Overview of the AI maturity model in networking

Stage 1. Hyperreactive operations

The first step in the AI networking maturity model is hyperreactive operations, which is how most legacy networking approaches operated. Rossbach said this stage of network operations relied on manual network management processes, which most organizations have moved beyond. Most businesses in this stage don't manage networks in-house, however, opting to outsource operations to major telecom service providers, such as AT&T and Verizon.

Organizations with hyperreactive operations have network stacks limited to the size of a LAN or WAN. While operations teams might use manual network monitoring tools, systems and databases aren't integrated. Data is siloed between network infrastructures, management requires expensive costs and administrators need a high skill set to run an enterprise network in-house. This leads to outsourced operations.

Stage 2. Reactive management

The next stage in the AI framework is a more reactive approach to network management. Somewhere between the first and second phases of the timeline, organizations regain control of their infrastructures from telecom service providers.

In this stage, network administrators manage networks with Simple Network Management Protocol. This enables them to collect a variety of metrics for monitoring, such as events, alarms, faults, logs and config files. This gives administrators full coverage of a LAN infrastructure.

Despite these improvements, however, network administrators are still only capable of addressing issues with a reactive approach. They can only begin to remediate issues within the network once users report them, and they must fix them as quickly as possible.

Stage 3. Proactive strategy

A proactive strategy is a more modern step within the maturity model, Rossbach said. In this stage, network administrators have full visibility of a LAN or WAN infrastructure. They can monitor networks with a variety of metrics, and they're able to do so across traditional and software-defined networks.

Rossbach said this stage is when networks become more proactive. Network administrators are now able to monitor specific aspects of the network, such as the threshold, network performance and memory usage.

Stage 4. Predictive intelligence

The fourth stage of the AI maturity model in networking enables the use of an analysis engine. Rossbach said this is a unified network monitoring tool that provides visibility into an entire network path across traditional and software-defined networks.

"An observability practice should have the ability to collect all this data, correlate it, normalize it across multiple vendors that speak different languages -- Junos OS, Cisco OS -- and surface it to the [NOC], as this is your global network health," Rossbach said.

This enables network administrators to monitor network performance across all devices, vendors and metrics. It also helps network administrators extend their monitoring capabilities to unmanaged networks, such as ISP- and cloud provider-managed networks. Network administrators will be able to glean enough information to observe the entire network path from end to end, Rossbach said.

Stage 5. Automated infrastructure

The fifth and final stage of the AI maturity model centers on full automation. At this step, network administrators use the data and metrics they collect from visibility tools to automate parts of the network. This enables them to fix simple network issues more efficiently.

For example, if network administrators receive multiple alerts about a network problem, rather than needing to fix each ticket that appears, AI consolidates the alerts through an alarm that helps them find the root cause. Other tasks that an automated infrastructure simplifies include ticket escalations and config rollbacks.

The capabilities of AI in networking today

According to Rossbach, these capabilities are the extent to which AI-ready networks can currently operate. While AI-ready networks will be able to incorporate additional capabilities in the future, they can only automate a few use cases today.

In the meantime, AI networks will need to collect as much data from an enterprise network stack, including unmanaged parts of the network, as possible. This means the network must view the entire network -- from end to end, from users to cloud networks and everything in between. The path to intelligent network observability will also be through a layered approach. When it fully develops, AI will be able to improve performance based on what the network's AI workload requires, Rossbach said.

"This is the maturity model for AI-ready networks, but really, it's the maturity model to a successful [NOC]," he added.

Neither organizations using AI nor AI systems have reached full maturity. Most organizations are in either stage 3 or 4 of the maturity model, and AI engines are not yet fully developed to deliver advanced use cases. According to Rossbach, current AI engines can only automate around 10% of these capabilities.

Although AI aims to automate many simple and routine tasks within a network, the technology won't replace engineers because human network engineers will still be necessary to train AI. Network administrators will need to train AI models by feeding adequate data to the engines.

"This is what today's network experience looks like, and it's important for us to gain visibility into this," Rossbach said.

Deanna Darah is site editor for Informa TechTarget's SearchNetworking site. She began editing and writing at TechTarget after graduating from the University of Massachusetts Lowell in 2021.