Predictive analytics

Alfred Sherk discusses new developments that can help enhance Lean capabilities.

The application of predictive analytics to extend the use of the Lean pull process into complex manufacturing processes offers great potential for improving the responsiveness and efficiency of global pharmaceutical manufacturing chain operations. Predictive analytics are substituted for the Kanban method used in assembly operations to provide the demand signal for the pull process.

Predictive analytics is the application of statistical data analysis to capture relationships used to predict future trends and behaviors. Credit card companies use predictive analytics in sophisticated fraud detection systems and leading retailers and e-tailers use predictive analytics to suggest additional items that may be of interest to the customer based on past behaviors and demographics. In process manufacturing, Rockwell Automation uses predictive control models that enable process systems to adjust more quickly and accurately than with actual measurements. Applying these predictive control techniques significantly improves performance.

The Lean pull process dramatically improved Toyota and Dell's assembly operations by eliminating dependence on inaccurate forecasts. The Lean pull process uses firm orders as a demand signal and propagates it along the production chain. However, manufacturers with high structural complexity require reasonably long production runs to achieve acceptable production economics. Accordingly, production-scheduling decisions need to be made before a complete demand signal from customers is available. In such environments, it had been assumed that the Lean pull process could not be used.

SherTrack's research on-demand patterns has enabled the successful application of predictive analytics to the order-to-fulfillment process. SherTrack's predictive analytics algorithms provide an accurate, synthetic customer order signal in the operational time horizon of the next two to three production cycles, enabling application of the very efficient Lean pull process in operations where it was thought too complex to use. 

The traditional approach to coping with poor demand visibility is to forecast demand, set inventory targets that incorporate safety stocks then optimize the production response. Unfortunately, forecast error renders this process incapable of handling the structural and dynamic complexity of multi-product facilities. Across all industries, best-in-class forecast error (such as demand variation) exceeds 40 percent mean absolute deviation, which means that while forecasts are appropriate for tactical planning issues, they really can't be effectively used in execution. Most advanced planning and scheduling (APS) solutions were developed to meet the needs of the production planning process with finite scheduling capabilities added later. However, the needs for very efficient execution level order fulfillment processes are markedly different than those required for tactical planning. 

In complex production processes, the most difficult constraints to resolve are the needs of competing products for the same production resource. The dynamic fluctuations in demand for individual products stress forecast-based processes beyond their capability limits. The most common operational response to complexity is to carry more inventories and to require extended lead times. Long end-to-end throughput times and operational complexity are critical barriers to operational excellence.

The combination of predictive analytics with innovative probabilistic inventory and scheduling solvers provide effective control of complex manufacturing, enabling performance close to the physical process limits. New predictive modeling capabilities use proven data based decision support methods to quantify process capabilities and identify improvement opportunities critical for attaining operational excellence. This approach also provides the ability to balance customer service, inventory levels and production costs to enhance operational excellence and reduces operational risk.

Today, innovations in predictive analytics and probabilistic inventory and scheduling solvers provide addresses demand visibility to enable deployment of the Lean pull order-to-fulfillment process in the pharmaceutical industry. In industries with similar characteristics, innovative manufacturers have recently leveraged the new hybrid of Lean pull and probabilistic algorithms to help transform their business processes and gain competitive advantage.

Alfred Sherk is the founder and CEO of SherTrack, provider of innovative, predictive analytic solutions for synchronizing supply with demand. He is a past member of the Technical Advisory Board for Michigan State University's Graduate School of Business and is a limited partner in North Coast Technology Investors LP.