Telecoms operators face the significant challenge of upgrading established systems to make use of AI insights and will need AI development teams that include members with in-depth knowledge of the telecoms industry.
Exaggerated claims about artificial intelligence (AI), specifically machine learning (ML) and deep learning (DL), are setting expectations for automation and reduction of human-based processes in the telecoms industry too high. Telecoms operators assume that AI can easily provide extreme automation to create new sales, reduce costs and improve customer experiences. However, significant challenges need to be addressed before these ambitions can be achieved. The two most common challenges that operators face are a lack of skilled staff and the poor quality of data available.
However, a more significant challenge is that of upgrading established applications and processes to make use of ML/DL insights. Legacy systems, processes, and culture will need to be updated or replaced at considerable cost and effort if operators are to benefit from any insights that AI systems can create.
The potentially millions of lines of code, carefully built processes and data flows all need to be reworked if data-driven AI automation is to be applied. Effectively, every process will need to be upgraded or transformed. In short, discovering an insight is an interesting academic exercise, but it needs to be acted on if it is to become useful to the operator.
Telecoms operators often view vendors as part of the 'problem' rather than the solution
Established telecoms-focused vendors are often seen as "part of the problem" because vendors have not always been able to incorporate ML/DL technology quickly enough at the right price and their solutions do not provide clear enough benefits for the telecoms operators. Vendors that are new to telecoms, or those that sell general-purpose AI tools, tend to be more likely to challenge the current systems, replace established technology and promise significant rewards for doing so.
The popularity of open-source tools is an added problem for established vendors. The low cost of these tools undermines the traditional business models of software vendors, inhibiting their ability to generate enough profit to finance new developments. The plethora of new technologies, companies and business models creates uncertainty in purchasers' minds. This has created a high level of industry 'noise' leaving operators paralysed in their decision making and unwilling to wholly adopt ML/DL technologies or software tools that may soon be superseded.
Vendors need to have telecoms expertise as well as skills in ML/DL
Telecoms-focused vendors, such as Ericsson, Huawei, and Nokia, have significant value to add for telecoms operators that want to adopt more ML/DL solutions. Vendors that specialise in ML/DL solutions are interested in selling to telecoms operators, but many of them have limited expertise in the telecoms industry. A lack of understanding of the industry can result in poor yields from ML/DL projects. The skills needed to understand the nuances within telecoms OSS/BSS operations are as rare as the skills needed for data science and AI. Skilled data science will continue to require a knowledge of what the data means, which data is available and where insights are needed. Without this knowledge, projects will fail or never progress from proof-of-concept (PoC) trials to deployment.
A typical scrum development team consists of a product owner, a scrum master and other team members. Such teams within telecoms-specific vendors include subject-matter experts, but vendors need to add data scientists to these teams. Data scientists can test core logic using data analytics or advanced ML/DL. Even before the development of a minimal viable product (MVP), concepts can be checked and validated where access to appropriate training data is available. This DevOps approach is sometimes referred to as AIOps when AI techniques are used. If a non-telecoms-specific vendor is used, the team will need to include an additional staff member to provide domain-specific skills. Where a telco is willing to support a vendor without telecoms skills in a co-operative development team, the telco can provide the telecoms knowledge.
During the development stage, programmers can call algorithms or insights as functions to provide logic for the use case, applications or workflow, effectively replacing potentially time-consuming and complex hard coding of logic in traditional development tools. In some cases, the complexity that can be achieved is beyond viable coding costs and timescales.
More significantly, the use of ML/DL can also provide an ability to continuously update logic as data inputs change. This 'soft' ruling allows for logic to evolve over time without necessarily needing traditional software releases or upgrades, and enables applications to learn from each unique environment in which they are deployed helping to reduce some of the costs associated with implementation.
Operators need to start small and build up skills incrementally if they are to successfully adopt AI solutions
AI tools have many hundreds of potential use cases and projects. Using AI to augment established processes rather than using it to change processes is often the easiest path to value. This 'augmenting' approach can be used to locate key data sources, build knowledge and test new tools, and may also help to overcome fears of AI replacing roles within organisations. Staff members are used to validate recommended actions, over time the recommendations become automated and staff can focus on more complex issues to validate.
Operators have a long journey ahead. Nokia suggests that AI solutions are just 1.5 years into a 10-year journey towards maturity. This is realistic, given the billions of lines of code and the increasing complexity of services and infrastructure, and daunting for most large enterprises including operators. However, operators can be assured that there are many short-term wins that can add value incrementally and do not require radical transformations.