One of the most important parts of being a retail store today, going against the likes of Amazon, Etsy, and other online stores 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 can be solved using machine learning.
Imagine that a large demand spike in cleaning supplies causes our inventory to be exhausted days before new shipments arrive. Machines can learn from this. Using this data to train our machine learning model, we can predict demand for certain items more accurately in the future, and ensure that our customers will be able to purchase what they want.
Using this scenario as the basis for our case study, we will take the view of the procurement manager. The procurement manager, Bob is notified by Lauren, the retail store manager whose inventory for certain cleaning supplies has been exhausted 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 will predict demand in order to optimize inventory and minimize procurement cost. Once we've predicted demand, we will use that demand as an input to our optimization problem. Our optimization problem will solve the problem of which plant to order items from in order to minimize cost. In the next few paragraphs, we will explain how the development team will use machine learning tools and techniques to solve these problems.
Michelle, the data analyst takes on the task to build a machine learning model using SPSS Modeler on IBM Cloud. 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 to use this predicted demand 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.
Once Joe receives the predicted demand from Michelle, he uses that as an input to the decision optimization problem, along with cost and capacity of the plants that produce the items that he needs to replenish. Using IBM's Decision Optimization engine, Joe is able to find the optimal combination of warehouses to select in order to minimize procurement cost while still replenishing inventory as suggested by the estimated demand.
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 UI-based modeling assistant.
Now that the Decision Optimization model has been made 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 cost and capacity, and get a result from the application. The result is the optimal plant strategy - which plant to order from, and the quantity of items to order from each plant in order to fullfil demand and minimize cost.
Use the Create a web-application to optimize plant selection based on cost and capacity to see how to build a web-application that accesses a deployed decision optimization model via API and displays the results for the manager to use.
In this case study, we've seen how a development team can help their procurement manager by building machine learning models to predict future demand, and an optimal procurement strategy. All in all, the manager is able to make data driven decision in seconds, by the use of a web-app that is enabled by machine learning models. The manager is confident he is making the best decision he can with the data at his disposal, and 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 re-trained to ensure accuracy.