Often you will need a variety of data at various levels of detail and aggregation to build a model. For instance, to build a customer churn model you would need basic contract details like term, product, contract start and end date, as well as demographic data. This detail level information can then be mashed up with aggregated data such as number of missed payments, number of inbound calls, or number of responses to outbound campaigns. Sourcing the required data elements from your business intelligence solution can help accelerate building the model. Often the business intelligence database contains most of the information to get started. If the data is not in the proper format or the required data is in source applications, you can leverage the logic or queries in the business intelligence ETL processes. Using these queries can help you understand where data is coming from and what transformations have been done on the data. A key to any successful data mining model is ensuring you understand the data and quality of the data that is being leveraged.

Applying the Models to Create Predictions

Once you have a working data mining model you can move from business intelligence to predictions. In this stage, the model is applied to your operational data. For example, a customer churn model would be applied to active customers. Each customer‘s probability of churn would be calculated by comparing the individual’s specific data to the model. Once the probability is assigned, operational processes can be created on modified data to utilize the new information. In the sample below the confidence (Churned) column shows the probability of each of these customers’ churning.

This information can be saved in a database and leveraged to improve your operations. CRM applications can integrate this information to give call center agents more insights into each customer.

Another example is creating marketing campaigns based on predicting the customers who are most likely to accept an offer. Once the model is built, apply your contact list to this model and each customer will be scored with a probability to accept the offer. In the example below, each customer is given a probability of responding to the campaign.

This will decrease your expenses and increase your conversion rate by matching the offer to the right potential customer.

Getting Started

To begin using predictive analytics in your operations, start by defining business problems where you have a good understanding of the data available. Then build out models to see what patterns or behaviors can be found in the data. Once a model is developed and run against your production data, ensure you have business processes ready to use the output from the predictions to improve your business. Data mining and predictive analytics can give your company an edge to improve revenue, decrease expenses or improve services.