With the growing recognition that supply chain is a strategic competitive differentiator, what would it mean if your planners could spend most of their day actually planning?
Demand planners are increasingly expected to be statisticians, not people who understand the business they’re predicting. A recent Demand Planner job posting by a multinational consumer products company lists statistical forecasting and troubleshooting statistical models as half the role; whereas strategic planning is only 10%. The ratio of planning activities is out of balance. Instead of tinkering statistical models, what would it mean to the business if planners had time to actually plan?
While understanding the limitations of statistical techniques is important for planners, the growing requirement for advanced statistical knowledge as a prerequisite has reached the point where the practice damages supply chain performance. If drivers were expected to understand advanced combustion principles and required to tune ignition timing as engine conditions change, the global automotive industry would never have grown to a $4 trillion market. Instead, efficient software runsquietly behind the scenes so that drivers can concentrate on their core task – i.e., safely driving the car. So why do weexpect demand planners to open the hood and tinker with ever more sophisticated statistical models when this is a task better suited to software?
The answer is that the current approaches to demand prediction are very limited and require considerable user intervention. Yet, accurate forecasts are important for successful supply chain operations. Prediction of future demand forms the basis for all manufacturing, procurement and inventory decisions –affecting overall supply chain costs – so getting forecasts right is pivotal. The need to manually tune models is a byproduct of the weak statistical analysis methods used by demand planning systems for decades and still relied upon by most of the industry. Over the years, the underlying logic has remained unchanged, despite the increased number of demand signals and greater market complexity.
To forecast future sales, traditional demand planning techniques fit historical shipments to simplistic seasonal time series models. The math behind these models performs best with large datasets spanning many years and is commonly used outside the supply chain for tasks like predicting weather. But any drought-stricken farmer whose livelihood is threatened by a string of heat wavesknows how hard it is to accurately predict weather even with 100 years of history. So it comes as no surprise that these models make poor choices for predicting future demand in industries like consumer packaged goods where half the products have less than two years of history, especially when volatile markets and economic uncertainty have created unprecedented shifts in consumer behavior. These limitations show in the numbers. For CPG, a typical baseline statistical forecast has60% error.
Sadly, the availability of inexpensive computing power has resulted in dozens of models being available to planners, each with various parameters. With so many options, planners can easily find themselves spendingconsiderable time fitting each series in the search for something better. The unspoken “elephant in the room” is that despite tuning, there is pretty limited information available in historical data. Achieving better forecast accuracy requires more data – much more. The good news is that the data is there and so too are automated software applications to sort through multiple demand signals, determine what is predictive and create a better baseline forecast for planners. Using the prior example, instead of a 60% error, forecasts from Long-Term Demand Sensing are in the 35% range – a 40% reduction in error. Using sophisticated mathematics augmented by multiple signals, Long-TermDemand Sensing automatically creates a better forecast. It’s realizing the promise of Big Data, to provide step-changes in performance by doing things that were not otherwise possible. And it frees planners from repetitive manual functions to focus on areas that are most important like strategic planning and business discontinuities. With the growing recognition that supply chain is a strategic competitive differentiator, what would it mean if your planners could spend most of their day actually planning?
