Overview
More and more businesses are beginning to leverage predictive analytics to increase revenue, reduce cost and improve service to customers. However, many others are not sure how to get started or believe they need to hire a new department of data scientists. With these unknowns determining the true ROI of implementing predictive analytics can prevent many companies from even trying. All companies can benefit from predictive analytics to improve operations.
To get started, determine a proof of concept use case that can be completed in 12 weeks. This POC enables the business team to understand the analytics process, outcomes and how to implement the model within operations.
Machine Learning
There are two types of machine learning algorithms used to create predictions, Supervised Learning, and Unsupervised Learning. Supervised learning makes predictions using past information and known outcomes. For example, predicting part failures by tracking historical part failures. Unsupervised learning makes predictions from only the data it has, no historical or known outcomes are known. An example of this is market basket analysis where all purchases made for the past year at a supermarket are analyzed to determine what products are frequently purchased together.
How does Predictive Analytics Work?
Predictive analytics leverages patterns and relationships in data by using historical data applied to current information. The core of predictive analytics is the data mining model. Data mining models are built using various machine learning algorithms that attempt to discover patterns, trends, and behaviors in the data. Using the best performing model for the individual use case, new insights can be gained by applying the model to new data. The following use cases explain how predictive analytics can improve business operations.
Use Case – Predicting Credit Risk Default
You can predict credit risk default by using a supervised learning algorithm that leverages historical data from all customers. A model can be built that scores each customer and provides the probability that they will default on their credit. The results of the model can be shared with the operations team to do further investigation and determine what actions to take to minimize defaults.
Use Case – Predicting Customers Most Likely to Churn
You can predict the customers most likely to churn by using a supervised learning algorithm that leverages historical data from all customers. A model can be built that scores each customer and provides the probability that they will churn or stop their service. The results of the model can be shared with the operations team to do further investigation and determine what actions to take to minimize churn and offer customers different products that lead to higher income.
Use Case – Predicting Machine Failures
You can predict which machines are most likely to fail by using a supervised learning algorithm that leverages historical data from all similar machines. For example, using the sensor data from a well pump. A model can be built that scores each pump and provides the probability of failure for each pump. The results of the model can be shared with the maintenance team to do further investigation and determine what actions to take to minimize downtime.
Use Case – Predicting Products that Sell Together (Market Basket)
You can predict which products are most likely to be used by customers using an unsupervised learning algorithm that leverages data from all customers. For example, using the product information from a financial institution such as saving account, checking account, ATM usage. A model can be built that determines which products customers commonly used by all customers. The results of the model can be shared with the product team to create new offerings and marketing efforts that get more customers leveraging.
© Derek WilsonCEO – CDO Advisorswww.cdoadvisors.com