Cover story
Cover story
Testing times
Capex and opex are Number 1 priorities for operators nowadays. How is T & M helping their bottom line? Is AI making a difference to diagnosis, even self-healing? Colin Mann sought the views of industry experts.
Given the learning abilities of AI, will this mean far more automated diagnosis and self-healing? Accedo: Yes, with important nuances. AI learning is fundamentally transforming test & measurement from traditional reactive monitoring to proactive, self-healing systems. AI can recognise e patterns that precede failures, identify root causes faster through correlation analysis across multiple data sources, and predict likely failure scenarios based on historical data. This predictive diagnosis capability enables us to intervene proactively rather than react after problems occur. With this self-healing approach, we have moved beyond simple diagnosis, to be able to take necessary and appropriate action. Aprecomm: GenAI is rapidly reshaping the customer support landscape, driving new levels of efficiency, personalisation, and scalability. AI-powered chatbots and virtual assistants are now handling routine inquiries with speed and accuracy— responding instantly to common questions and issues, reducing wait times, and freeing human agents to
8 EUROMEDIA focus on more complex customer needs. But the real power of AI goes beyond reactive support. It lies in preventing problems before they occur— and resolving them automatically when they do. Through advanced machine learning, AI systems can detect and correct common connectivity issues in real time, ensuring a seamless experience for customers. This proactive approach is already delivering measurable results. Bitmovin: Yes, AI is already transforming how testing and monitoring systems detect and respond to issues. Traditional T & M relied heavily on manual analysis and predefined alert thresholds, which often meant problems were only discovered after they had already impacted viewers. With machine learning and continuous data ingestion, monitoring platforms can now identify subtle behavioural changes that indicate a potential fault before it escalates. For example, AI can recognise that a sudden rise in buffering events on a specific device model or region may point to a player configuration issue, while an increase in failed segment requests might indicate a CDN or delivery path problem. This level of pinpointed insight enables real-time diagnosis and allows thresholds to be set for automated remediation, creating the foundation for genuinely self-healing streaming environments. Bridge Technologies: There is certainly potential in that arena, and no doubt automation will play an increasing role in network diagnosis and recovery. But Bridge Technologies’ position is not to rush in: inevitability doesn’ t mean immediacy. AI will absolutely find its place within monitoring, but we believe that foundations come before fanfare. Without integrity of data, AI-led decision-making risks compounding errors rather than resolving them. The groundwork needs to be right first- coherent, synchronised, and truth-telling data across the entire chain- and that’ s where our focus lies. Self-healing systems are only as intelligent as the inputs that guide them. Leader Electronics: Artificial intelligence and machine learning are already helping engineers identify trends and anomalies that would have taken far longer to detect using traditional methods. In the future, we see T & M systems evolving toward assisted and, in some cases, autonomous diagnosis, where the instrument can recommend corrective actions. Live production remains highly time-sensitive, so human oversight will continue to play a vital role in validating system behaviour before any automated correction is applied. Telestream: Yes, and this is nothing new. Telestream monitoring products have used ML models for years to perform quality analysis on video without a reference image for comparison. Additional diagnostic, root cause analysis, and predictive models are also at the forefront of our product direction. Torque Video Systems: I suppose so, but it greatly depends on economies of scale. Properly training an AI to be useful requires tremendous amount of relevant training data. Estimates for GPT-5 range from $ 1 billion to $ 2 billion. Of course, a simple diagnosis model on a constrained subject does not need that much training data, but each input data must be tagged with the particular type of problem. No-fault data must also be input to the trainer so that the model learns what proper behaviour looks like. Finding the training data( and getting enough of it) is expensive and time-consuming. Furthermore, the complexity