In such times, general top-down adjustments of a manually or statistically generated demand forecast could create a significant over- or under-supply at item level.

Data show that the COVID-19 situation has caused significant and sudden shifts in demand within and across markets, customer segments, product categories and sales channels. They also clearly show that the nature and severity of the impact on demand also varies significantly on a detailed product item level.

Many companies have therefore introduced and are leveraging demand sensing technologies to generate a realistic demand forecast at item level on a continuous and frequent basis.

In short, ‘demand sensing’ is a technology that you can apply to produce as realistic a short-term demand forecast as possible. The technology collects and analyses internal and external data in near real-time, such as open orders, point-of-sale data, weather data, etc. Demand sensing is thus a way to help bridge the typical gap between medium- to long-term demand plans and the short-term forecast required for frequent (daily or even hourly) demand and supply planning.

A word of caution: Be careful not to be too ambitious when implementing a demand sensing solution. Many companies have been too ambitious in terms of the number and type of data sources to include in their demand sensing solution and thus failed to establish a functional solution. Implementing and leveraging a solution based on a ‘minimum viable product’ design principle therefore is recommended, taking the following approach:

  1. Identify and clearly define the data that are relevant, easily accessible and representative for demand sensing purposes
  2. Set up the demand sensing solution and start collecting and analysing data on a daily or even hourly basis. Alerts should be incorporated in the demand sensing solution to enable immediate attention and reaction to any significant changes in the demand within a timeframe of hours, days and/or weeks
  3. Continuously monitor demand signal changes as compared to the detailed statistical demand forecast pattern
  4. Evaluate the statistical significance of the ‘sensed’ demand changes to avoid a ‘random walk’. In other words, avoid setting up a demand sensing process in which there is no observable pattern or trend that is significant, i.e. where the movements of a forecast object or the values taken by a certain variable are completely or partly random. The statistical significance must be evaluated on two key variables: Sample size and effect size
  5. Analyse actual demand and perform adjustments to the statistically generated, short-term demand forecast using automated routines
  6. Identify and respond to any supply issues or opportunities based on the revised short-term forecast as part of your Sales & Operations Execution (S&OE) process

Again, implementing an integrated demand forecasting solution based on advanced statistical forecasting and demand sensing technologies should not be a major investment in terms of time and resources. Speed is more important than ever, so go for a solution that can be up and running sooner rather than later and then build on that.