Photo by Brett Sayles on Pexels

Elisa’s AI Transformation: When Machines and Engineers Work Side by Side

Finnish telecommunications operator Elisa has quietly been engineering one of the most significant operational transformations in European telecom — and the results are hard to ignore. According to Sami Komulainen, Elisa’s Chief Operating Officer, the Helsinki-based carrier has reduced network incidents by more than 80% through the strategic deployment of agentic artificial intelligence, operating in concert with a comprehensive digital twin of its live network. The achievement is not just a headline — it’s a potential roadmap for an industry desperately seeking to reduce costs, improve reliability, and manage ever-increasing network complexity in the 5G era.

What Is Agentic AI — and Why Does It Matter for Networks?

Unlike traditional AI systems that observe and recommend, agentic AI takes autonomous action. In Elisa’s case, the system doesn’t simply flag anomalies for human engineers to address — it identifies potential issues, diagnoses root causes, and in many scenarios, resolves them without waiting for human intervention. Think of it as the difference between a weather alert on your phone and a self-driving car that automatically re-routes when it detects traffic ahead.

The foundation enabling this capability is Elisa’s network digital twin — a real-time, high-fidelity virtual replica of its physical network infrastructure. By continuously ingesting live telemetry data, the digital twin allows AI agents to simulate the impact of any change or anomaly in a risk-free environment before acting on the live network. This dramatically reduces the chance of automated interventions creating cascading failures — historically one of the biggest concerns operators have had about autonomous network management.

Human-Machine Collaboration, Not Replacement

Critically, Elisa’s model is not about removing engineers from the equation. Komulainen has emphasized that the agentic AI operates alongside human network teams, handling high-frequency, repetitive troubleshooting tasks while freeing skilled engineers to focus on complex, strategic challenges that genuinely require human judgment. This collaborative model — sometimes called “augmented operations” — is increasingly seen as the pragmatic path forward for operators who cannot afford the risk of full autonomy but also cannot scale human workforces to meet 5G network demands.

The Scale of the Problem Elisa Was Solving

To appreciate the magnitude of an 80% incident reduction, it helps to understand the operational reality modern telecoms face. A nationwide 5G network can encompass tens of thousands of individual network elements — base stations, routers, core network functions, edge nodes — each generating continuous streams of performance data. As networks have evolved from relatively static 4G infrastructure to dynamic, software-defined 5G architectures with network slicing, open RAN deployments, and cloud-native core functions, the sheer volume and velocity of operational events has exploded well beyond what traditional NOC (Network Operations Center) teams can realistically manage manually.

Industry research consistently shows that a significant proportion of network incidents are recurring, pattern-based events — precisely the kind of problems where machine learning and automated remediation excel. By training AI models on years of historical incident data layered on top of the digital twin’s real-time network state, Elisa has essentially automated its way out of a large category of operational toil.

Industry Implications: Is This the Autonomous Network Tipping Point?

Elisa’s achievement arrives at a moment when the broader telecom industry is under intense financial pressure. Capital expenditure on 5G rollouts remains high across Europe and North America, while revenue growth from traditional services has plateaued. Operational expenditure — particularly around network management and staffing — has become a key battleground for improving margins.

Standards bodies and industry groups have been laying the groundwork for exactly this kind of transformation. The ETSI Zero-touch Network and Service Management (ZSM) framework and the TM Forum’s Autonomous Networks initiative have defined maturity models that categorize networks from Level 0 (fully manual) to Level 5 (fully autonomous). Most operators today hover around Level 2 or 3. Elisa’s results suggest it may be pushing toward Level 4 — a “conditional autonomy” state where the network handles most operational scenarios independently with humans available for exception management.

Other Operators Taking Note

Elisa is not alone in pursuing this direction, but it appears to be ahead of the curve on measurable outcomes. Operators like Deutsche Telekom, SK Telecom, and SoftBank have all announced AI-driven network operations initiatives, and Nokia and Ericsson have embedded AI-powered network management capabilities into their latest platforms. However, documented, quantified results at the scale Elisa is reporting remain relatively rare in public disclosures — making this a benchmark moment for the industry.

Challenges That Remain on the Road to Autonomous Networks

Despite the impressive numbers, the path to broader industry adoption is not without obstacles. Data quality and integration remain persistent challenges — a digital twin is only as accurate as the data feeding it, and many legacy operators struggle with fragmented, siloed data architectures. There are also growing questions around AI explainability: when an autonomous system makes a network change, can engineers understand and audit why that decision was made? Regulatory scrutiny in some markets may also impose limits on how autonomously critical infrastructure can be managed.

Additionally, training agentic AI models requires substantial investments in both data engineering and specialized talent — resources that smaller regional operators may find difficult to mobilize.

Looking Ahead: A New Operational Standard

Elisa’s 80% incident reduction figure is likely to reverberate through the industry for some time. It provides a concrete, credible data point that autonomous network management is not a futuristic concept but an operational reality delivering measurable ROI today. As 5G networks grow denser and more complex, and as the industry begins laying the groundwork for 6G, the ability to manage networks intelligently and autonomously will shift from competitive advantage to operational necessity.

For telecom operators still debating whether to invest seriously in AI-driven automation, Elisa’s results make the business case harder to ignore. The question is no longer whether agentic AI belongs in the network operations center — it’s how quickly the rest of the industry can catch up.