One of the most important parts of retail stores today is optimizing inventory. If you have too much inventory, you’re losing money on the items that you have not sold. If you don’t have enough, you are losing money and confidence from your customer. Finding the perfect balance of inventory by predicting demand is a problem that machine learning can help you solve.
Imagine that a large demand spike in cleaning supplies causes the inventory to be exhausted days before new shipments arrive. Machines can learn from this situation. By using this data to train a machine learning model, you can predict demand for certain items more accurately in the future, and ensure that your customers are able to purchase what they want.
Using this scenario as the basis for this case study, you’ll take the view of the procurement manager, Bob. Bob is notified by Lauren, the retail store manager that the inventory for certain cleaning supplies has been depleted days ahead of schedule. Bob gives the task to the development team to take the past demand data and train a machine learning model to predict future demand. The model predicts the demand to optimize inventory and minimize procurement costs. After predicting the demand, that demand is used as an input to the optimization problem. The optimization problem solves the problem of which plant to order items from to minimize costs. This article gives an overview of the use case and explains how the development team will use machine learning tools and techniques to solve these problems.
Predict future demand using IBM SPSS Modeler
Michelle, the data analyst takes on the task of building a machine learning model using IBM SPSS Modeler, which is available as part of IBM Watson Studio on IBM Cloud and IBM Watson Studio Premium for IBM Cloud Pak for Data. After she builds a model, she uses the model to predict future demand for specific products in the retail store. Michelle visualizes the demand and sends the output to Joe, the data scientist, to use as input to his decision optimization model.
Use the Predict future product demand using SPSS Modeler tutorial to see a step-by-step approach of building a machine learning model using a flow-based editor.
Create a machine learning model to optimize plant selection based on cost
After Joe receives the predicted demand from Michelle, he uses that as an input to the decision optimization problem, along with the cost and the capacity of the plants that produce the items that he needs to replenish. Using the IBM Decision Optimization engine, Joe is able to find the optimal combination of warehouses to select to minimize procurement costs while still replenishing the inventory as suggested by the estimated demand. The Decision Optimization feature is available as part of IBM Watson Studio on IBM Cloud and IBM Watson Studio Premium for IBM Cloud Pak for Data, which supports multicloud environments.
Use the Optimize plant selection based on cost and capacity with Decision Optimization tutorial to see a step-by-step approach of building and deploying a decision optimization model using a UI-based modeling assistant.
Create a web application for the procurement manager
Now that the Decision Optimization model is available, Joe creates a web application. The input to the Decision Optimization model is the demand generated from the SPSS model, and the SPSS model runs periodically to get the latest predicted demand. The user can input their demand and plant costs and capacity and get a result from the application. The result is the quantity of items to order from each plant in order to fulfill demand and minimize costs.
Use the Create a web application to optimize plant selection based on cost and capacity code pattern to see how to build a web application that accesses a deployed decision optimization model through an API and displays the results for the manager to use.
In this case study, you’ll learn how a development team can help a procurement manager by building machine learning models to predict future demand and develop an optimal procurement strategy. The manager is able to make data-driven decisions in seconds by using a web app that is enabled by machine learning models. The manager can be confident that he is making the best decision possible with the data at his disposal, and that his company is able to serve its customers well and increase profits. As demand changes, new data is added to the machine learning model, and the model is retrained to ensure accuracy.