Shrivastava tried to demystify digital twins by arguing there isn’t one twin, but many – each tied to an outcome (parameter-change impact; configuration/ service-introduction impact; utilisation and pathing effects). He said the concept is maturing, but valuable today as a change-management confidence tool that augments engineers rather than replacing them. Dashboards and the “single pane of glass” also got a reality check. Bridge Technologies chairman Simen Frostad rejected the idea of adding yet another “intelligence layer” on top of the OSI model, arguing simplification matters more than new abstraction. Shrivastava was equally sceptical about the classic promise of a unified view: “We all talk about single pane of glass. They cannot be a single pane of glass ever. The glass is always broken or scratched.” Instead, he argued the future is dynamic, role-based views – potentially generated on the fly using AI models curated to the operator’s network “truth” (with the ability to save or discard views) – while accepting that confidence scoring and verification remain essential. Preventative maintenance and cost reduction were framed less as “fix faults faster” and more as better capacity, performance and commercial insight. Kingdom suggested operators need clarity on what they truly control (owned backhaul vs third-party wholesale, interconnect capacity), plus visibility into OLT port utilisation and customer usage to predict both problems and upsell opportunities: “easier to upsell to an existing customer than… bring in a new customer.” He went as far as suggesting the creation of “10 hottest customers” reports.
Training Places
and spotting patterns humans can’t, but dangerous if it’s trusted to make changes without strong guardrails, accurate data, and verification loops. A field story about an HFC outage reinforced the human factor: a technician knew how to run a meter, but not what the results meant – evidence that tools don’t replace understanding. To close, Paddison asked for one capability likely to matter in the next 18 months. Gannon pointed to the sheer volume of modern telemetry (for example, OFDM complexity) and the need to learn where errors truly are, warning about false positives — like historical leakage detection programmes that were disabled because they created too much noise. Shrivastava picked “closed-loop assurance, with verification”: not just detecting and recommending, but proving the fix worked. Kingdom emphasised the looming skills gap: installing is one thing; fault-finding and diagnosing fibre issues at scale is another — and it will take years for altnets to embed that mindset. Frostad closed by arguing management also needs “training” to understand that resilient networks are worth more than fragile ones — and that the industry must invest ahead of the curve to remain sustainable.
Training and de-skilling became the other major thread. One audience member questioned whether digital twins and automation deskill “common sense engineering”, and asked for evidence that it improves time-to-delivery or type approval. Gannon responded that adoption has to be “show and tell”, and that automation can free scarce experts from repetitive testing – while still correlating automated outcomes against manual tests until trust is earned. Shrivastava described “workflow-based training” that teaches engineers which signals matter in a scenario, how to reduce alarm floods to one actionable root cause, and – critically – how to capture evidence of the fix and feed it back into the system. There was also a strong caution about AI reliability standards in telecoms and broadcast. Frostad put it starkly: “In our industry, you know, 85% is not fine, 95% is not fine, 99% is not fine. 99, 999. We’re starting to get somewhere.” AI Requires Guardrails That led to a broader debate: AI is powerful for analysing huge datasets
Volume 48 No.1 MARCH 2026
21
Made with FlippingBook - Online magazine maker