Data Strategies That Drive Revenue Growth

AI excitement remains rightfully high. Recognition of unending potential has sent stakeholders racing off in virtually every direction in efforts to conquer challenges and pursue new opportunities.??

In this piece, we take you behind the curtain of the AI strategies we have deployed to target revenue growth and expand revenue streams, discussing how they can be applied to global telco efforts.??

As with any AI strategy, effectiveness and outcomes are directly tied to the quality of data, robustness of data preparation and implementation of sound governance strategies.??

Dataset decisioning: Top down or bottom up??

Every great AI outcome begins with great data. How that data will be stored, ingested and accessed has implications on the entirety of the work that will follow.??

We considered data integration and management from two perspectives: top down and bottom up. This starting point gave us a structured methodology to address data preparation and governance complexity while ensuring high-quality and effective management. It forced decisions around all data management aspects, from acquiring and integrating high-value datasets to optimizing existing processes and workflows. It also provided an opportunity to weigh the benefits of each so that a balanced approach could be taken in service to strategic and operational needs.??

Our top down approach focused on ingestion of high-value datasets, including:?

• Deep Packet Inspection (DPI) data providing insights into what users are doing.?

• Call Detail Records (CDR) data that logs how much activity is happening, where it is happening and the billing info associated with it.??

• Customer experience data that shows the value of services to customers and can include highly sensitive information.??

The goal is to make these datasets universally accessible to enable comprehensive data analysis and AI model training for actionable insights and strategic decision-making. Our understanding was to partner universal access with strong access governance. ?

Our bottom-up approach is focused on reinventing existing processes, tooling and workflows to find new efficiencies and performance enhancement. In contrast to the top-down approach, here we started with “what can we do better,” driven by a desired outcome. ?

Remote Electrical Tilt (RET) was one of the use cases we focused on. Mobile operators continuously make RET adjustments to optimize mobile network performance, maximize coverage and capacity, and provide great customer experiences. Our goal was to optimize and automate this process with daily instead of weekly adjustments.??

The step-by-step approach included:?

1. Identifying the desired outcome.?
2. Figuring out what data was required to achieve it.?
3. Determining availability of the data and ways to acquire it.?
4. Ensuring the data was correctly formatted and met quality standards.?
5. Develop an automated approach to ingesting and analyzing it with continued iteration.?
6. Implement data privacy measures and access permissions.?

When we focus on energy efficiency, we take the same approach. It is important to understand though that each model seeks a different outcome, requiring different data and different access. That is the beauty of the power of the space. ? ?

Challenges on our AI-powered journey?

Despite not being hamstrung by legacy processes and infrastructure, Rakuten Symphony still faced numerous challenges on our AI-powered automation journey. In this way, we are no different from other operators and vendors who face similar challenges. It is the nature of dealing with massive datasets and using them in new ways. ?

We immediately faced challenges related to siloed data lakes and traditional vendor-led governance structures. We knew the end state was a scalable data lake that had already been proven by technology leaders beyond our industry.?

Even simple tasks like comparing fault and performance data to identify correlations wasn’t possible because each application had its own siloed storage.??

When it came to governance, managers didn’t have a clear view of who was using which data sets. This introduced challenges in tracking data usage and revoking access when necessary. The high volume of requests and the approvals surrounding each became overwhelming. ??

We understood that a system-level governance system could help us overcome these hurdles and provide a foundation for robust data governance policies that better managed access. ?

Scaling a unified approach?

Harnessing AI’s power and overcoming the data challenges we faced hinged on scaling a unified data lake.??

At Rakuten Symphony, we prioritized a vendor-independent data model. S3 compliance was important because it ensured standardized ways of storing and accessing data, including seamless API plug-and-play with applications and third-party vendors. This was a critical component of creating a comprehensive data repository to support diverse AI use cases—even ones we hadn’t yet conceived of.??

This access simplification had to occur while simultaneously increasing security given sensitive DPI, CDR and customer datasets require stricter governance protocols. We evaluated every data access policy, implementing robust governance frameworks to manage access permissions and protect sensitive data. ?

Next, it was time to empower our data scientists to succeed within this new governance structure.??

Setting up our data scientists for success?

The right storage and access policies go a long way to helping teams be successful. We implemented several strategies and tools to enhance capabilities of our data science teams, streamline model development and deployment and significantly reduce time to market.??

• AI studio for model development and deployment. Our AI studio gives our data scientists tools and infrastructure to efficiently develop, deploy and manage AI models. It includes the libraries they need to develop solutions and infrastructure that hosts models and fosters creation of recurring pipelines.??

• Scalable Machine Learning Operations (MLOps) to manage the full model lifecycle. Scaling MLOps capabilities is an important part of full lifecycle management of AI models, including registry deployment, inferencing and performance monitoring so any degradation can be addressed through model retraining.??

• Reducing AI model implementation time from months to days. The process of determining data needed for a certain use case, getting access approvals and moving it to storage could take weeks or even a month. The same is true for development and deployment. Our new approach minimized non-value added processes so our teams could simply focus on the actual algorithm.??

This approach enabled us to rapidly and effectively pursue AI-driven revenue opportunities with a focus on monetizing data, integrating AI capabilities into network operations and targeting ARPU growth. ?

At every turn, our teams can rapidly evaluate revenue growth opportunities tied to available data with fast deployment to understand whether desired outcomes can be achieved. By combining DPI, CDR and BSS data, customers can be targeted based on insights into preferences and behaviors. Then, personalized services and packages can be developed to enhance customer experiences and increase spend.???

Global data lessons emerge?

Our journey has been powered by universally adaptable strategies pioneered and proven by technology leaders outside our industry.

There are no wrong questions, just an unending focus on how to get to the desired destination from where we are today. With the proper data infrastructure and policies in place, telcos can war game scenarios based on specific goals to better understand the required steps to achieve them. By simulating various scenarios, new challenges and opportunities are revealed. ?

Then, it is up to strong leadership and dedicated centralized teams to drive initiatives forward. In our experience, cross-departmental collaboration is crucial to achieve data democracy. Operators can create a more integrated and efficient data ecosystem by encouraging teams to share data and adopt AI across applications.??

Unified adoption will hinge on a cultural shift that ensures AI initiatives remain a priority as every team member has a stake in fostering a culture of innovation and collaboration.??

We are on the road to autonomous networks, constantly opening doors to new opportunities along the way. Preparing the path that takes us there is today’s priority.???

Mention Rakuten Symphony AI team members Gaurav Jain and Piyush Khurana in the comments to start a conversation!