AI and ML in the networks: risky business or critical to success?

STL Partners and Futurenet World are running a free survey for telecoms operators, benchmarking the industry’s progress at automating the networks and operations domains. Everyone who takes part will receive a personalised report of the findings and multiple submissions from one operator are welcome. Take the survey here: https://survey.alchemer.eu/s3/90293580/a55353a49f6d
Much has been written about the future of telecoms networks, and a significant chunk of it relies heavily upon the idea that networks will become “self-driven machines” – self-organising, self-healing, self-optimising and so on. To achieve this, would mean moving beyond rules-based automation and into the world of artificial intelligence (AI) and machine learning (ML).
This ambition is, in some ways well-placed. STL Partners analysis highlights implementing A3 (AI, data analytics and automation) within the network and operations domain will have the biggest financial impact on an operator. The question is, how much of this can be attributed to AI and ML versus more simple rules-based automation and RPA? And why is it that, despite the ambition, progress implementing AI and ML within the network domain has moved slower than in other areas?
Figure 1: Core network operations accounts for the largest proportion of value unlocked by A3, according to STL Partners modelling
Source: STL Partners, Charlotte Patrick Consult
The value of rules: guaranteed stability and quality of service
There are some crucial differences between the domains that have seen most early success in implementing AI and ML (customer care, sales and marketing and HR) in comparison to the network and operations domain:
- For operators, the network is their one core service – best efforts will not be tolerated. Within sales and marketing, if a campaign generated by an AI algorithm does not create the leads that are desired, it is an ineffective campaign. However, if an AI algorithm causes issues with a network outage in a particular region, it is a major failing. This means that new implementations within networks must be proven to work, at scale, across heterogeneous networks supplied by multiple different parties; a higher barrier to adoption than we see in other domains.
- Linked to the above, AI and ML introduces a degree of uncertainty. It can be most useful when there are fewer set procedures or workflows, and instead there is a desired end goal that can be achieved in multiple ways. AI and ML can also be most usefully applied to less structured data sets, where rules are less easy to implement. For a significant number of potential A3 implementations within the network domain, these criteria just aren’t met and that degree of uncertainty of cause and effect will not be tolerated. This too slows down the adoption of AI and ML within the networks.
Next generation networks: do operators even have a choice in implementing AI and ML?
It is widely agreed that for operators to roll out and manage next generation networks (e.g. 5G) in a cost-effective manner, automation will be required. The most bullish of operators within this domain have created stiff targets for themselves, like Telenor’s aim to be “touch free” within their networks by 2023, cutting around 15% of jobs and looking to save about 3% of their total operating costs. Or Elisa’s target to keep their CAPEX sending flat at 12% despite investing in 5G rollout. These goals will only be achieved if automation becomes pervasive throughout their networks, from network planning and provisioning, all the way to in-life management and fault resolution.
However, for many of these processes, rules-based automation and RPA will produce a large proportion of the cost savings, with AI / ML implemented down the line to squeeze even further efficiencies. This is not to say that we will never see successful AI and ML use cases within the network domain. There are already some concrete examples where these technologies can deliver value, such as in the management and prioritisation of incoming network alarms. But these implementations are likely to be in pockets, rather than saturated throughout the networks.

So what’s next: the role for AI and ML
There are three things that telecoms operators must get right to navigate the unique benefits and challenges of implementing AI and ML within the networks:
- Choose carefully when it is the appropriate technology that will bring the biggest benefit. Where rules-based automation will create a similar impact, the stability and predictability it can bring will make implementation faster, cheaper and less complex.
- Where AI and ML will deliver real value over and above simple automation, ensure that trust in the algorithms is built. In many cases, this will mean a staged implementation where decision makers can compare what a human would have done with the algorithm output. If the two decisions do not match, time must be taken to understand why the algorithm has suggested something different from the usual human practice.
- Make clear the new role that employees, such as network engineers, will have when automation (and AI-supported automation) is widespread. This must be as well articulated and structured and operators’ technology roadmaps.
STL Partners is a boutique research and consulting house focused on the TMT sector. We are currently running a free survey,in partnership with FutureNet World, for telecoms operators globally to understand more about automation strategies and implementations in the network and operations domain. To take part, please click here.