Three Critical Questions That Guide An AI Initiative
As enthusiasm for artificial intelligence in IT operations continues to grow, particularly in network operations (NetOps), it’s easy to get swept up in the excitement. But before launching an AI initiative, two foundational questions must be answered clearly and critically:
What problem are you trying to solve with AI that AI would solve better than a less expensive, simpler solution?
What is the nature of the available data? Is network and network-related data stored in disparate locations? Has it been cleaned and processed for analysis? How is the data updated? How is real-time data ingested and stored?
What is the makeup of your technical staff? Do you have access to developers, data engineers, and data scientists?
These questions are not theoretical. They are essential guardrails that help determine whether an AI initiative is feasible, valuable, and aligned with business objectives.
For the first question, clarity is often lacking. In many organizations, AI initiatives are passed down from leadership without a well-defined problem to solve. In other cases, the problem is clear, but AI may not be the best or most cost-effective solution. AI should not be deployed simply for its novelty. Instead, it must offer a distinct advantage over existing tools or processes. Otherwise, there is a real danger of being mired in neverending proofs of concepts, and an imbalance in the cost-benefit of implementing an AI solution.
The second question is equally critical. Network operations data is complex and spans both structured formats (like metrics, logs, and flow data) and unstructured formats (like tickets, email threads, TAC transcripts, and system logs). This data is often spread across disparate systems, arrives in different formats, and changes rapidly. Real-time telemetry in particular requires serious infrastructure and data engineering to clean, normalize, store, and analyze quickly enough to be actionable. In practice, solving these data challenges often constitutes the majority of the effort, which many experts say is up to 80% or more of implementing an AI solution.
These questions also surface additional considerations, such as whether the existing NetOps team has access to developers, data engineers, or data scientists. The composition of the team will influence the build vs. buy decision, the timeline, and the viability of long-term support for the AI solution. And with that comes the need to justify the return on investment, whether through direct cost savings, improved operational efficiency, or new revenue-generating capabilities.
In NetOps, this is often in terms of reduction of MTTR, new predictive insight, and overall improvement of operational efficiency. However, without the mechanisms in place to quantify and measure these outcomes, the benefits of an AI solution will remain a mystery.
Any AI deployment in NetOps will require iteration. Use cases evolve, data changes, and models must be tuned. Success doesn’t come from a one-time integration but from a commitment to continuous improvement and clearly defined goals.
AI holds incredible promise for network operations, but only when approached with purpose, realistic expectations, and an understanding that the problem and the data must dictate the solution. Without clear answers to these two core questions, any AI initiative risks becoming an expensive and complex experiment in search of a justification.
