AIOps to the rescue: How to get there

In the previous blog, I argued that AIOps is not just about running operations faster or with fewer people. It is a deep change that requires a fundamental technological, organizational and cultural transformation towards AI that will unfold over the years.

Getting ready: Avoid the hype

It is tempting to plunge into AI-driven operations with AIOps, but we need to accept that the technology is still maturing and, perhaps more importantly, most organizations are not yet ready for it.

The first step is to unplug from the hype and look inward. What are your most pressing needs and priorities? What can you realistically implement and when? What network domains should you focus on first? What results can you expect? Do you have a business case that validates your implementation strategy? How should you measure success?

Each operator will have its own answers and roadmap. The challenge lies in striking a balance between doing too much too quickly, which can be overwhelming and lead to disappointment, and waiting for a complete overhaul, which might mean missing out on immediate and mid-term benefits of AIOps.

Choosing what you need: Models and data

Once you have set a strategy, you need to pick AI models and platforms that fit your needs. What are the models that are best suited to your data, goals and processes you want to automate? How close to the edge should they be located?

Equally important is selecting the right data to feed these AI models. Not only must the data be accurate and properly cleansed; it must also have the right granularity. Too much detail or overly high-frequency data can introduce delays and reduce accuracy. AI models will also need to access data from different vendors and network elements from multiple domains, ideally through standard or common interfaces.

Measuring the outcomes: Trust and ROI

A successful transition to AIOps requires the ability to measure its impact quantitatively using multiple metrics and KPIs. This lets you validate your initial choices, adjust course when necessary, and drive the expansion of AIOps into new domains.

According to several analysts, 2025 was the first year in which opex efficiency improved, with several telcos reporting opex growth below revenue growth, thanks to AI-driven closed-loop assurance, predictive maintenance, and energy optimization as main examples.   

Measuring outcomes is essential for building trust in the new processes, too, especially since they may differ from established ones and require operators to relinquish direct control over an expanding range of tasks. Developing trust is a mutual, incremental learning process. Humans must learn how the AI infrastructure works, and the AI systems must adapt to human requirements.

Integrating AI technology and culture: A two-way path

Trust is essential because implementing AIOps demands cultural and organizational changes that run even deeper than technological changes. These shifts can be more complex and harder to absorb. AI is more than the efficiency layer we have today; it can, and should, drive an all-encompassing business model shift.

Predictive capabilities will make field technicians’ schedules more manageable and less driven by emergencies. At the same time, more skilled talent will be needed to manage the AI infrastructure, potentially requiring new hires or reskilling of existing staff. As the scope of AIOps widens, collaboration among business units is crucial to ensure consistent implementation across the organization.

AIOps should be integrated into existing network operations and introduced gradually from within, rather than as a separate add-on run by a separate AI team. This approach fosters the growth of internal championship for AIOps among legacy teams, which is necessary to avoid fragmentation and duplication of effort.

And now repeat

Adopting AIOps is an iterative process that gradually extends the scope of AI-driven automation in operations. This expansion occurs in two dimensions: horizontally, from individual network elements or areas to the entire end-to-end network spanning multiple domains, and vertically, revealing deeper patterns and transforming processes in ways humans could not have anticipated.

AIOps will only lead to truly autonomous networks if it continues to improve and adapt to meet operators’ evolving goals and rising network complexity.

All these steps must be revisited and refined to reach each new phase of automation. Each iteration will deliver greater efficiency, deeper trust, and improved performance.