Case Study 1 (cont.): Transitioning to Stochastic Planning

The new probabilistic solution brought about Four key changes to the client's supply chain:

1.Enabled Detailed Demand Modelling: Demand models contain highly detailed stochastic information, preserving full variability and volatility details—not just about the demand quantities, but also the order-line frequencies and quantities per order-line. This is the building block for Demand Sensing, MEIO, and Multi-echelon Replenishment Planning, achieving end-to-end synchronization and zero planning latency.

2.Enabled Demand Sensing: Unlike Timeseries methods, Stochastic Demand Sensing transcends quantity-based forecasting by employing line-order level details. It models influencing factors and their impacts at the most granular levels, providing short-term projections. This approach resulted in reduced errors, improved inventory accuracy, and optimal deployment of downstream inventory, such as in distribution centers, 3PL and sales depots.

Utilizing downstream data such as customer, point-of-sale (POS), and channel data allowed the client to discern demand trends, detect signals early, and eliminate the lag between planning and real-time supply chain activities. Rapid identification of deviations enabled swift and intelligent responses. Going forward, in the next phase, Demand sensing can also incorporate a broader range of demand signals and respond to real-world events such as market shifts, weather changes, natural disasters, and consumer buying behavior etc. more efficiently.

Generally, there are two ways to handle eventualities in supply chains. The first, “prepare” is a proactive approach, where many demand-supply behaviors are anticipated in advance. Statistics and their predictability play a significant role in this so called "Upstream Proactive Planning". However, this is difficult when using deterministic methods. When statistics become weak, the focus shifts to supply-side tactical measures, which are reactive to eventualities, known as "Downstream Reactive Planning". Both approaches are required to achieve the right results.

To get upstream planning right, a thorough understanding of demand at the order line level is essential. This understanding is also crucial for primary-level Demand Sensing, which serves as the transition zone between upstream (proactive) and downstream (reactive) planning. Enabling this transition is crucial in developing cohesive response strategies.

3.Enabled Probabilistic Inventory Optimization (MEIO): Previous Inventory Planning relied on simple ABC classification. This was based on internal operational or logistics perspective. There was no customer centricity. Sales, marketing, business, or customer perspectives were never considered in deciding ABC segments.

In the new solution use of “service class,” was introduced. This is the grouping of similar products like Oncology, Respiratory, Rheumatology, Critical Care product groups etc. This type of categorization is much more relevant to the business and its markets. It is at this “Service Class” levels aggregated service level goals were defined, instead of defining service levels to ABC classes.

At the core of inventory optimization is reliable "stock-to-service" models, for each SKU-Location individually. Unlike traditional ABC matrices, this method treats each SKU-Location comprehensively, considering factors such as demand/order line variations, lead times, and other operational constraints. With “stock-to-service” models dynamic differentiation of service levels within defined limits ensures flexibility, enabling automated management of vast and complex supply chains with optimized stock investment.

By using “stock-to-service” (STS) models, software optimizes service level and safety stock level of each SKU-Location, through mix optimization. Employing the concept of marginal contribution, the software automatically distributes service levels among competing SKU/L’s. within a Service Class. In one variation of STS, the overarching service class goal is attained with the lowest possible stock investment. In another scenario, the focus may be on maximizing net margin, maximizing residual shelf-life, minimizing obsolescence risk, among other objectives. This approach ensures that each SKU-Location is assigned a customized service level for calculating safety stock levels, aligning with specific business objectives.

The automated differentiation of service levels in each service class was set within defined limits. As an example, the aggregated service level goal for “Rheumatology” could be 93% with a lower limit of 89%. The inventory optimization can then subscribe any service level from 89% and above in a way that the stock investment is minimized. In another example, “Critical Care” could have a goal of 99.5% with a lower limit of 95%.

Despite improving service levels to 99 percent at their 3PL distribution sites, inventory levels have been reduced by up to 15 percent. On this measure alone investment will have paid off in the same year of implementation.

4.Multi-echelon Replenishment: The new solution created a service-model hinged on three key elements: Demand Sensing, Probabilistic Inventory Optimization (MEIO), and multi-echelon replenishment planning. These elements, working in synch and in a tightly integrated manner, ensured accuracies throughout the supply chain and delivers optimal service levels, ongoing basis. These service-driven stochastic models stood apart from the previous deterministic planning where monthly quantity forecasting, ABC inventory norm setting, and single-echelon supply planning were used to drive the supply chain.

With multi-echelon replenishment planning, diverse demand-supply status vectors are reconciled to generate predictive and actionable proposals. These proposals also factor in dynamic safety stock, order-up-to levels, and reorder levels modeled on an ongoing basis to keep replenishments in the optimized zone. When executed, these proposals enable the supply chain to attain target service levels through the optimal allocation and deployment of inventory.

Replenishment planning involves processing Stochastic Demand Signals across the Network. Propagating not only demands and dependent demands upstream but also detailed demand profiles, including variability profiles related to demand quantities, demand order lines, and order sizes, at each node along with supply side uncertainties. Such profiling is conducted for each SKU-node combination, retaining the probabilistic nature of the demand propagated throughout the network. By implementing such detailed probabilistic profiling created a transparent view, end-to-end, that enables efficient and dynamic synchronization.

Outcome: In the case study discussed, typical benefits included improvements in service levels by 8% on average (from 90% to 98%), whereas the client never achieved more than 92% while using the old system. There was a 9% reduction in inventory at the end of the first year, and by the second year, there was an additional 5% reduction despite sales growth. They also experienced reduced lost sales and a 50% reduction in planner workload during the first year. By the end of the second year, manual interventions were required in less than 2% of transactional orders. Overall, the client achieved a positive ROI in about 14 months after go-live.

What Next: In the adaptation of outside-in refinement, machine learning is the main enabler and plays a pivotal role in enhancing probabilistic planning by analyzing vast amounts of external data. Factors like weather, social media impacts, and outside events can all be seamlessly integrated into the stochastic end-to-end model. This capability enables more accurate proactive and responsive planning, thereby supporting better-informed decision-making across inventory management, supply chain operations, and strategic planning.