Inventory management gets complicated when a company keeps numerous SKUs with unique product characteristics. This is especially true for medical device companies. The products they carry are of high value, and service level is critical since patients’ lives are involved. Consequently, companies in this market often rely on conservative inventory models that maintain excessive levels of inventory. This practice results in high inventory carrying costs.
The goal of our research project was to find the right level of inventory to maintain excellent service levels. To achieve this goal, the key is to understand demand and supply using lots of data. We demonstrated how this information helps improve inventory management, based on data from MedCo, a leading medical device company.
In addition to helping one of MedCo’s regional distribution centers set the right amount of inventory, the research findings also yielded insights that help further improve the company’s operational performance.
Most companies design replenishment policies that assume demand is normally distributed, to simplify safety stock calculations. But this is not always the case. With the benefit of today’s information technology and data infrastructure, companies accumulate a pool of data from their transactions, which allows them to thoroughly understand demand characteristics.
To start exploring the available data, we used clustering to extract insights from a noisy dataset. Through the data exploration process we arrived at two main findings. First, demand characterization should be aligned to performance metrics. Second, our model needed to use the actual distribution found in the demand data, instead of assuming normality.
Considering both the relevance and data availability, we chose the distributions of order quantity, order frequency and replenishment lead time to characterize demand and supply. These inputs were used in a simulation model to determine the right inventory strategy.
The simulation model we built is straightforward, but has a powerful advantage: it directly links the way service performance is measured with the way inventory level is determined.
The model determines the inventory level that satisfies the target service level at least 90% of the time. That is, the recommended inventory level lies at the 90th percentile of the distribution.
We ran the simulation program with a sample of material numbers chosen from different ‘clusters’. The results varied. In some clusters, the program recommended a higher level of inventory to hedge higher implied risk. In others, it showed that there is excess inventory that is not required.
By applying this technology to a sample of material numbers, we showed that there are huge cost savings and risk avoidance opportunities across multiple SKUs. MedCo can use these results and adjust their inventory plans accordingly.
No two supply chains are identical; therefore, inventory reduction plans should be targeted to each companies’ product and process features. In MedCo’s case, we showed that most unfilled inventory requests were caused by demand peaks that were not fully anticipated and depleted the stock in the DC. Curiously, these peaks are caused by a few customers. MedCo can segment their customers by ordering pattern, and influence buyers by providing accurate forecasts or smoothing the reordering patterns.
Another major challenge many supply chains face is the management of critical but long-tailed SKUs. In such cases, it is difficult to establish a reliable forecast due to insufficient data. After considering MedCo’s product and process characteristics, we suggested that the company should stock spare inventory in conjunction with postponement strategies at manufacturing sites, and use air transportation to move product when orders are received. Thus, MedCo can reduce the risk of demand uncertainty while improving service levels.
This article was written by Xiaofan Xu and Maria Rey, and is based on their MIT SCM Master’s thesis Simulating Inventory Versus Service Risk in Medical Devices. The thesis was supervised by Dr. Omar Sherif Elwakil, MIT Center for Transportation & Logistics. For more information on the research please contact Bruce Arntzen Executive Director, Supply Chain Management Program, at: firstname.lastname@example.org.
The article was first published online by Supply Chain Management Review.