When a customer encounters a service problem that cannot be resolved by regular triage with a call or chat agent, a fulfillment truck roll is scheduled. A standard technician, who specializes in repair issues located inside the premise and/or related to the service drop, performs onsite troubleshooting. If they discover an issue within the broader service network typically impacting multiple customers, they escalate the problem to a line technician. Line technicians expertise is in repairing and maintaining the Outside Plant (OSP) network, such as the nodes, amplifiers, passives, and hardline cables that provide service to multiple customers. This two-step troubleshooting practice incurs the cost of dispatching both types of technicians when an OSP network maintenance impairment is determined to be the root cause of a customer’s service issue. More importantly, the follow-up escalation often delays the problem resolution for the customer.
In this paper, we will show that resolution efficiency can be improved by harnessing machine learning (ML) to predict when a customer’s service issue has a high likelihood to be escalated to the line technician. Our approach leans on network and device telemetry as well as monitoring processes that assess the integrity of our service network (such as checking for outages, impairments, and performing problem segmentation). A key component of our ML modeling is the integration of graph features derived from network topology, that help identify issues related to equipment within the network that serves multiple customers. This model is in the trial stage and is being tested by selected regions. We also discuss how we can evaluate model precision to incur savings for implementing the model. Finally, we describe how the model will be integrated into the internal troubleshooting software.