3 key AI use cases in telco networks

5G is going to enhance the field of AI, but AI can also play a key role in the rollout of 5G itself. This article explores the different types of use cases for AI as applied to telco networks.

3 types of network AI use cases

In previous work by STL Partners, we identified 3 core categories of AI use cases as applied to networks:

1. Fault detection, prediction and resolution
2. Network optimisation
3. Network planning and upgrades

In this article, we will run through several of the key sub-use cases in each of these categories and explain why they matter.

Fault detection, prediction and resolution

Service impairments and faults are inevitable in a telco network, so this is a critical area in which AI can play a key role. While optimising your network may be a secondary consideration, having a functioning network is the primary consideration. Faults will also often result in large costs, whether the operations and maintenance costs themselves or fines for breaching SLAs.

Given these concerns, 3 KPIs emerge for measuring fault detection, prediction and resolution:

1. Labour cost: the cost of engineers fixing the problem, as well as customer-facing roles dealing with complaints
2. Speed: how quickly the telco can identify the problem and therefore solve it; measured as mean time to repair (MTTR)
3. Customer experience

The use cases that we will run through all seek to target these 3 KPIs, whether directly or indirectly.

Root cause analysis

Historical data is essential for detecting the root cause of network faults, in fact ML models can learn to predict the likely causes of new network faults. This dada includes customer calls, type of customer premise equipment, firmware, trouble tickets and historical data on customer premise visits.

For relatively common faults, there is a richer bank of historical data and models are able to identify the cause of the fault more quickly, hence improving all 3 relevant KPIs. If using vendor-developed solutions then the learnings from other telcos’ networks can also be applied, shortening the time to find the root cause further still.

However, for less frequent service issues which are more operator specific, it is harder to fall back on vendor knowledge. Rich and deep data must exist for more sophisticated root cause detection. The more data that the models have access to, the more likely it is that they can predict when a specific faut is likely to occur before it actually does.

Fixing the problem

Automation to fix network problems has existed in the form of fixed policies written by network engineers for over a decade. Detail in the data is required to automate the recommendation of fixes without any human input. Engineers go to sites to fix a problem, logging every step they took to fix the problem, but it is essential that they also log what the actual problem was. Richness in the data is key for automation at this level.

As models learn they are able to provide more complex recommendations, detailing the fastest solution, most cost-effective, or the least impact on customer experience.

Field engineers would only now be required for the instances when the model is unable to match a solution to a problem with a high enough level of confidence. However, a level of buy-in is required from the field engineers, and it is a sensitive issue to move from human to machine solutions. As a result, many operators may see this sub-use case as one to introduce more gradually than others which are less likely to displace human workers.

Make network innovation the engine of growth

Our Network Innovation service supports telcos and vendors in the creation and adoption of innovative technologies and services

Book a demo

Network optimisation

Network optimisation is about how to route traffic and balance workloads across the available infrastructure and assets to try and deliver the highest quality of most cost-effective service. Elisa rolled out a self-organising RAN and reported improvements in efficiency and quality of service:

  • 20% reduction in mobile network customer complaints
  • 2% improvement in CAPEX efficiency

It is possible to optimise the network manually, but with thousands of radio sites this would mean the whole team of network engineers doing nothing but re-optimising the network. Hence, the benefit of a self-optimising network which follows a similar process as for fault resolution:

  • Real-time, event-based network data highlights a service degradation, relating to a specific root cause (e.g. sharp rise in traffic in a specific area)
  • The recommendation engine consults the policy engine to find out what the operator’s pre-defined intent is in the given situation (e.g. deliver as high quality of service as possible)
  • The recommendation engine then suggests which fixed policies, also stored in the policy engine, to implement in order to meet the intent, while adhering to any constraints (e.g. boot up any assets on standby, re-route some traffic through longer paths to reduce congestion, prioritise SMS and calls over video streaming, etc.)
  • An automated system re-optimises network equipment in line with the recommendations

It is clear that this type of network is desirable, so what are some of the sub-use cases that make it achievable? The sub-use cases address both cost and quality of service, and are broken down as such:


  • Maximise use of existing network assets / recommendation engine in where to deploy base stations
  • Predict potential service level objectives violations and prioritise traffic to reduce the risk of SLA breaches, or prioritise traffic for breaches that would incur higher costs
  • Prioritise and plan firmware updates, for times that will be least disruptive for customers and dependent network equipment

Quality of service

  • Optimise distribution of traffic between cell sites and/or service disruptions, and revert to default configuration after event/peak is over
  • Service prioritisation (e.g. calls over video streaming) during demand peaks or service disruption
  • Detect hungry VNFs that are hogging shared cloud infrastructure resources (CPU, memory, storage, networking) and re-adjust service prioritisation / spin up additional resources dynamically

With 5G and a move towards more cloud-native networks, it is expected that more use cases for AI and automation will be discovered.

Network planning

Network planning had a period where it was seen as less of a priority for many operators. The operators leading the way on AI are generally Tier 1 operators who had mostly completed roll out of 4G networks and hence were less concerned with network planning. However, with the roll out of 5G this can be expected to change. In a survey conducted by Ericsson, 70% of solution providers stated that it was in network planning where they expected to see the highest returns from AI adoption.

In a previous STL Partners report, we highlighted the example of Telefónica using AI image recognition of satellite images to try and identify rural populations not currently served by their network. By comparing mapping of unknown communities with real network coverage they were able to identify underserved areas and then further use AI to deploy networks there as well. We can expect to see more uses of AI in network planning as more operators have rolled out their 5G networks

AI and the rollout of 5G

These use cases can play a role in helping operators deploy their 5G networks. We generally associate 5G with enhancing AI, but actually AI can play a key role in the deployment of 5G itself and in improving the service offering of 5G, particularly in key areas of 5G network management. This is also the case with the deployment of 4G in less developed countries. As we look to the future, AI can play an important in dynamic 5G network slicing in enabling telcos to intelligently provision network resources so that slices can be scaled up and down as needed whilst maintaining SLAs.

A more intelligent and automated approach to networks will increase margins and increase customer satisfaction. For this reason, operators should seek to place AI use cases at the forefront of their minds when deploying their 5G networks.

At STL Partners, we have extensive research and consulting work within automation, analytics and AI (collectively known as A3) both within the network domain and beyond. We will be sharing our perspective and key insights from our work very soon, so stay tuned!

Matt Bamforth


Matt Bamforth

Senior Consultant

Matt is a Senior Consultant at STL and has experience in consulting projects across a wide range of topics. These span areas such as 5G, private networks, telco cloud, and edge computing. Matt has previous experience in strategy consulting, as well as in the Fintech sector. He holds a BSc in Economics from University College London.

Network APIs: Unlocking new value in the telco cloud

Network APIs may offer an answer to the question of how to monetise recent and upcoming telco cloud deployments. Virtualised networks upgrade APIs and enhance the value they offer to developers and customers. To unlock their potential, telcos should focus on optimising their commercial models.

Progress in telco cloud: How do we measure agility?

In the January 2023 update to our Telco Cloud Deployment Tracker, which was looking back at the entirety of calendar year 2022, we recorded an overall slowdown in deployments.

Advanced analytics for telecoms: how to unlock AI and ML capabilities

With more data than ever before, advanced analytics forms the base for AI and ML capabilities to play an essential role into the day-to-day operations of major telcos enabling smarter, more agile networks.