-term forecast decisions are critical for being able to produce and distribute products with a high customer service level, without building stockout of the roof or risking obsolescence.
Demand sensing provides higher short-term forecast accuracy, which is an enabler for efficient short-term decision-making. Moreover, and likewise importantly, demand sensing frees up time for the planners not having to supervise the forecast for products where demand sensing performs well. Time can thus be spent on areas that are truly difficult and unpredictable such as new product introductions.
Demand sensing brings a lot of good potential. However, we do also believe that it is very important to focus on how it SUPPORTS the planners – not replaces them with a black box of intransparency. In order to keep the planners’ trust in the figures provided by demand sensing, it is critical that the planners understand how to manage and keep the sensed demand in a leash, e.g. setting thresholds and control parameters and reacting to well-designed exception alerts. Otherwise, we risk losing trust in the forecast and create frustration and thereby cause even more manual work.
We also believe that an important point is that demand sensing only works on the short-term horizon where the short-term demand signals live. For the mid- and long-term horizon, we still need to focus on keeping the statistical forecast simple and focus on the demand planning process itself rather than complex algorithms - see the article: Sales forecasting is not about complex algorithms.
Customer x-perience on purpose
Implement Consulting Group