Artificial Intelligence Reaches the Core of ISPs
17/11/2025
Internet service providers across the region are beginning to integrate AI into the management and monitoring of their networks. The expectation is clear: detecting failures before they impact users, and real-time automation of responses. But success depends on one critical factor— how reliable and well-structured their internal data is.
Artificial intelligence is no longer an abstract concept for Internet Service Providers (ISPs). As networks grow more complex, traffic volumes continue to rise, and customer-experience pressure increases, companies are turning to tools that can analyze events, predict outages, and trigger automated actions. Yet that excitement comes with a structural challenge: data quality.
“More than 80% of AI initiatives fail today— and it’s not due to a lack of technology, but because of poor-quality information,” said Epafras Schaden, Chief Technology Officer at EdgeUno, a network services company rapidly expanding in Latin America.
“If we don’t know for sure which customer is connected to which interface, or how a circuit is configured, no AI model can operate effectively.”
At LACNIC 44, EdgeUno’s CTO highlighted the potential of using AI for intelligent anomaly detection, predictive analytics, and the automation of repetitive tasks. “AI allows us to move from reactive operations to truly proactive ones. We can anticipate potential issues and address them before they affect customer experience,” he explained.
This has a direct effect on end-user experience: fewer incidents and shorter response times allow support centers and NOCs to focus on complex troubleshooting instead of dealing with repetitive, low-value tasks.
But getting there requires putting internal systems in order first. Integrating data from CRM, BSS platforms, monitoring tools, network inventory systems, and operational logs is essential. “There’s no point in collecting every router log if I can’t link that information to the customer database. Isolated data doesn’t create context—and without context, AI cannot understand the network,” he noted.
(Free access, no subscription required)
A LONG ROAD AHEAD. One of the tools highlighted was the use of large language models (LLMs)—both commercial and locally hosted—integrated through protocols that allow AI to query real-time network data and suggest automated actions. “Our goal is to move from a manually operated network to an autonomous one, capable of self-correcting and optimizing without direct human intervention,” the CTO said.
The transition is not immediate. It calls for investment, time, and technical teams willing to develop entirely new skills. “Network engineers now have to adopt a software-architect mindset. This is a substantial cultural change,” Schaden emphasized. The lack of talent with this hybrid expertise is becoming a major obstacle for operators.
SOVEREIGNTY AND PRIVACY. Data sovereignty and privacy concerns are pushing many companies toward fully local AI deployments. “Security and cost are critical. Running models locally lets us control both,” he emphasized.