Date Published: June 26, 2017 - Last Updated 5 Years, 3 Days, 23 Hours, 32 Minutes ago
Advances in data mining and predictive analytics are enabling companies to gain a deeper understanding of their customers. By leveraging these techniques companies are able to create focused marketing campaigns, understand likelihood of a customer to churn or create better customer segmentation. Companies can also use these same techniques to better understand their workforce. The effective application of data mining techniques in a contact center can decrease agent turnover and improve working conditions, or insights to find the best candidates. Applied data mining and predictive analytics can highlight what traits make a high performing employee as well as the traits that lead to employee turnover. Initiating a predictive analytics project to predict agent turnover starts with reviewing available data.
Getting Started by Selecting Data
Start with the data in your organization regarding your agents’ demographics and performance. Data mining models can be built utilizing data from HR systems, agent performance information, work schedules or any other data available. For this process to be successful you need to know which agents are still employed and which agents departed along with why they left (voluntarily or involuntarily).
Discovering Patterns in the Data
By combining various pieces of information from these sources, a model can be built. The model will discover patterns or relationships in the data that are not able to be seen through traditional reporting or business analysis methods. This step in the process will take the longest. It requires experimenting with different combinations of data and algorithms to create a data mining model that performs the best to predict results. Each type of algorithm has its own evaluation criteria to determine how effective the model is performing. The evaluations apply sample data to the model to test how many predicted answers correctly match the actual result and how many are misclassified. Once a working model is created it can be applied to your active employee agents. The active agent’s data points will be input and applied to the model. The results for each agent can be output and classified with a probability of churn percentage and a flag indicating if an agent is likely to Churn.
How to Leverage the Prediction
In the sample output shown above each row represents an agent and their likelihood of churn between 0 and 100%. The column confidence(Yes) on the results shows Agent 342 has an 83.9% probability of churn in this predictive model. Using this information, a manager can know which agents are most at risk of quitting. In addition, the model will provide the significant factors that determines the confidence of churn for each agent. For example, the model could show that agents who work 2nd shift and live over 20 miles from work and do not receive overtime are most likely to churn. This is just one of many combinations of information that the model will be able to provide. Managers will be able to utilize this information to maximize their workforce. If you have high performing employees that are at risk of churn. You can apply the information from the model to modify their work environment to retain top talent.
Modifications could be as simple as a single change or combination of changes such as switching shifts, changing roles, moving to a new team or increasing hourly rates. Could other changes be made to reduce churn such as disciplinary policies, break frequency or length, or increased training in certain areas. The information provided by predictive analytics will enable the appropriate changes to be made that can amplify the impact to your team.
Using Predictive Analytics
Once a predictive analytics model is built and ready to be used. A process needs to be put in place to update and maintain the model along with the results. As changes are made to the workforce based on information learned through the model updates to the model may be necessary. For example, if you change the frequency of breaks for agents. The model will interpret that change and its impact on the model, however, it could be many weeks to months before the change does have an impact. As you utilize models to improve operations be aware that updates need to be made to ensure the best performing model is available for decisions.