Getting Mobile Network Operators to a Network That Runs Itself

*Blog by Omar Shahdad, SVP, Operations, Mavenir

I’ve lost count of how many times I’ve heard an operator ask, “what level are we at?” It’s the wrong first question. TM Forum gives the industry a good scale for autonomous networks; L0 to L5, manual to fully autonomous, but advancing in levels isn’t the be all and end all.  What matters is what actually changes on the ground at each step, and whether it moves within the business.

I keep coming back to this because I see the same pattern in almost every conversation with mobile network operators (MNO). Someone in the room asks where they sit on the maturity scale, gets an answer, and moves on to the next agenda item without ever asking the harder question: what would it take to move up one level, and what would that actually buy us? Now that’s the conversation worth having. 

Allow me to break-down the climb, in plain terms.

At L0, everything is manual. Engineers watch dashboards, chase alarms, and touch every change by hand. Nothing wrong with that, it’s where most networks start, but it doesn’t scale as complexity grows. I’ve sat in NOCs where a single congestion event triggers a scramble of phone calls, spreadsheets, and tribal knowledge that lives in one person’s head. It works until that person is on vacation or retires.

L1: Add scripts and rules. Alarms get quieter, and reports get easier to read, but someone still has to work out what’s wrong and fix it. This is usually where operators feel like they’ve already automated a lot; dashboards look cleaner, tickets are categorized, but the actual diagnostic work hasn’t moved off a person’s plate.

By L2: predefined workflows start handling the routine stuff; scaling, config changes, upgrades, and a human just has to say ‘yes’. This is where AI starts earning its keep: recommending the right action instead of asking someone to build it from scratch. It’s a small shift on paper but a big one in practice, because it’s the first time the network has been doing more than reacting to rules someone wrote months ago.

L3: is where it gets interesting. A single domain starts closing its own loop; known problems get fixed without waiting on a ticket. This only works if the automation has actually learned from how your best engineers operate, not just from a static rulebook. I’d argue this is the level most operators get stuck at, because closing the loop safely means trusting the system enough to let it act without a person in between, and that trust must be earned, not assumed.

L4: is cross-domain. Core, transport, and service layers start correlating with what they see and catching problems before a customer ever notices. Mean-time-to-repair drops from hours to minutes. This is also where the economics start to shift meaningfully, because the cost of a degraded experience (churn, support calls, brand damage) starts showing up in the numbers before it ever reaches a customer’s phone.

And L5: the level most talked about, is where an operator just states what it wants (“keep latency under X in this region”) and the network figures out how, executes it, checks its own work, and keeps adjusting. People supervise outcomes, not tasks. Very few networks are actually here today, and that’s fine. L5 isn’t a finish line so much as a direction to keep walking toward.

None of this happens by flipping a switch. It happens by giving each stage the right tooling, and by being honest about which stage you’re actually at rather than which one looks good in a board deck. 

This is where national operators feel the squeeze more than most. MNOs are usually running networks stitched together over a decade or two, with equipment from multiple vendors, layers of OSS bolted on top of each other, and an engineering team that’s been shrinking as senior staff retire and their replacements are harder to find. Add regulatory obligations, coverage mandates, and margin pressure from competitors with newer infrastructure, and the case for automation stops being a nice-to-have. It becomes the only realistic way to keep the network running well with the people and budget available.

That’s honestly what I find most interesting about where Mavenir is putting its energy right now. MAVcore and MAVapps expose open APIs, so there’s something for automation to grab onto in the first place, you can’t automate a black box. MAVscale’s Cloud-Native Automation handles the GitOps-style deployment grind: spinning up Kubernetes clusters, rolling out network functions, and managing lifecycle changes without an engineer manually stepping through a runbook. Network Intelligence as a Service turns raw telemetry into AI-recommended actions instead of another dashboard to stare at, using everything from deep learning to digital twins to work out what’s actually going on inside the network.

The newer piece, and the one I think matters most for L3 through L5, is the Agentic Service Assurance Framework. It pairs with an Intent Orchestrator with an Intent Ops engine that captures what good NOC engineers do and turns it into automation the network can run on its own. Not a static playbook but built on the judgment calls skilled engineers make when triaging faults. —Every workflow gets validated against a human baseline before it’s allowed to run unsupervised, so operators aren’t being asked to hand over the keys blind. It’s also built on TM Forum’s own interoperability standard, IG1453, which means it talks to an operator’s existing OSS through agent-to-agent communication instead of asking them to rip out systems they’ve already paid for.

To understand how this looks in action, you can see what’s already happening in live networks.

Deutsche Telekom: cut energy use in its 5G Core by up to 65% through cloud-native automation and lifecycle management, a number large enough that it shows up in sustainability reporting, not just an ops dashboard. See announcement here: Telekom pioneers new cloud approach: Up to 65 percent energy savings in the 5G core network

Tune Talk, Malaysia: became the first fully cloud-native mobile network operator in ASEAN by modernizing its OSS and BSS on Mavenir’s platforms. See announcement here: Tune Talk Launches Malaysia’s First Cloud-Native Mobile Network on AWS, Driving Next-Generation Network Modernization

A Tier-1 brownfield operator: applied AI-native network intelligence to live 4G/5G traffic and saw real, measurable gains in performance and user-experience KPIs — proof that this works on networks that already exist, not just greenfield builds designed for automation from day one.

Mycom is now connecting its NOC Copilot directly to Mavenir’s domain agents through agent-to-agent integration, aimed squarely at getting CSPs from assisted automation to genuinely autonomous operations faster. See announcement here:: Mycom and Mavenir announce strategic collaboration for delivering Agentic AI benefits to CSPs to support Autonomous Networks L4/5

For MNOs juggling legacy complexity, an aging workforce, and rising expectations, none of this is theoretical anymore. The path to a network that runs itself is walkable, one measurable stage at a time — and the operators who get there first won’t just be running cheaper networks. They’ll be running better ones, with the people they have, not the people they wish they had.

What’s kept your organization from climbing faster — technology, budget, or trust in the automation itself? I’d like to hear it.