Telcos collaborate to deliver AN business benefits faster

TM Forum members, including Entel, Orange, Telefonica and Vodafone, explored how to move faster on achieving and demonstrating business benefits from network autonomy, during TM Forum’s AN Workshops on 20th & 21st October, hosted by Google & Zinkworks,

Several sessions underlined why it is a pressing issue: While many telco stakeholders have a mandate to progress network autonomy they also face internal obstacles to securing investment for AN initiatives.

The promise

TM Forum has defined six autonomous network levels from 0 (manual operations) to 5 (fully autonomous), with over 80% of telcos indicate they want to achieve AN level 4 (highly autonomous) in 2030 (TM Forum 2026 AN Survey). But adherence tends to be qualified. Vodafone, for example, aims to achieve AN level 4 by 2030 “where it makes business sense.”

Challenge 1: Linking AN technical progress to business benefits

Today, however, only 37% of telcos can link the operational benefits of AN, such as faster mean-time-to-repair (MTTR), to a tangible financial benefit (TM Forum 2026 AN Survey, see this article).

Efforts are underway within TM Forum to strengthen the measurement of business benefits. One of the most interesting sessions on this topic came from Nigel Portley, Account Director, Zinkworks, who presented a use case for autonomous service assurance for fixed fibre home broadband services, based on a tree of indicators:

• Key Capability Indicators (KCIs): Indicators on how things are being done, in this case task operation automation ratio and fault management automation ratio
• Key Effectiveness Indicators (KEIs): Indicators on how well operationally things are being done, in this case zero-touch trouble ticket resolution ratio, truck roll volume, and mean time to repair
• Key Business Indicators (KBIs): Indicators that materially affect the business, like reduced operations & maintenance costs, time saved in network operations, employees reskilled, revenue from operations

These three levels of indicators are encapsulated in TM Forum’s IG1256 published in July. This wasn’t just theory. Nigel was able to talk through how the program with a tier 1 telco customer was approximately a quarter through, and these impacts were being felt, including at a business level. Not every session was like this, underlining the challenge many telco stakeholders face internally to get investment for AN initiatives, even when they have been given a mandate to progress network autonomy.

The progress

Examples of telcos achieving higher AN levels with single domain or high value scenario (HVS) are increasingly well documented. Within TM Forum’s AN Level Assessment Validations, 17 results published in 2026 in the year to November scored 3.3 and above, with six accredited at level 4. However, the major technical challenge, according to workshop members, is scaling autonomous networks cross-domain.

Challenge 2: Data architecture and governance the lynchpin to scaling AN

Many of the more technical workshop sessions focused on the ubiquitous issue of data management and governance. In critical network operations environments the price of failure is extremely high. To increase the accuracy and efficacy of autonomous decision making in networks telcos need a series of reasoning and governance structures.

The workshop dedicated several sessions to how TM Forum members are building ontologies for autonomous networks. An ontology is a formal, structured description of the concepts in a particular domain or knowledge base and the relationships between them. If an AI agent is configuring a network, it would need an ontology to understand the relationship between different data inputs.

One ontology, however, is unlikely to be enough for an autonomous network. The presentation by Richard Kilmurray, Principal Network Automation Architect, Vodafone, provided an overview of how different depths of knowledge and understanding are required at different network layers and domains:

• Granular ontologies are required at a network and system domain level to ensure that autonomous decisions taken within each domain are accurate and trusted.
• A cross-domain ontology must also be in place at the service layer, which includes enough detail so that cross-domain decisions can be taken.
• A service layer ontology needs to be balanced, too little detail and agents making autonomous decisions at the service layer could waste tokens creating their own ontologies, with a greater risk of introducing errors into the decision-making process
• Ontologies are only one piece of the puzzle, apart from getting the underlying data in order autonomous decision making also requires knowledge graphs, and could utilize constructs like digital twins as well

Other presentations highlighted additional business challenges to scale, from developing promising pilots that end up being technically impossible to expand, to a multi-country operator which may have the same vendor mix in two operating countries but be unable to replicate AN implementations due to operational nuances.

The problems

The prize of solving one issue is usually a new problem. The workshop highlighted the increasing need to integrate security inputs into developing autonomous networks.

Challenge 3: Autonomous capability ≠ secure capability

The ability to act is not the same as the right to act. This makes governance frameworks that manage autonomous decision-making critical for scaling AN across boundaries.

Autonomous systems have to manage conflicting intents, such as energy saving and customer experience. To run a network efficiently power usage should be as low as possible, but run too low and it may not meet customer needs.

Hans Neff, Senior Director within the CTO Group, ZTE delivered a presentation on “The autonomy envelope - Defining when an autonomous network is allowed to act” which thoughtfully explored the different boundaries that telcos should consider for governing autonomy in their networks.

“An autonomy envelope is a governed, machine interpretable specification of the authority delegated to an autonomous actor defining the decisions and actions it may perform, the applicable scope and conditions, and the limits under which that authority remains valid,” according to Neff.

Practically this takes into account:

• Actor: what/who is empowered to act?
• Scope: which resources is an actor empowered to manage?
• Action: what actions can be taken to change?
• Impact: what level of risk is permitted for actions taken?
• Time: when and for how long can action be taken?
• Evidence: what must be proven to what confidence level for an action to be authorized?
• Conditions: when should an action stop?

Ideas like the autonomy envelope are the kind of governance mechanisms that the industry must adopt if it is to manage potentially thousands of AI agents configuring networks with conflicting intents.

The next steps

So, can the industry move faster to achieving the business benefits of network autonomy? The sense from the workshops was that momentum is building as evidence accrues from diverse implementations. But more work needs to be done to connect the business to the autonomous network. This will be explored more in upcoming TM Forum deliverables, including the TM Forum AN playbook for CxOs, and a TM Forum Insight report on the economic impact of AN.