Demand Sensing: Part 2

7 Essential capabilities of Demand Sensing were summarized in the previous edition. Demand Sensing in action encompasses several steps aimed at accurately capturing and responding to short-term demand fluctuations: These are,

Importing Short-Term Demand Data: Instead of relying on weekly or monthly data, Demand Sensing involves importing short-term demand data on an hourly or daily basis and at an order line level. This allows for a more granular understanding of demand patterns.

Immediate Sensing of Demand Signals: Demand Sensing promptly detects changes in demand signals compared to detailed statistical demand patterns. This agility enables swift response to emerging trends or deviations.

Evaluating Statistical Significance: The system evaluates the statistical significance of these changes in demand signals. This assessment helps distinguish between meaningful shifts in demand and random fluctuations.

Analyzing Partial Period Actual Demand: Demand Sensing analyzes partial period actual demand data and executes short-term forecast adjustments using automated routines. This ensures that forecast predictions remain in alignment with the evolving demand patterns.

Responding to Replenishment Issues: Advanced statistical analytics enable the identification and rapid response to replenishment issues or sudden changes in customer demand. This proactive approach minimizes disruptions in supply chain operations. To ensure its operability, it is essential that all relevant status vectors across the channels are synchronized.

To gain potential network benefits, the platform must seamlessly integrate planning and execution processes while efficiently replenishing high-frequency demand signals with optimized execution. Crucially, there should be zero latency between planning functions, such as between the demand module and inventory and replenishment modules. Unfortunately, many supply chain planning solutions have evolved through acquisitions. Consequently, forecasting may utilize a different data structure from replenishment, with later additions like inventory planning modules introducing yet another data structure. Such discrepancies can severely hinder Demand Sensing capabilities. This underscores that for the expected levels of agility and responsiveness, mere sensing of demand is not enough. In this context, only very few solutions can offer true Demand Sensing, though everybody talks about it.

In the same vein, another obstacle to effective demand sensing is the contentious notion of “functional excellence”. This arises from the belief that no single platform can offer the best of capabilities across all functional modules. Consequently, relying solely on an integrated platform is deemed insufficient. Hence, many companies opt for separate tools: one for demand forecasting, another for S&OP, a third for supply planning, and so forth. While there is some validity in this argument, it reflects remnants of the traditional “functional silo mindset” that once dominated supply chain operations. It's questionable whether such a fragmented landscape can effectively harness the benefits of demand sensing.

Moreover, it's ironic that numerous organizations have now established dedicated data science teams that often seem detached from real business objectives. For these teams, the focus tends to be solely on data, algorithms, and achieving forecast accuracy. For them, anything below 90% accuracy is not acceptable and they seem to miss the business context altogether.

Similarly, for many supply chain professionals with backgrounds in data science, Demand Sensing is perceived as applying an array of demand forecasting algorithms at will. They try best-fit methods, and through tweaking algorithms hope that somehow, they can pull out a magical combination that will give accurate forecasts. While this reflects their academic curiosity and initiatives, the real benefits are extremely marginal. They are missing opportunities, and more importantly delaying their business from achieving radical innovation.

While Demand Sensing, with the right technology, data, and processes, can indeed enhance short-term forecast accuracy, it is important to recognize that forecast accuracy measures directly impact (on a one-on-one basis) service levels for regular items only. For intermittent, erratic items, conventional forecast accuracy measures need not directly translate to service levels. For such items inventory accuracy (please see previous editions) and multi-echelon replenishment becomes more crucial in achieving service level targets.

With the continuous expansion of the "long tail," (volatile, erratic, and intermittent items) it becomes imperative to clarify our objectives: are we primarily focused on enhancing forecast accuracy, or on fulfilling customer orders efficiently. In truth, Forecasting is not the end but is the first step to provide service levels to customers, and the service levels are measured on a daily and order-line basis. Even with a very high forecast accuracy, achieving high service levels are not guaranteed. This is because accuracy measures are ex-post and monthly aggregates, while order fill rates are assessed at the line order level and within the month and on a daily basis.

In summary, Demand Sensing is not about applying an array of time series methods more frequently. Such traditional demand forecasting methods often struggle to capture sudden changes in demand due to change in customer behavior and external factors. In contrast, demand sensing utilizes real-time data and advanced analytics to generate more accurate short-term forecasts by capturing and responding to dynamic market and supply chain conditions. This is an essential capability for service-driven supply chains. While terms like "Demand-Driven," "Supply-Driven," or "Market-Driven" are commonly used, it is crucial to recognize that supply chains ultimately revolve around the service levels provided to customers and consumers.