100% Automation of Support Tickets Processing in ServiceNow for a Large Construction Company

Goal: To fully automate the processing of support tickets in ServiceNow    

Solution: Developed and deployed ML models to accelerate ticket classification and assignment    

Benefits: 100% automation of processing the financial, IT, and equipment rental requests    

Services delivered: AI/ML, Intelligent Automation, Custom Software Development

Infopulse developed and implemented a hyperautomation solution for the client to optimize the manual processing of tickets in ServiceNow. Namely, we deployed a series of Machine Learning (ML) models to fully automate the classification and assignment of financial, IT-related, and equipment rental requests. This resulted in a 6X reduction of effort the HelpDesk team had spent earlier for their processing and faster overall ticket resolution.

Applying its extensive AI expertise and working closely with the client’s subject matter experts, Infopulse initially achieved 82% ticket classification accuracy. However, by establishing automated re-training ML models, the system reached 96% ticket classification accuracy in a year. This ensured little-to-no human intervention and optimized the Help Desk team’s workflow, redirecting their effort and time to more high-value tasks.    

Here are some improvements the construction company received from implementing this ML-powered solution:

• 90% reduction of administrative effort and ticket dispatching
• 15% reduction of MTTR (Mean Time Taken to Resolve) for ServiceNow tickets
• 35% increment of FCR (First Call Resolution) rate
• 20% reduction in ticket bouncing
• Higher data quality
• Faster ticket resolution

Case Study in Detail 

When the client first reached Infopulse as part of Tietoevry, their main goal was to reduce the Help Desk workload, automate routine support tasks, and improve the overall ticket resolution time. The Help Desk team spent all its effort and time, manually sorting out incoming ServiceNow tickets by 60 classes and dispatching them to an accountable employee. The faster the process of classification and dispatching goes, the faster the request is resolved, and the more optimized business operations are.     

Infopulse suggested applying machine learning for ServiceNow ticket classification, which would result in 96% ticket classification accuracy due to further automated re-training of the machine learning model.    

After the developed ML model for financial requests showed its high accuracy and efficiency in production, Infopulse proceeded to automate IT and equipment rental requests. However, because of contextual differences, the other two ML models were also custom-built from scratch, using other approaches to their development and training.         

What Infopulse did to serve the solution:

• Analyzed historical data of ServiceNow tickets and studied their context leveraging the domain expertise of the client’s subject matter experts
• Created and trained ML models for automated ticket classification using ensemble learning model, logistic regression techniques, and such technologies as Google BERT and LightGBM
• Provided a rule-based classification of ServiceNow support tickets for over 60 classes within 3 categories
• Automated dispatching of ServiceNow support tickets to specific assignment groups
• Automated daily training of the ML models with new data
• Developed a simple web user interface using Azure services for the Help Desk to review and correct the classification results if manual intervention was needed
• Integrated ServiceNow API to enable reading and updating of tickets by the Help Desk via a single web dashboard.

Azure integration

Since the client already had the subscription for Azure Cloud Services, Infopulse, as Azure Expert MSP, cost-effectively integrated Azure capabilities:

• When training a model, we used Azure VMs
• Azure MS SQL as a database for tickets training
• Storage Account for storing the files of the ML model
• Azure Function for prediction scheduling
• Azure Application Insights for monitoring the deployed ML model in production
• Azure Front-end and Back-end App Service for building a web user interface
• Azure Active Directory for user authentication.