December 15, 2020

Resolve Support Tickets Faster with Predictive Risk Algorithms

Business Case:

Our client, a global technology company, receives approximately 15 million customer support tickets a year. With an overwhelming volume, it was impossible for support engineers to identify which tickets were at highest risk of breaching service-level agreement (SLA). As a result, high-risk tickets could go unresolved for weeks, creating a major backlog.

At the time of partnering with us, our client had approximately 10 million open support cases in their ecosystem. Our client needed a solution that enabled support engineers to resolve tickets faster.

Key Challenges:

  Automatically assign ticket priority  
  Accurately predict high-risk cases  

Our Solution:

We implemented a CatBoost algorithm that uses machine learning to prioritize tickets based on risk probability.

Using historical ticket data, we identified the categorical and numerical attributes that our algorithm would evaluate to determine risk probability. The identified attributes included ticket type, ticket severity, initial response time, total time in queue, SLA state, and customer satisfaction score.
After comparing numerous algorithms, we concluded that CatBoost (algorithm for gradient boosting on decision trees) provided the most accurate results. Compared to other models, CatBoost:

  Allows the most categorical attributes  
  Saves time with faster model loading (does not require time-consuming one-hot encoding and pre-processing)  
  Is available at a lower cost  
  Requires a minimal learning curve  

To launch the model, our team used data outliers to inform the algorithm. With CatBoost’s machine learning capability, the algorithm increases its accuracy over time.
Figure 2: Mobile Notification

To improve support engineers’ response time, we implemented an integrated web feature that sends high-risk ticket notifications to engineers’ cell phones.

Business Outcomes:

Our CatBoost algorithm implementation enabled support engineers to resolve high-priority tickets faster. The algorithm uses machine learning to accurately assign ticket priority based on comprehensive risk evaluation. Prior to our solution implementation, our client had a backlog of 10 million tickets. Now, support engineers can reduce their backlog while keeping up with incoming tickets. With our added cell phone notifications, support engineers can immediately respond to high-risk tickets and ensure SLA compliance. Overall, our solution has enabled our client to significantly improve customer satisfaction.


    Implemented a CatBoost algorithm that automatically prioritizes tickets based on risk probability
    Reduced ticket resolution time
    Added a cell phone notification for high-risk tickets

For more on customer support, check out how our custom Dynamics 365 portal improved our client's gaming support operations