Rakuten Mobile was the first greenfield mobile network to fully embrace Open RAN and a software-based architecture. It aggressively invested in automation, eventually running the entire Japan-wide network with only 250 network engineers.
Process, tools and people challenges emerged along the way. We know our customers will face the same hurdles as they embark on digital transformation journeys. We have identified nine challenges and realities that are likely to appear. Here’s our view on what to expect and how to get ready.
AI needs high quality data so it is important to have a unified data strategy whereby all data is discoverable, understandable, organized and sharable. Data governance requirements play a role, necessitating a unified data platform with controlled, secure access. This is especially true with sensitive data that introduces regulatory consequences tied to privacy laws and initiatives.
When an AI model is written and the results are computed, a user interface that allows business users to make sense of the results and incorporate them into network or business activities is a must.
Ideally, these models would be embedded into the tools users access daily. Think about a network operations center (NOC) engineer that could use AI to provide insight on a trouble ticket. If that AI insight is not available on the screen of the trouble ticket application, then it's of little use to the engineer. The answer is not always to increase the number of tools that a user needs to collect and use AI, but rather to embed those insights into the existing tools.
Having a backlog of use cases separates the MNOs that talk about implementing AI from those that are serious about it. A backlog means management has analyzed the network and the business operations to uncover problems that can be solved using AI.
It is the start of the AI transformation journey. With a backlog, an MNO can determine the priority of the projects based on network or business system impact. This analysis can also guide decisions around personnel and AI tools.
One way to create the backlog is to insert a new step into the incident review process. As the team reviews an incident and answers questions about why it happened and how it can be prevented, the question of “How can AI solve this?” can be broached. This helps develop an organizational understanding of how AI can be used.
It is important to establish trust and security around AI use cases to guide the use of AI and ensure it remains a positive and trusted tool for solving problems. Trust and security are related to data governance but stand apart in our list due to importance to customers. Proper security will vary by tool but is important to build into all AI systems and processes.
Beyond security, it’s important that data scientists access, implement or use data and insights in a way that aligns with trust, security and even responsibility principles.
One real world example is a use case we are implementing to better manage battery backups during natural disasters. Earthquakes, typhoons or other disasters can cause power outages that take down mobile networks. We’re using AI to understand how we can reconfigure the mobile base stations in an affected area in order to save battery power.
We could use AI to maximize battery power by cutting the transmit power thus reducing transmission radius dramatically. But the AI system needs to balance this battery power optimization with the need for people to connect during an emergency. By focusing too much on extending battery life, an operator could lose the trust of users who find they can’t access the network just when they need it most.
Creating an AI model is a multistep process that includes discovery, training, testing, evaluating and deploying. Today these steps take place in siloed tools or on a data scientist’s laptop. This provides no traceability or tracking and it can slow down the model development process significantly as data scientists have to search for the data they need.
In our experience, it can take up to nine months for a data scientist without a platform to go from a proof of concept to production. But an AI platform can cut that time to weeks because all data is in one place and the platform provides a single workbench for an entire project.
What’s harder than finding a qualified data scientist? Finding the GPUs and compute power they need to do the job. There is a scarcity of GPUs and it’s an industry wide problem.
One advantage of working with Rakuten Symphony to build AI models is that the company has already trained some of the most useful models based on data from Rakuten Mobile. These models don’t need as much compute power because they have been trained on Rakuten’s compute infrastructure which is also available to customers for training new models.
We have access to the infrastructure that powers Rakuten Group which includes banking, e-commerce and other digital businesses. The company runs its own hyperscaler compute infrastructure. Rakuten Symphony can access this infrastructure to help solve unique MNO problems.
AI is good at identifying problems and giving insights, but graduating to a Level 5 autonomous network, requires an MNO’s AI solutions be integrated into the tools users are already using in order to convert AI insights into actions.
The way to get this integration is through standardized APIs that allow the AI model to feed alerts and fixes to southbound operations support systems that can take action to solve the problem. This also means the AI tool must have standards compliance from industry leading standards bodies such as O-RAN Alliance, 3GPP and TM Forum. It’s important to select an AI platform that complies with telco standards, otherwise the insights will not be autonomously actionable.
Data scientists for telecom are a rare breed. To become one, a data scientist needs to be trained in AI model development and they also need domain expertise to understand the specific problems facing telcos and the data models available to fix those problems. This shortage is a challenge in every geography around the world with the possible exception of Silicon Valley.
Rakuten Symphony can help here as well by providing insights into customer network challenges from our data science team that has been embedded for years in the MNO industry, becoming experts in solving specific MNO AI challenges.
The skill of the data scientist and the quality of the AI model are equal only to the quality of data as inputs that lead to a successful AI outcome. But data scientists, by and large, don’t have telecom specific data to be able to fine tune models. In building our AI models, we relied heavily on our access to data from Rakuten Mobile which provided us with large quantities of real-world mobile network data. This allows us to build high-quality models that can be deployed quickly.
Revolutionizing your network with AI involves changing many of the processes, personnel and tools that have made the network run for many years. The scope of the change is bound to cause some speed bumps along the way, but armed with the right strategy, our industry can take them head-on in swift pursuit of AI-powered transformation.
For more information on how autonomous networks are reshaping telecom, be sure to subscribe to our Zero Touch Telecom newsletter on LinkedIn.
