Automatic Ticket Resolution in Telecom
How Daisho’s text mining module automatically identified categories of issues from tickets and streamlined resolution
What Daisho did
Daisho identified hundreds of categories of issues across multiple telecom operators from textual descriptions of issues in tickets. Daisho replaced the painful, manual and weeks long exercise of categorizing tickets with a fully automated approach that gave results in minutes.
The Problem - Sifting through a sea of tickets
Lots of applications run a telecom network. Some applications are customer facing, like apps on phones or portals. Others are part of the IT infrastructure which tracks servers, routers, switches, databases, storage, etc., at different points of customer management - like order processing, onboarding, billing, promotions and offers, upgrades, etc. All these applications generate tons of tickets and alarms - and they come with issue descriptions, either fed manually by one who raises the ticket (e.g. customer tickets on recharge or payment issues) or automatically generated but with lots of contextual information (e.g. ticket raised automatically due to circuit breakers).
For our partner, which works with large telecom companies around the world, managing tickets is an especially vexing problem because of the volume of tickets across business lines like wireless (voice, data, and internet), fixed line (voice and internet), and even satellite TV and dish TV. Our partner had two primary goals:
Once tickets are categorized, they can be resolved using automated standardized resolution workflows.
With patterns in anomalies that emerge, the customer will be better prepared for future changes. Evolution of ticket volume over time helps to allocate resources.
The Solution - Automatically categorizing tickets to bring burning issues to focus
Using Daisho’s text mining recipe, our partner was able to automatically categorize tickets, bringing big issues to focus for multiple telecoms.
For one telecom, 50 categories of tickets were automatically identified by Daisho -- these included tickets related to missing usage alerts by customers, SMS failures, issues related to billing, payment and recharge, login, etc. But categories contributing to the biggest volume of tickets were related to failures in order processing (component failures, failures in communication, node failures, port servicing and provisioning issues, etc). This was a major insight since failure to process orders results in loss of revenue.
For another telecom, about 20 ticket categories contributing to more than 80 percent of the tickets were automatically identified. While other categories of ticket were also identified (e.g failures in billing and revenue applications, degraded response times for certain operations, etc.), the top issues were related to mobile virtual private network operators (MVNO) e.g. delay in getting MVNO orders completed. This would point to technical and business communication gaps between the access provider and the MVNOs.
Similar approaches were also applied to service request tickets across multiple telecoms.
Daisho’s text mining recipe also identified when certain ticket categories showed anomalous behavior based on completely automated time-series models. For one telecom, it identified node issues related to application components had spiked during a specific week. It also identified specific times of the year when tickets from the mobile app related category spiked - correlating it with types of app updates, provides a more direct view of the types of updates causing more serious impact on customer experience.
From tickets to predictive resolution strategies
Based on automatic ticket categorization one can build automatic standard resolution procedures for each category, making a direct impact on the quality and cost of ticket resolution. Using historical data on how each ticket was resolved, Daisho can also predict the most effective way to resolve any new ticket.
Daisho’s text mining module also has forecasting insights to predict the expected volume of a ticket category in the next time period making it easier to plan resources ahead of time.
Often new and unexpected tickets also show up. Daisho’s text mining recipe automatically extracts keywords and flags them if they are completely new or based on expected occurrences from historical data. This serves as an early warning system for new issues in the network.
How Daisho categorizes tickets automatically
Daisho uses a class of models called topic extraction models in text mining to automatically extract categories of tickets from the textual description of the issue. The algorithm searches for groups of words which co-occur together to extract patterns and surface them as a possible category. It is completely automatic with Daisho taking into account stop words (automatically filtering very common context specific words which need to be excluded from the analysis) to give more insightful topics, reducing different words to their common roots, etc.