The extensive availability of data along with recent advancements in machine learning tools and techniques have provided unprecedented opportunities for service parts planning organizations around the globe.
At Dell EMC, for example, the service parts planning organization has significantly improved its processes in order to have parts available at the right time and in the right location across the globe to meet the customer demand – meaning the request of replacing a defective part under a service agreement.
Autonomous Planning (AP) is one of the principal machine learning projects in service parts planning. AP intends to reduce human intervention on the entire parts planning process - from forecasting the future demand to approving the order plans - by automating where it is feasible to do so.
Main Components of Autonomous Planning
AP involves multiple subprojects, each using historical information and various machine learning and operations research methodologies to address a unique problem in the planning process. However, from a strategic perspective, AP has the following three major components that are critical to achieving the most benefit:
- Forecasting: This focuses on augmenting the capability of projecting the future demand of parts in each location for various time horizons. Forecasting enables companies to harness the variability of future demand by learning the behavior of demand in the past. Therefore, a high-quality forecast enables an excellent customer experience while maintaining an efficient parts supply across the globe. The high variability of demand and various forecasting types—i.e., forecasting for high volume and sporadic demand, and short/medium/long term forecasts—require an extensive amount of research and effort to enhance the forecast quality. To improve the future demand forecast accuracy, organizations should focus on innovating and applying various forecasting methodologies. These include conventional time series analysis, Monte Carlo simulation, and other statistical and mathematical approaches to improve the future demand forecast accuracy.
- Inventory Optimization and Distribution (IOD): This is the heart of AP and generates inventory plans based on the inputs from the forecasting process and considers many other variables. The ultimate goal of IOD is to optimize inventory and minimize expedited shipping costs while maintaining high service levels. Due to the complex network structure of inventories in many services organizations, traditional inventory management techniques fail to provide the best solution for their inventory plans. Therefore, more powerful methodologies should be explored. For IOD, services organizations should pursue mathematical modeling and multi-echelon network optimization tools and techniques to enhance inventory planning.
- Auto Execution: The ultimate directional step in AP is auto execution and involves answering a binary decision-making problem: whether the generated plan is robust enough to be passed to procurement and operations or if it requires further reviews from planners. Today, in most cases, once a plan is generated by the planning tool, planners are responsible for approving, rejecting or adjusting the plans based on experiential knowledge. Auto execution intends to not only mimic the planners’ behavior but also take the accuracy of this decision-making problem to the next level. This requires both comprehensive statistical analysis of all the potential variables affecting the accuracy of the decision-making process and thorough domain knowledge, which is gained by interaction between data scientists and experienced planners. In auto execution, services organizations should pursue various supervised machine learning techniques to identify important variables, build predictive models and ultimately enhance the decision-making process. As the forecasting process and IOD become more accurate and efficient, it is expected that the ability to systematically and automatically execute the order plan will drive efficiencies within the organization.
Benefits of Autonomous Planning
Once these components are understood and addressed, there are three main advantages organizations will experience with AP:
- Increased Service Levels: As companies better understand and capture variability of future demand through forecasting, they can predict customer behaviors more accurately and meet their demand with a higher level of confidence.
- Optimized Inventory Levels: Companies realize that optimizing inventory levels does not necessarily mean decreasing inventory costs, but in some cases, using AP to find a better way of utilizing the inventory to increase service levels. IOD works to obtain the optimal plan with respect to inventory and expedited costs subject to a required service level.
- Improved Planners’ Efficiency: Auto execution equips an organization with a powerful tool that allows demand planners to shift focus to more complex issues and improve organizational efficiency.
New Machine Learning Methods Address Large, Complex Projects
The advent of machine learning has empowered service parts planning organizations to make the planning process more efficient, fluid and automated.
However, due to the large scale and complex projects involved in this process, multiple phases are needed to reach ideal Autonomous Planning. For many services organizations, this complexity has brought a timely opportunity to innovate new machine learning methods to address these challenges.