Probabilistic Supply Chain Forecasting
The enormous complexity and volatility businesses face today requires probabilistic supply chain forecasting that accommodates the reality of demand uncertainty. Supply chains face enormous challenges in product line proliferation, shorter product life cycles, increased complexity, huge volumes of data and ever more demanding and unpredictable customers. Despite all this, many businesses still use planning systems and forecasting approaches that were designed for simpler times. These aren’t engineered to integrate, analyze and take advantage of increasingly available data.
What is probabilistic supply chain forecasting?
At its core, probabilistic demand planning helps you decode uncertainty. Sometimes called probability forecasting, it works together with machine learning engines to analyze multiple demand variables to identify the probabilities of a range of possible outcomes. This probability distribution is derived by modeling both order lines frequency and order lines size; this deeper information is ignored by most traditional forecasting packages, and allows you to generate accurate demand behavior much more quickly than considering demand history alone.
Probability forecasting is the only reliable approach to plan for unpredictable, slow-moving, long-tail SKUs, and those with limited or no order history. The beauty of the probability forecast is that it helps you manage the risk that comes from demand volatility. It’s not just about improving average demand predictions but assessing the entire range of possible outcomes, which have the biggest impact on service levels. This self-tuning approach allows you to predict demand behavior much more accurately than considering demand history alone. By extracting the demand signal from the noise, probabilistic supply chain forecasting helps companies reduce demand volatility and risk.
“One of the biggest issues for SCP is uncertainty — especially unknown uncertainty. The application of machine learning helps to convert this unknown uncertainty to known variability.”
Why supply chains need probabilistic forecasting
The world is full of forecasting algorithms that are often offered in “best fit” groups. The problem with having a variety of forecasting methods is that they are each designed for one type of thing. Holt-Winters (or Exponential Smoothing and the derivatives) are great for fast movers. Croston’s Method (or Poisson Distribution) are designed for slow movers. Holt-Winters is slow to adapt and it cannot do slow movers at all. Croston’s Method cannot well handle spikes in demand. “Best-fit” algorithms assume that the forecast starts after demand has been measured. They also tend to focus on a simple count of shipped or sold goods, and struggle to extend to “causal” factors.
As opposed to algorithms designed a century ago, modern probabilistic demand planning systems work much like the human mind. Seasoned buyers and planners know their customers, seasonality, performance of their vendors and which products require higher service.
How to succeed in forecasting intermittent demand
Achieving a high service level in an intermittent demand world requires probabilistic supply chain forecasting that can master demand behavior and probability distributions across wide ranges of demand patterns. Look for these capabilities:
Reliable demand modeling that automatically understands all types of SKU behavior and elements of demand uncertainty and considers not only forecast error, but also the variability caused by order-line frequency and order-line size distribution.
Advanced inventory modeling that provides highly reliable descriptions of statistical inventory behavior by eliminating the gross approximations of traditional inventory management theory.
Logic for demand signal propagation that combines the superior capabilities of both demand modeling and inventory modeling to model the impact of replenishment policies and constraints at each echelon of the supply chain. This will ensure alignment of plans and actual performance, reduces the need for planner intervention, and enables better service performance.
Take probabilistic demand planning to the next level with demand sensing
Further refine your probabilistic demand planning by applying machine learning-driven demand sensing. In addition to requiring continuous tuning, traditional forecasting typically forces you to work with larger time buckets and for a longer time with more uncertainty. This means your forecast senses short-term trends immediately. You can get started with “short-term forecasting” using sell-in data. Then you can begin to use sell-out data such as customer, PoS, channel data or other external variables correlated with the demand. This self-learning automated model identifies demand trends, provides advanced warning of problems and removes the latency between the plan and what’s actually happening in your supply chain.
This is only a short summary of the benefits of a probabilistic approach to forecasting. Check out our blog: Probabilistic Forecasting–a Primer to learn more.