April 2024 | www.tmforum.

org

driving intelligence
in network lifecycle
automation:
focus on Asia-Pacific

sponsored by:

Author: Charlotte Patrick, Contributing Analyst

Editor: Dawn Bushaus, Contributing Editor
 contents
03 the big picture
06 section 1: the need for intelligence in network lifecycle automation
11 section 2: how much AI and machine learning are CSPs implementing today?
14 section 3: examples of intelligence in network lifecycle automation
22 section 4: make it happen – strategies to take advantage of new intelligence
26 additional resources

Find out more from our sponsor about AI-driven intelligence

in network lifecycle management in Asia-Pacific:

ServiceNow helps Telcos accelerate growth, maximize network investment, and

reduce costs across the entire digital ecosystem – all while delivering seamless
experiences. Be it planning for 5G or fiber buildouts, managing data centers, launching
and delivering new services, or automating network operations, our purpose-built
solutions help make work easier across the entire service lifecycle with automation
and AI. Built for the fast-changing world of Telecom, the Now Platform connects
people and data for greater productivity and innovation.
the big
picture
3 inform.tmforum.org
 Since the launch of ChatGPT in late November 2022, intelligence (defined in this research
as the deployment of both AI and machine learning) has moved into the mainstream
consciousness of communications service providers (CSPs), giving staff the ability to
interact with a large language model (LLM) and discover its strengths and weaknesses for
themselves. In the Asia-Pacific (APAC) region, large operators are implementing a range of
active intelligence projects, while smaller CSPs are moving more slowly.

Large CSPs in countries such as Australia, India, Indonesia, are well positioned to exploit the potential of AI and
Japan, Malaysia, New Zealand, the Philippines, Singapore, machine learning. Mark Sanders, Telstra’s Chief Architect,
South Korea and Thailand (note that this report does not notes that many of his engineers are curious about the
include China) are seeing some early value from domain- possibilities of AI, and he adds that they are running
specific deployments that allow them to build the skill sets projects outside regular working hours.
needed to work with intelligence, without the complexities
of cross-domain data gathering or processes. Successful Where’s the value?
projects include anomaly detection for service assurance The graphic at the top of the next page from our AI
and task-specific models in the radio access network Benchmark report highlights an interesting aspect of this
(RAN). Smaller CSPs, or those in developing markets, do enthusiasm in that respondents expect high potential
not have as many intelligence projects underway, and impact from adding intelligence in all three survey
it is unclear whether they will invest heavily unless the categories: customer experience, operations and new
use case solves a significant pain point or provides well- product development. It’s unclear whether this means
documented return on investment (ROI). that there are equally exciting opportunities across the
company or that respondents don’t have a firm idea of
TM Forum surveys conducted this year and last have where the real value will come from yet.
found much enthusiasm and confidence in AI and machine It is unclear
learning among CSPs in the Asia-Pacific region – although This issue of value recognition is one of several challenges whether smaller
the number of respondents from the region is small for CSPs operating in Asia-Pacific. On the technical side, CSPs will invest in
compared to the full group of global respondents. Level 3 onwards of TM Forum’s Autonomous Network
intelligence unless
maturity model, where machine learning begins to be
Our AI Benchmark survey, for example, found that a full deployed across the network, is much more challenging
the ROI is clear.
85% of APAC CSP respondents believe their organizations than earlier levels.

4 inform.tmforum.org
Problems include but are not limited to legacy Where will AI / ML most positively impact
infrastructure, data quality, lack of organizational skill sets,
and several issues related to governance, data privacy and
the telecoms business? (APAC responses)
risk management.
Customer experience / CRM

The survey conducted for TM Forum’s Digital Transformation

Tracker (DTT 7) gives some insight into the maturity levels
of APAC CSPs, with about half not having a real plan or
Operations / network operations
vision in place to tackle the coming complexity (see pie chart
opposite). This research, therefore, looks at the scope
of intelligence deployments that will be needed to meet
corporate goals such as improving customer experience, saving New product development n Low
n Medium
operational costs and underpinning new revenue streams. n High

0% 10% 20% 30% 40% 50% 60% 70% 80%

Our aim with this report is to look at the progress to date
in the Asia-Pacific region and consider the future steps TM Forum, 2024

necessary to support successful rollout of intelligence.

APAC CSPs’ approach to OSS / BSS automation
Read the AI Benchmark to learn more
about telcos' AI strategies:

47%

n We have an architectural
roadmap / vision which we
adhere to strictly

n We are pragmatic and

seek to build automation
into processes as needed

53%

TM Forum, 2024

5 inform.tmforum.org
 section 1

the need for

intelligence
in network
lifecycle
automation

6 inform.tmforum.org
 Network lifecycle automation (NLA) is the use of intelligence and automation to create end-
to-end processes across multiple domains within a CSP’s environment such as the RAN, IP
transport network and mobile core network, and across multiple vendors’ solutions. Adding
intelligence and automation results in many new capabilities such as fault resolution for the
self-healing network and prediction to allow better network planning.

The six main NLA processes are shown in the graphic (TM Forum defines intent as “the formal specification
on the next page with a rough timeline of intelligence of all expectations, including requirements, goals, and
deployment depicted by the arrow. Rollout typically constraints given to technical systems”.)
starts with descriptive intelligence used in service
assurance scenarios and progresses downward and to Network deployment and maintenance. This process
the right to become prescriptive, encompassing many includes all activities enabling the deployment,
operational processes. The black text shows use cases commissioning, self-configuration, testing and
with the need for machine learning while pink text validation of networks – including RAN-specific
highlights future use cases for Generative AI (GenAI) or actions such as neighbor relations.
other LLM tools. A description of each process follows.
Service orchestration. Service orchestration
Service assurance, trouble management and designs, creates, delivers and monitors service
self-healing. The assurance process includes the offerings, requiring automation that spans order
deployment of AI in operations (AIOps) and other handling, service or resource provisioning, network
intelligence to handle the high volumes of data orchestration, policy management, and assurance. It
generated in 5G networks and to provide complex aims to move from early automation in order handling
anomaly detection, correlation and causal analysis. towards the automated deployment of services
Automation using AIOps will also be necessary to on the (increasingly hybrid and complex) network
move to a self-healing network. and automated assurance of those services against
Adding intelligence
customers’ service level agreements (SLAs) and
Network management. This category contains all intents.
and automation
processes for managing and optimizing network results in many new
domains, including specialized machine learning use capabilities.
cases in the RAN and power management. It also
includes the development of intent-based networks

7 inform.tmforum.org
 Network planning/capacity management. Examples of intelligence deployed in
This process includes the strategic design of the
network to meet current and future needs, including
network lifecycle automation
topological design, network dimensioning and Descriptive Diagnostic Predictive Prescriptive
simulations. Capacity management, requiring the
Assurance, Data management Root cause analysis Prediction of future Automated raising
automation of traffic analysis and diagnostics, for monitoring and (automated) issues and their and remediation of
trouble management
forecasting of future capacity needs, and automated and self-healing reporting Anomaly detection impact trouble tickets

configuration of the network also happen here.

Network Visualization of Identification of Prediction of future Provide instructions
management device-level energy network configuration failure points and to move network
Intelligence to support operational teams. This consumption on that breaches policy best resolution workload to most
the network appropriate cloud
category brings together a range of intelligence deployment
needed to support humans working in network- Multi-agent systems
related areas, including the network operations center
(NOC) and service operations center (SOC), service Network deployment Prediction of trends Prescriptive
and maintenance Generation of Diagnostics in affecting network suggestions to field
help desk, and field services. These capabilities will 3D coverage maps network testing configuration services deploying
often be delivered as part of a solution in one of the Predictive maintenance cell-site assets

areas above, but Gen AI-supported use cases and

Service Selection and Troubleshooting Prediction of best Optimization of
other AI solutions that help humans with specific tasks orchestration display of complex service to be cost, performance
are included here. network configuration deployed and CX
suggestions
Generate vendor-specific
instructions
According to Anthony Rodrigo, Group Chief Information
Officer, Axiata Group, building machine-learning models Network planning / ML to stitch Network diagnostics Capacity Next-best-action
capacity management together multi-layer to find poorly forecasting from simulation tool
has been a “hobby” development for CSPs over the last topology graphs performing
few years. However, intelligence has been creeping into network elements Support for Creation/updates
Network data predictive models of topology and
the telco in waves, driven by the successful development management architecture
of individual models for a particular task. These tasks Management of
digital twin
include:
Gen AI
Intelligence Design suggestions
Anomaly detection, where problems that have to support Scheduling for field services AI models
Digital assistants
complex data sets and invisible issues cannot be operational Optimization of travel or supply chain Computer vision
Creation and maintenance of
Next-best action suggestions VR or AR used on site visits
teams documentation
easily solved with rules-based models; in this area, Optical character recognition
Declarative statements
machine learning has been applied first in assurance Code creation

and testing TM Forum 2024 (source: Charlotte Patrick Research)

8 inform.tmforum.org
 Autonomous network levels Specialist mathematical tasks such as those used in
the RAN to calculate tilt on an antenna

Simple predictive problems where results are fed

to humans for decisioning; for example, predictive
maintenance.

Virtualization of network functions and service

Fully autonomous network: orchestration are pushing forward the current wave

5
The system has closed-loop automation of intelligence rollouts. Cloud deployments come from
capabilities across multiple services, multiple
domains (including partners’ domains) and the the IT world with some intelligence in their lifecycle
entire lifecycle via cognitive self-adaptation.
automation processes, and with 5G Advanced on the
Highly autonomous network: horizon, intelligence will be needed within the service
In a more complicated cross-domain

4 environment, the system enables decision-making

based on predictive analysis or active closed-loop
management of service-driven and customer
orchestration process to support 5G network slicing.

experience-driven networks via AI modeling The next few years will see the broad deployment of
and continuous learning.
intelligence across NLA areas. Some use cases have solid

3
Conditional autonomous network: ROI, such as the potential to save network capex and
The system senses real-time environmental
changes and in certain network domains opex using models that optimize capacity or improve
will optimize and adjust itself to the external
environment to enable, closed-loop management network design and planning. Other deployments
via dynamically programmable policies. underpin a move up the levels of the autonomous

2 Partial autonomous network:

The system enables closed-loop operations
and maintenance for specific units under
network as defined by TM Forum (see infographic left)
and will take longer to roll out due to the complexity of
certain external environments via statically deploying cross-domain, multi-vendor automation.
configured rules.

1
Assisted operations and maintenance:
The system executes a specific, repetitive
subtask based on pre-configuration, which can
be recorded online and traced, in order
to increase execution efficiency.
The next few years
Manual operations and maintenance: will see the broad
0 The system delivers assisted monitoring
capabilities, but all dynamic tasks must be
executed manually.
deployment of
intelligence across
NLA areas.
TM Forum, 2024

9 inform.tmforum.org
Focus on GenAI When will GenAI / LLMs have a significant
GenAI is currently mainly used in improving human
interactions. A first wave of automation using non-
impact on your business? (APAC responses)
intelligent technologies such as robotic process
automation (RPA) and rules-based analytics has 11%
significantly reduced the time it takes for humans to
complete repetitive tasks. This has helped CSPs make
improvements in areas such as scheduling of field
services.

The introduction of GenAI offers new types of help

including more advanced digital assistants, code n Over the next 1-2 years
creation and anomaly detection in areas such as testing. n Between 2 and 5 years

Most APAC respondents to our AI Benchmark survey n More than 5 years

said they expect the technology to make a significant

58%
difference to their businesses in the next few years
32%
(see pie chart opposite).

The next section analyzes current deployments of AI and TM Forum, 2024

machine learning in Asia-Pacific.

10 inform.tmforum.org
 section 2
how much AI
and machine
learning
are CSPs
implementing
today?

11 inform.tmforum.org
 TM Forum’s AI Benchmark survey finds that 65% of CSP respondents in the Asia-Pacific region
agree with the statement: “We have made good progress in using AI and machine learning in
different parts of our business.” However, the reality on the ground is that a relatively small
number of machine-learning models are actually in production.

The graphic on the next page shows the results of our 5. Network deployment and maintenance including
analysis into APAC CSPs’ use of intelligence in the six predictive maintenance and self-configuration
main NLA processes outlined in the previous section. of the network
This analysis comes from primary and secondary research 6. Intelligence to support humans including digital
material from TM Forum, recent (and past) discussions assistants
with vendors and CSPs, plus publicly available vendor
case studies. These steps were then divided further into the primary
tasks to be completed in each step. For example, service
To conduct the analysis we split each of these six provisioning requires tasks such as “automatic creation
processes into multiple steps: of technical work orders” and “automated testing of
service”.
1. Automated service orchestration including
service provisioning, resource management and Each task was then considered against two questions:
services design
2. Automated and intelligent network planning 1. Does intelligence need to be added? If the task
and capacity management including automated needed AI or machine learning, it was added to others
planning tools within the process that also needed intelligence. Based
3. Intent-based networks and automation in network on this, we calculated the percentage of tasks within
management including domain orchestration power each process that need intelligence (see the primary A relatively small
management, RAN management and axis shown in the graphic). number of machine-
service orchestration learning models are in
4. Automated assurance, trouble management and 2. What is the progress to date? Are the most advanced production in
self-healing including automated resource and fault tier-1 CSPs that are claiming they have deployed the Asia-Pacific.
management, proactive fault management, and functionality really deploying it, or is it still only on
assurance of customer experience vendor PowerPoint?

12 inform.tmforum.org
Our measurements of progress were as follows: Progress of intelligence deployment in NLA
1 = nothing much developed, but discussions seen in
100% 5.00
industry groups and academic settings
2 = some intelligence available from vendors for a 90%
single use case but not as sophisticated as it could 4.50
be (meaning, there are some diagnostic capabilities 80%

available, but a full closed-loop, prescriptive version is

% of processes requiring intelligence

70%
not yet available) 4.00

Progress seen to date

3 = sophisticated intelligence is available from vendors 60%
4 = deployed in a production network for certain
domains or use cases but not everywhere; or deployed 50% 3.50

by a few tier-1 CSPs only

40%
5 = widely deployed.
3.00
30%
The results shown in the graphic reveal that a significant
20%
proportion of processes underpinning NLA (about 60% 2.50
to 70%) could benefit from the addition of intelligence.
10%
The noticeable dip for service orchestration is due to the
availability of simpler automation techniques that only 0% 2.00
1. Assurance, 2. Network 3. Network 4. Service 5. Network 6. Intelligence
require rules-based decisioning. trouble management management deployment and orchestration planning/capacity to support
and self-healing maintenance management humans

The black line on the chart, which shows progress

TM Forum 2024 (source: Charlotte Patrick Research)
to date, gives a score of 3.84 to assurance for the
quite sophisticated use of AIOps in many assurance
deployments. This drops for the other processes, with
some having pockets of deployment but many proving
challenging to deploy. The final area of intelligence used
to support humans has suddenly been brought into
focus with the availability of GenAI but is still relatively
immature in deployments so far.

The next section expands on this analysis.

13 inform.tmforum.org
 section 3

examples of
intelligence
in network
lifecycle
automation

14 inform.tmforum.org
 This section of the report builds on our assessment of the use of intelligence in network lifecycle
automation, providing examples from CSPs and suppliers operating in the Asia-Pacific region.

1. Assurance, trouble management and self-healing Self-healing networks. Self-healing networks

Several trends and requirements are impacting the require a range of monitoring solutions
deployment of intelligence in assurance: that observe the network against a set
of key performance indicators (KPIs) and
Observability across customers and domains. benchmarks. Data is collected when anomalies
The last few years have seen continuing are detected, and intelligence is used to predict and
improvement in network resilience mainly from prescribe remediation. Self-healing capabilities require
the adoption of virtualization and automation. active assurance using time-series databases, with
This means less focus is required on individual storage and monitoring happening in a distributed
network issues in the NOC. Along with preparations for architecture, to support massive amounts of data from
new services that take advantage of 5G standalone’s multiple domains. There is also a need for machine
cloud-native core, these improvements have shifted the learning to predict and prescribe the best way to
focus to service assurance. resolve issues in near-real time.

Data from the network and many other sources (test Getting assurance data to the right
data, weather patterns and customer sentiment, for destination. To meet enterprise customers’
example) is needed in service assurance, alongside new requirements for new services and to ensure
machine-learning models to deal with this increasing satisfaction for all customers, CSPs must
volume of data and to undertake anomaly detection, integrate their data and operations across
The last few
prediction and optimization. In addition, virtualized diverse and complex ecosystems. Delivering the right
data (a mix of assurance, inventory and external third-
years have seen
networks require assurance to span both enterprise IT
and traditional networks, and 5G will increasingly blur the party data) to the right people and processes at the right improvements in
line between the two as enterprise workloads run at the time will provide new visibility and automation to support network resilience
network’s edge. Machine learning models are needed to these new services. from the adoption
deal with the volume of data from multiple domains. of virtualization and
automation.

15 inform.tmforum.org
An example of AI-driven service assurance from An example of a CSP using intelligence in RAN
Thailand’s AIS is featured in TM Forum’s Autonomous management again comes from AIS, which must deal
Networks Journey Guide. AIS designed a proactive with frequent power interruptions during Thailand’s rainy
approach that consolidates near-real-time customer season. Storms trigger thousands of cell outages monthly,
experience data, network events and process insights causing degradation of customer experience and revenue
using AI models to enable proactive intervention and loss. AIS has introduced the Cell Outage Detection and
optimization. The strategic initiative aims to promptly Compensation (CODC) solution to ensure swift detection
detect and address poor customer experience, utilizing of cell outages, precise coordination of compensation and
intelligent and automated integration to identify problem optimized network performance (see graphic).
areas or potential root causes and integrate this capability
into the complaint process to resolve customers’ network
problems and improve satisfaction. Remote repair in cell out-of-service scenarios
2. Network management
Intelligence in network management is needed across 1. Outage detection 2. Cell recovery (D-SON)
domains such as:

RAN management. This has been an area of

progress in intelligence deployment in the last
few years, particularly with the addition of new
machine learning to solve complex problems FAIL

in near-real time. It includes RAN-specific activities Action

• Cell recovery
such as beamforming, channel estimation, interference SUCCEED • Move UE to other cell

management, handover optimization, dynamic spectrum

sharing, scheduling, random access channel (RACH) and 4. Roll back 3. Compensation (MAE-OPT)
physical cell ID optimization. The addition of new machine
Fast RF Adjust
learning and automation offers an opportunity to reduce (<15 min)
ENGINEER REPAIR
opex and improve customer experience.

Published case studies from network equipment providers

offer examples, although the capabilities are most likely
Typically recover
not operational outside the largest global CSPs. outage cell
60% area

TM Forum, 2024 (source: AIS)

16 inform.tmforum.org
The solution constitutes a remote repair mechanism, In addition to device-level power management, CSPs are
which orchestrates a near-real-time response to cell looking at network power balancing, which uses traffic
outages after triggering an alarm when a cell outage steering to provide the optimum network footprint
is identified. The system determines the exact point of for energy-saving, as well as management of power-
traffic impact and selects the most suitable neighboring related assets such as batteries or air conditioning and
cell for load balancing. Mechanisms for optimization of management of power backup to ensure optimum
radio frequency coverage are put in place, along with battery backup for emergencies.
ongoing monitoring of KPIs.
Intent-based networks. The process of
Power management. This is a popular deploying intent-based networks (see sidebar
area for intelligence deployment, given its on p.18) has started with services like network
ability to decrease opex in an era of rising slicing. Still, automation capabilities, new
energy costs. A proof of concept from intelligence and orchestrators are required
Globe Telecom in the Philippines showed how to across the network to increase the kinds of intents that
automatically switch off idle equipment when not in can be fulfilled. Intelligence will be required in intent-
use, using AI and machine learning to understand the based networking to define the behavior expected from
energy utilization pattern of network elements. Using the network, creating end-to-end service goals and KPIs
energy efficiency software-as-a-service, Globe Telecom for each domain. It is also needed to dynamically optimize Learn more about network
automation use cases in
was able to achieve the equivalent of annual power and remediate services; to deal with multiple, potentially
TM Forum’s Autonomous
savings ranging from 3% to 6%. conflicting intents; to calculate the benefit to customer Networks Journey Guide:
experience or cost in taking a particular action; and to
The solution “uses AI/ML to learn the energy utilization control risk by predicting the outcome of decisions.
< Back to Contents

pattern of each network element, like the base station”, AN JOURNEY GUIDE

Gerard Ortines, Head of Network Solutions and Capex Digital Nasional Berhad (DNB), a special-purpose vehicle Autonomous
Management at Globe Telecom, explained to TM Forum.
“In turn, it reduces energy cost without compromising
company and wholesale mobile operator established in
2021 to serve the six national operators of Malaysia, is
Networks
the network performance and is easy to implement… implementing intent-based network operations, with a Empowering digital
transformation –

[The solution] enables dynamic energy consumption goal of ensuring that 5G connectivity is affordable for SEPTEMBER 2023 evolving from Level 2/3
towards Level 4

through advanced algorithms based on real-time Malaysians. DNB is adding automation and intelligence to
network traffic while maintaining premium mobile network management, and the deployment is delivering Objective: to share the latest updates of business
requirements, architecture, capabilities and collaborative

impressive results.
activities, best practices, and reference implementations on

user experience,” he added. Autonomous Networks to enable value-driven transformation.

17 inform.tmforum.org
For example, auto-analysis and correlation of alarms and
actuation in the network led to a 500% reduction in alarm
count six months after its introduction, and AI-based
predictive 5G network management using network data
resulted in network uptime greater than 99.8%.

DNB implemented a trial of a new intent-based model to

handle RAN partitions for each RAN cell site, while also
managing the network’s resources. This resulted in:

8% to 10% better throughput of premium services with

no intent violations Defining ‘intent’ and showing how to implement it
Avoidance of 12 SLA breaches per cell per day
15 to 17 hours of partition overuse prediction shared In TM Forum’s Autonomous Networks Project, CSPs and their suppliers are working
with the customer per week on how to implement intent-based automation and integration between autonomous
Network slices with individual guaranteed SLAs. domains. They are building on and extending previous standards work on intent
wherever possible. For example, the team’s definition of intent-based networking is
3. Network deployment and maintenance compatible with that of the Internet Engineering Task Force (IETF).
Opportunities to deploy AI and machine learning to
assist with network deployment and maintenance can be The systems in autonomous networks are governed by intent, which sets expectations
divided into three main groups of use cases: as requirements, goals and constraints. These are abstracted from the technical inner
workings of the network. Put more simply, intents are the “what” not the “how” –
Planning and deployment management meaning, you tell the system what goal or outcome is required without having to tell it
tools. These tools enable all processes around how to achieve it. This decoupling increases agility, allowing autonomous systems to
site deployment – from site selection and fulfill the intent using data, machine learning and AI, while enabling supplier innovation
surveying to supply chain management. that can be easily integrated and adopted.
Vision recognition tools, for example, gather images and
information from drones about cell site configurations. Intent can be expressed at various levels (business intent, service intent and resource
This data can then be input into a 3D model or digital intent, for example). In the TM Forum AN Reference Architecture intents are used with
twin of the site, enabling engineers to make engineering closed-loop management to automate the full lifecycle of services. Closing the loop
decisions remotely based on the model. means collecting and analyzing data to figure out how networks can be optimized,
and then implementing those changes in an automated way. To learn more about
TM Forum’s work on intent and autonomous networks, please contact Alan Pope.

18 inform.tmforum.org
 Analytics and machine learning can also 4. Service orchestration
optimize tools for the allocation of design and Automation has been available for some time within
field resources needed for making changes and certain parts of the service lifecycle such as service
emergency repairs to the network. Coverage assurance, but the end-to-end provisioning and
optimization tools allow antenna attributes such deployment of the service on the network depends on
as height, tilt and azimuth to be calculated from the availability of suitable orchestration and the ability to
a three-dimensional photograph, and tools for modify multiple network elements. With the introduction
capacity auditing and acceptance compare of service orchestration solutions, vendors are articulating
sites’ current performance with baseline goals, intelligence use cases for:
capturing possible issues such as installation
problems or equipment defects. Prescriptive and predictive capabilities to enable
deployment of the complex set of new network
Ongoing site management tools. Intelligence technologies, devices and features
solutions in this area include “humans in the Detection of unusual patterns or outliers in
loop” with use cases such as technicians inventory data
running power into a cell site, installing Understanding patterns of change in the inventory
equipment, connecting cables and troubleshooting database to provide information on how the network
connectivity. is evolving
The ability to define new services using GenAI –
Testing. CSPs are beginning to include humans would use declarative statements to provision
testing as part of closed-loop service and new services, such as “provide a connectivity solution
network orchestration, allowing end-to-end for secure communication from point A to point B”.
service orchestration and rapid deployment
of network functions and devices. Intelligence provides Telstra has been using a service orchestration solution
anomaly detection for diagnostics and a range of and TM Forum Open APIs to integrate new domains
predictions, including future performance and potential such as cloud and SD-WAN with established wireless and
failure scenarios. transport domains, with an average seven day shorter
order fulfilment time. Service composition is first defined
in a declarative template, with the solution automatically
designing the service based on specified parameters such
as latency before instantiating it. It can modify or terminate
the service using workflows dynamically generated from
the service template and provide end-to-end management
and visibility of services.

19 inform.tmforum.org
Closed-loop operation then enables the composite Intelligence used in network planning and capacity management
domain to maintain the required service level throughout
the lifecycle of the service as conditions and capabilities
change in the underlying domains. Visualization & Design Analysis & Diagnostics

5. Network planning / capacity management Machine learning to provide an Detailed traffic analysis to find
Planning and capacity management are areas where understanding of the network and high-value areas and bottlenecks
optimum connectivity of network
intelligence and automation can result in significant capex
components based on factors like
and opex savings. The past four to five years have seen Network diagnostics providing
cost or customer experience
real-time fault analysis and
the emergence of the first tools adding intelligence to understanding poorly performing
the planning team’s activities. The infographic opposite Improved network dimensioning network elements
(determining how to meet minimum
describes the typical use cases for intelligence in network capacity requirements at peak hour)
Load balancing predictions used
planning and capacity management. Sanders notes that in demand forecasting and long-term
Telstra’s team tended to be very experienced but was still Simulation tools including ‘what-if’ predictions of network behavior
analysis and future use of digital twin
using multiple manual planning and design documents,
which created a large labor cost to support.

6. Intelligence to support operations teams

Support for operations teams historically has included RPA
to complete simple tasks at the right moment for a human.
These capabilities have expanded over time to take on
many repetitive tasks, particularly in large teams where
the time savings add up to a significant cost benefit. The
appearance of GenAI offers new functionality via simple,
digital assistants. Principally the use of natural language
understanding and generation brings a new sophistication Optimization RAN Planning
in human-machine communications.
Sophisticated models for optimization Reinforcement learning (e.g., the
(e.g., a desired base station location is autonomous configuration of tilt
There are no publicly available examples yet of CSPs unavailable due to a lack of electricity and power of antennas in a business
or the landowner refusing access; this district to achieve the best possible
using GenAI or other LLMs in network operations. means that the optimal network may coverage or quality per the mobility
Indeed, TM Forum’s research for the AI Benchmark found not be possible, but machine learning patterns at that time)
that network operators are largely skeptical about the can be used to revise network designs
based on limitations)
technology’s usefulness in the day-to-day running of
telecoms networks.
TM Forum 2024 (source: Charlotte Patrick Research)

20 inform.tmforum.org
“I’ve met a lot of network experts in this industry that Onsite technology. There are a range of newer
are trying to find some real-time data monitoring or root AI technologies for field technicians undertaking
cause analysis use cases for GenAI. But I don’t think it’s tasks such as running power, installing equipment
possible at this moment,” Chung Suk-guen, Chief Global connecting cables and troubleshooting
AI Officer at SK Telecom, said during an interview for the connectivity, which require intelligence. For
report. “GenAI is not fast enough to handle tons of data in example, video-based tower inspections can
real time. And also, the data is not human readable.” be conducted using a drone to stream video to
the ground operator’s mobile device. This uses
That said, there is a range of (mostly) AI-related computer vision to detect issues with the radio
functionality that could be added in the future to support tower installation in real time.
operations teams, such as:
Computer vision can also identify potential
Support for more complex problems. problems with cable weatherproofing or color
Increasingly, intelligence will be built as digital coding (to identify crossed cables). This AI
assistants to support more complex problem can be augmented with a convolutional neural
solving. These will use data from inside and network (CNN) to identify “bad” versus “good”
outside the organization, as needed. cabling installations. And optical character
recognition can locate the sticker or tag area
More intelligent interactions. Here, machines in images to bring back information about the
start to interact on a more ad-hoc basis, offering details of installed equipment.
suggested corrections (or making automatic
corrections) where human error has crept in. Finally, deep neural networks could also check
A machine may also proactively undertake tasks for installation issues: Are cables neatly bound?
as it perceives a need based on human activity Is equipment that was expected to be deployed
in the recent past. evident in photos of the base station taken as the
engineer leaves?
Support for specialist tasks. GenAI is used
to support declarative statements (allowing The next section offers CSPs advice for increasing Increasingly,
humans to express their high-level objectives in deployment of AI and machine learning in operations.
natural language). It can also be used for code
intelligence will
creation (for example, real-time creation of API be built as digital
script), or it may be involved in the creation and assistants to support
maintenance of documentation. more complex
problem solving.

21 inform.tmforum.org
 section 4

make it happen –
strategies to
take advantage
of new
intelligence

22 inform.tmforum.org
 CSPs operating in the Asia-Pacific region are shifting the focus of intelligence deployment
towards the task of industrializing AI practices. In doing so, they are moving from an early
phase of experimenting in multiple AI projects to developing and implementing an
AI and machine learning strategy that enables them to make good decisions commercially
and technically. Here are four actions that require immediate attention.

Determine the real benefits of implementation intelligence skill sets without needing complex
Creating intelligence roadmaps is difficult. A cross-domain data gathering
large CSP may have developed hundreds of
machine learning models, of which only a small Place within a particular value stream – again,
handful are currently in useful deployment. the project may not bring much immediate
One executive we spoke with commented on financial value but may be an integral part of a
the disjointed approach to AI that is often seen wider NLA goal that will fail without it
today and the tendency to focus on the next new
and exciting technology. Intelligence projects Board-level enthusiasm – many early
should be selected based on a mixed set of intelligence projects have failed to develop
measurements, including: the necessary models, so the highest levels of
management may need to evangelize AI more
Financial value to the organization, both widely across the company.
individually and as part of the total vision
for network lifecycle automation Take a mature approach to road
map development
Immediate feasibility given the maturity of Building out a new technology without an
models and experience within the organization overarching vision risks creating a whole new
layer of technical complexity and debt. CSPs
Immediate usefulness for building knowledge should take a step back to understand any
and experience – for example, small domain- weaknesses in their current road maps for using
specific automation that includes machine intelligence. Questions to ask include:
learning may not offer a huge financial upside
but could provide an opportunity to build

23 inform.tmforum.org
 Have we selected the simplest possible solution Create a robust intelligence discipline
to each problem, given that this may also be the One of CSPs’ key strengths is their background
most cost effective and have the highest chance in developing sophisticated capabilities in
of success? network engineering. This kind of thinking will
also be needed when deploying intelligence
Are we moving swiftly down the path of into NLA. The creation of a small, centralized
creating reusable data assets? Without these team supporting intelligence and automation
there is a risk of creating silos of data or efforts should be considered. The team might
duplicating requests from the network or other include technical experts and business-focused
equipment for the same data. employees responsible for ongoing development
of vision and strategy, business cases, and risk
Do we need a plan (in conjunction with others management.
in the industry) to introduce a more open
ecosystem around intelligence that limits the An example of this type of team has been
reduction in the value of our data by allowing running in Telstra for the last five months.
third parties, such as vendors, to benefit from The team begins by reaching out to areas
ingesting it into their models? where intelligence and automation could be
deployed to answer questions and demonstrate
Is there an evaluation process that removes potential value to skeptical team members. It
in-progress projects where the intelligence is brings technical expertise (as needed) as well
failing? This could be because the problem as commercial expertise, drafting business
requires more mature models than used or cases and analyzing the operational benefits of
because the models are proving problematic (for projects.
example, they are prone to creating subtle and
difficult-to-spot errors).

One of CSPs’
key strengths is
their background
in developing
sophisticated
capabilities in network
engineering.

24 inform.tmforum.org
 Build an intelligence-native team New job titles should be considered for
As the dust settles around the hype of addition to the organization. For example,
GenAI, CSPs are beginning to understand the the “knowledge engineer” will obtain
technology’s uses and limitations (some of understanding from the process expert in the
which may be permanent or temporary until operating unit and incorporate this human
more sophisticated or different models become understanding into the intelligence within
available). This knowledge, backed up with automation or digital assistants. Staff issues,
practical experience from deploying AI models including insufficient headcount or expertise,
into NLA will need to be instantiated in the remain a significant headache for APAC
organization and spread across departments CSPs, particularly smaller organizations. Our
to reuse it. Actions might include encouraging research finds that there is not one agreed-
individuals with an early interest in testing upon path to overcoming these problems,
intelligence technologies to get together with CSPs expecting to build up new skill sets
in informal groups to explore topics and through recruitment, reskilling, and acquiring or
exchange ideas. partnering with specialist AI firms.

Our AI Benchmark report discusses some of

the new AI skills needed, and an upcoming
report focusing on how telco-to-techco
transformation impacts people will look at
where CSPs can find the talent they need.

Staff issues,
including insufficient
headcount or
expertise, remain a
significant headache
for APAC CSPs.

25 inform.tmforum.org
 tm forum
open digital
framework

26 inform.tmforum.org
A blueprint for intelligent operations fit for the 5G era

The TM Forum Open Digital Framework provides a

migration path from legacy IT systems and processes to
modular, cloud native software orchestrated using AI.
The framework comprises tools, code, knowledge and
standards (machine-readable assets, not just documents).
It is delivering business value for TM Forum members
today, accelerating concept-to-cash, eliminating IT
and network costs, and enhancing digital customer
experience. Developed by TM Forum members through
our Collaboration Community and Catalyst proofs
of concept and building on TM Forum’s established
standards, the Open Digital Framework is being used
by leading service providers and software companies
worldwide.

Core elements of the Open Digital Framework

The framework comprises TM Forum’s Open Digital
Architecture (ODA), together with tools, models and data Maturity tools & data
that guide the transformation to ODA from legacy IT Maturity models and readiness checks to
systems and operations. baseline digital capabilities
Data for benchmarking progress and training AI
Open Digital Architecture
Architecture framework, common language and Goals of the Open Digital Framework Learn more about
design principles The Open Digital Framework aims to transform business collaboration
Open APIs exposing business services agility (accelerating concept-to-cash from 18 months If you would like to learn
Standardized software components to 18 days), enable simpler IT solutions that are easier more about the project or
Reference implementation and test environment and cheaper to deploy, integrate and upgrade, and to how to get involved in the
establish a standardized software model and market TM Forum Collaboration
Transformation tools which benefits all parties (service providers, vendors and Community, please contact
Guides to navigate digital transformation systems integrators). George Glass.
Tools to support the migration from legacy
architecture to ODA

27 inform.tmforum.org
 tm forum
research
reports

28 inform.tmforum.org
 February 2024 | www.tmforum.org February 2024 February 2024 | www.tmforum.org
February 2024
February 2024 TM Forum | December 2023 TM Forum | December 2023

REPORT
REPORT
REPORT

closing
REPORT REPORT

closing
the
theloop
loop a numbers game:
CSPs aim to exploiting data to BSS
CSPs aimservice
automate to
automate service
orchestration and drive future for B2B operators generative
strategies
AI:
orchestration and
assurance diverge on the
assurance path to cloud
Author: Sponsored by:
Dean Ramsay, Principal Analyst Author: Sponsored by:

operators
Author: Sponsored sponsored
by: by: Richard Webb, Senior Analyst sponsored by:
Dean Ramsay, Principal Analyst
Editor:
Author: Dean Ramsay, Practice Lead, TM Forum Author: Richard Webb, Senior Analyst Editor:
Ian Kemp, Managing Editor
Editor: Ian Kemp, Managing Editor Editor: Ian Kemp, Managing Editor
Editor: Ian Kemp, Managing Editor

take their
Ian Kemp, Managing Editor

first steps

Author:
Author:
Ed Finegold, Contributing Analyst
Mark Newman, Chief Analyst

Editor:
Editor:
Dawn Bushaus, Contributing Editor
Ian Kemp, Managing Editor
September 2023 | www.tmforum.org August 2023 | www.tmforum.org
September 2023 | www.tmforum.org September 2023 | www.tmforum.org August 2023 | www.tmforum.org
Sponsored by:
TM Forum | September 2023 August 2023 Sponsored by:
TM Forum | December 2023
August 2023

REPORT

Reigniting
REPORT
REPORT BENCHMARK REPORT

levelingup:
up:
REPORT

Cybersecurity strategies: leveling wholesale how software-as-a-service

wholesale
REPORT

cybersecurity strategies: achievingLevel

Level33 changes: telecoms growth
the telco risk management moves
firmly into the telcomoves
spotlight
achieving
autonomousnetworks
autonomous networks changes:
wholesale is reshaping
sustainable revenue
risk management
firmly into the telco spotlight
andbeyond
and beyond changes:
rethinking
rethinking
for new
support systems
support systems
fiber operators
a Playbook business support
systems
for CEOs
telco:
for new fiber operators
growth: Author:
Patrick Donegan
Principal Analyst
HardenStance
rethinking support systems
for new fiber operators
Author:
Mark Newman
Chief Analyst
Sponsored by:

sponsored by:
Author: Sponsored by:
Editor: Mark Mortensen, Contributing Analyst Sponsoredsponsored
Author: by: by: Editor: Authors: sponsored by:

navigating the
Dawn Bushaus, Author: Sponsored by:Thinking Labs
Mark Mortensen, Contributing Analyst Dawn Bushaus, Sangeet Paul Choudary, Platform
Contributing Editor sponsored by: Contributing Editor Mark Newman sponsored by:

time for
TM Forum Editor: Chief Analyst
Nik Willetts, CEO, TM Forum
Author: Patrick Donegan, Principal Analyst, HardenStance Author: Dr. Mark H Mortensen, Contributing Analyst, TM Forum Author: Mark Newman, Chief Analyst Anthony Rodrigo, CIO, Axiata Author: Ed Finegold, Contributing Analyst
Dawn Bushaus, Contributing Editor
Editor:

maze of scope
Editor: Dawn Bushaus, Contributing Editor, TM Forum Sponsored by: Editor: Dawn Bushaus, Contributing Editor, TM Forum Editor: Dawn Bushaus, Contributing Editor September 2023 | URLEditor:
TBC Dean Ramsay, Principal Analyst, TM Forum Editor: Ian Kemp, Managing Editor
Dawn Bushaus, Contributing Editor
September 2023 | URL TBC Dawn Bushaus,

operators to Contributing Editor

3 emissions place new bets

September 2023 | URL TBC

Authors:
Mark Newman, Chief Analyst
Author: Dean Ramsay, Principal Analyst
Joanne Taaffe, Editor in Chief, Inform

Editor: Editor:
Ian Kemp, Managing Editor Ian Kemp, Managing Editor
June 2023 | www.tmforum.org
June 2023 | www.tmforum.org June 2023 | www.tmforum.org
June 2023 | inform.tmforum.org
TM Forum | June 2023 TM Forum | May 2023

standout
REPORT REPORT REPORT

REPORT

strategies:
REPORT REPORT
CSPs take
Digital key steps to
transforming how telcos are
Transformation modernize BSS:
Tracker 7 network inventory
innovating in a
crowded market making
cutting
racing to a flexible,
customer-focused future waves: partner
complexity the future for
with automation
Author: Mark Mortensen

Editor: Dawn Bushaus

Sponsored by:
sponsored by:
Author:

Editor:
Sponsored by:

sponsored by:
Author:

Editor:
Sponsored by:

sponsored by:
Open RAN
technology
ecosystems:
and AI
Author: Dr. Mark H. Mortensen, Contributing Analyst
ISBN: Author: Teresa Cottam, Contributing Analyst Author: Mark Newman, Chief Analyst
Editor: Dawn Bushaus, Contributing Editor

who does what

Editor: Ian Kemp, Managing Editor Editor: Ian Kemp, Managing Editor
ISBN: ISBN:

and can CSPs

compete for more?
Author:
Dean Ramsay, Principal Analyst, TM Forum

Editor:
Ian Kemp, Managing Editor, TM Forum
Author: Author:
Dawn Bushaus, Contributing Editor October 2022 | www.tmforum.org
Ed Finegold, Contributing Analyst
November 2022 | inform.tmforum.org December 2022 | inform.tmforum.org December 2022 | inform.tmforum.org
March 2023 TM Forum | March 2023
Editor:
Ian Kemp, Managing Editor Sponsored by: Dawn Bushaus, Contributing Editor

Sponsored by: REPORT BENCHMARK

the
REPORT REPORT
REPORT REPORT

DIGITAL OPERATIONS
MATURITY:

establishing links:
counter a roadmap for sustainable evolving achieving
platform models
in the Open intelligence Asia-Pacific
telecoms
5G core: telco:
engineering
business support systems
business value
from transformation
API economy using AI to
improve growth exploring CSP networks for for future
sponsored by:

net zero services

customer approaches
Author: Teresa Cottam, Contributing Analyst
Editor: Dawn Bushaus, Contributing Editor

experience
Authors: Author: Author:
Mark Newman, Chief Analyst, TM Forum Joanne Taaffe, Editor in Chief, Inform Mark Newman, Chief Analyst
Dawn Bushaus, Contributing Analyst, TM Forum

Editor: Author: Editor: Editor:

Ian Kemp, Managing Editor, TM Forum Teresa Cottam, Contributing Analyst Ian Kemp, Managing Editor, TM Forum Ian Kemp, Managing Editor

Dawn Bushaus, Contributing Editor

Authors: Sponsored by:
Sponsored by:
Authors: Dean Ramsay, SPONSORED BY: Sponsored by:

29 inform.tmforum.org
Mark Newman, Chief Analyst, TM Forum Principal Analyst
Dawn Bushaus, Contributing Analyst, TM Forum
Sponsored by: Joanne Taaffe, Editor in Chief, Inform Rahul Gupta,
Senior Analyst
Editor:
Ian Kemp, Managing Editor, TM Forum Editor:
Ian Kemp, Supported by:
Sponsored by: Managing Editor
 meet the
research
& media team

30 inform.tmforum.org
Meet the Research & Media team

Report Author: Report Editor:

Charlotte Patrick Dawn Bushaus Published by:
Contributing Analyst Contributing Editor
dbushaus@tmforum.org European Office
TM Forum, Uncommon,
34-37 Liverpool Street,
Managing Editor: Editor in Chief, Inform:
London EC2M 7PP
Ian Kemp Joanne Taaffe
ikemp@tmforum.org jtaaffe@tmforum.org United Kingdom
Phone: +44 207 748 6615

US Office
Chief Analyst: Global Account Director:
Mark Newman Carine Vandevelde 181 New Road, Suite 304,
mnewman@tmforum.org cvandevelde@tmforum.org Parsippany, NJ 07054 USA
Phone: +1 862-227-1648

Practice Lead: Marketing Manager: ISBN: 978-1-955998-81-9

Dean Ramsay Ritika Bhateja
dramsay@tmforum.org rbhateja@tmforum.org Report Design:
Paul Martin

Head of Operations: Digital Media Coordinator:

Ali Groves Maureen Adong
agroves@tmforum.org madong@tmforum.org

Commercial Manager:
© 2024. The entire contents of this
Tim Edwards
publication are protected by copyright. All
tedwards@tmforum.org rights reserved. The Forum would like to
thank the sponsors and advertisers who
have enabled the publication of this fully
independently researched report. The
views and opinions expressed by individual
authors and contributors in this publication
are provided in the writers’ personal
capacities and are their sole responsibility.
Their publication does not imply that they
represent the views or opinions of TM Forum
and must neither be regarded as constituting
advice on any matter whatsoever, nor be
interpreted as such. The reproduction of
advertisements and sponsored features in
this publication does not in any way imply
endorsement by TM Forum of products or
services referred to therein.

31 inform.tmforum.org
To find out more about TM Forum's
AI and data work please contact
Yvonne Kuimba