A Cognitive Automation Solution for Responding to Service Desk Requests

We consider the problem of automating responses to service desk requests in the IT department of a company. Employees get support, when issues arise, by sending an email to a Service Desk team. Some of these requests are forwarded to relevant parties, including a software development team for second-tier support. Providing this function takes developers away from their project development tasks and this tends to delay responses and hence problem resolution. We use cognitive automation to determine the nature of email requests and respond to them thus reducing the time developers spend on troubleshooting and support. A Machine Learning classification model, trained on a labelled set of emails, is used to perform this task. We focused on two approaches, Support Vector Machines (SVM) and Multi-Layer Perceptrons (MLP). The SVM model achieved an 86% accuracy score and a 92% micro F1 score. This solution speeds up the process significantly and we show that this translates to a potential time saving of 8 weeks of developer man-hours and 70 weeks of additional user man-hours per year.