Where do you want to go?
- How We Help
- About Us
Bullhorn is one of the leading providers of applicant tracking and customer relationship management software globally. They deliver a stream-lined platform that enables recruiting firms and corporate HR departments to manage the entire recruitment and applicant management process. Low barriers to entry make it easy for small recruitment firms to switch software platforms or abandon the recruiting business altogether.
How can we use advanced analytics to forecast which customers are likely to unsubscribe to a software as a service product?
With this notion, Bullhorn is experiencing relatively high customer churn of their flagship applicant tracking software. The cost associated with replacing a churned customer is much higher than retaining an existing customer. Between system setup, integration, and training, there are a lot of resources used up front which results in a loss, or at best a marginal gain, if a customer churns quickly.
When Bullhorn was having a difficult time getting a hold of their churn problem internally, they turned to Axis Group to help provide them an operational solution.
Axis group sought to have a deep understanding of the client’s business operations, and through experimentation and iteration learned what features are driving customers to churn.
As an initial step, Axis met with and interviewed Bullhorn’s key business users and account managers to better understand the problem at hand and the data available to solve it. After gathering information and insights from the internal teams, Axis began mining through multiple datasets to uncover patterns that may be indicative of customer churn. This allowed Axis to dial in to the most important features that drive customers to cancel their subscription. Axis then experimented with various machine learning algorithms to predict whether a customer was likely to churn in the near future.
Axis Group honed in on the highest-performing machine learning prediction model and built an automated production pipeline to predict which customers are likely to churn in 30 days.
After experimenting with various modeling techniques, Axis group selected and pre-trained the model with the highest predictive accuracy on unseen customer data. The chosen model was able to correctly predict the churn customers 76% of the time and the non-churn customers 89% of the time. Ultimately, the model produced two metrics to help Bullhorn target the customers at risk of churning. The first was simply a yes or no prediction indicating that yes a customer is likely to churn or no they are not. In addition, the model produced a probability of each customers likelihood to churn. For example, customer A has a 92% probability of churning, while customer B only has a 32% probability of churning. Churn prediction are made at the beginning of each month to indicate whether a customer is as risk of churning in the following month. Subsequently, a production pipeline was built and employed to automate the churn prediction each month, without the need for human instigation.
Actionable recommendations if a customer was likely to churn
Share this story