A Path Toward Trusting Agentic NetOps

Network operations teams are increasingly seeking AI tools to improve their ability to build and manage infrastructure. In fact, they now expect AI features and products from their network infrastructure and network management vendors.

Enterprise Management Associates (EMA) recently published the research report AI-Driven NetOps: How Enterprises are Embracing Intelligent Network Management Solutions, based on a survey of 458 IT professionals. The report found that AI features are very inflectional over vendor selection for network management and network infrastructure products in 59% of IT organizations.

Expectations for AI are growing. Most network operators believe that AI will help them

• Complete tasks faster
• Reduce errors
• Eliminate busy work

Network Teams Won’t Use Tools They Can’t Trust

Terminology for this transformation is volatile. Whether you call it AIOps, AI-driven network management, or agentic NetOps, the opportunity is certainly there. Unfortunately, agentic NetOps transformation will only occur if network operators trust these AI tools.

EMA research found that only 31% of IT professionals completely trust the AI tools that their organizations are applying to network management. Soft trust in AI correlated with:

• Less success with AI-driven network management tools and products
• Weaker uptake of AI tools by network operators

Notably, trust is not about AI technology perfection. Many research participants who trusted their AI tools told EMA that they were aware that those tools often made mistakes. So, what’s going on here?

How to Build Trust

In our research, we asked research participants to tell us how vendors could best earn their trust with AI capabilities. They identified the following:

1. Be transparent about your approach (56% of respondents). Vendors make mistakes. They have breakthroughs. Network operators want to know about this journey. They want vendors to be open about how thy arrived at the AI solution they deliver to the market. They don’t want secrecy.
2. Connect insights to the data (50%). How can a skilled engineer know that an AI insight is based on reality? They need to see the data. Any conclusion or recommendation made by AI should be attributable specific data. This data is the evidence Of course, one will need to ensure that these AI tools are working with good data, which is presents its own set of challenges.
3. Visualizations of reasoning (47%) Network operators want to see decision trees and graphs that explain how AI reached its conclusions. These visualizations can help operators understand the logic behind the results. They might even be interactive, with operators being able to tweak certain data to see how insights shift based on changing conditions. Notably, respondents who trusted AI the most were more likely to select this option.
4. Manual confirmation workflows (45%). Many operators want to go beyond data attribution and work with the data. They want to manual workflows that allow them to recreate the process that AI automated. AI tools that offer such workflows or connect to them can help.

Earlier, we noted that AI mistakes did not necessarily lead to distrust. EMA believes network operators can learn to trust AI if AI tools are transparent and verifiable. If AI does make a mistake, that mistake should be easy to detect through many of the attributes described above.

Transparence, data attribution, visualizations, and manual confirmation workflows will also allow network operators to track AI trustworthiness over time. As AI tools improve, network operators will see that improvement through day-to-day interaction. This is how AI solution providers can earn the trust of network operations teams.