Previously, automation typically processed data periodically in a batch rather than in near real-time. It was characterized by manual fallouts and hardcoded, inflexible rules that on occasion created inefficiencies. For example, Martini highlighted instances where older automation continuously switched services between two paths while attempting to balance the network, incurring unnecessary cost on multiple service oscillations.

Martini analyzes 3.5TB of data daily from real-time data sources like streaming telemetry and alarms, and batches, such as call data, that include more than 10 types of data and sources. Its analytical model has identified major service outages that before were masked by normal user behavior. For example, an outage impacting broadband services in the core network was severely restricting traffic’s flow through a port. As this was in the early hours of the morning, low levels of traffic are the norm and previously would not have been detected.

The model also automatically generates a case in less than half the time taken before Martin was deployed and by tracking deviations in ‘calls attempted’ and call connected’ data picks up from the baseline profile for voice services, as well as anomalies in the ratio of ‘attempted’ to ‘connected’ calls.