Human-Centered AI in Telecommunication

The power of Social Network Analysis in churn prediction and beyond

Earlier this year, I earned my PhD in human-centered AI transformation in banking. Since then, I’ve been invited to give keynotes and run in-company training programs, helping business leaders from many industries understand the impact of this AI revolution on their companies—and how to lead through it effectively.

In the coming weeks, I’ll be sharing a series of blogs on my insights from various industries from these activities. These are the same learnings we apply in our AdviceRobo AI-risk platform, but I believe they’ll also be valuable for anyone navigating the AI landscape. Stay tuned!

The first industry in today’s blog is the telecommunications industry. The telecommunications industry has rapidly evolved into one of the most significant sectors in developed countries. As technology advances and more operators enter the market, competition has become fiercer than ever. To stay ahead, telecom companies are adopting various strategies to boost revenue. The three main approaches are: (1) acquiring new customers, (2) upselling existing customers, and (3) increasing customer retention.

Recent studies have shown that, when considering return on investment (RoI), the third strategy—customer retention—is by far the most profitable. In fact, retaining an existing customer is not only more cost-effective than acquiring a new one, but it’s also easier than trying to upsell services. In a competitive market, keeping your current customers happy and loyal is the smartest move for long-term success.

One of the most critical strategies for telecom companies therefore is to reduce customer churn—the tendency of customers to switch providers. Churn is a major challenge in highly competitive industries like telecom. However, predicting which customers are likely to leave early can open a significant revenue opportunity.

Recent research has shown that machine learning is incredibly effective at predicting churn. By learning from historical customer data, companies can anticipate potential losses and act early to retain customers.

In one study, researchers used a dataset containing nine months' worth of customer information, which amounted to a massive 70 terabytes of data. This data came in various formats—structured, semi-structured, and unstructured—and flowed in quickly, requiring a powerful big data platform like Hadoop to process it. The data was aggregated to create features of each customer, including aspects like their social network connections and customer behavior.

To predict churn, the team employed several tree-based machine learning algorithms, such as Decision Trees, Random Forests, Gradient Boost Machines, and XGBoost. These algorithms were tested on telecom companies, with the results showing significant improvements in accuracy when Social Network Analysis (SNA) features, like degree centrality and customer network connectivity, were included.

Social Network Analysis (SNA) is a powerful tool for predicting customer churn in telecom. By analyzing the relationships and interactions between customers within a network, SNA can reveal patterns that indicate which customers are likely to leave. In fact, the use of SNA has shown to significantly improve predictive accuracy—boosting model performance from 84% to 93.3% on the AUC standard. This makes SNA an invaluable tool for identifying at-risk customers early and taking proactive steps to retain them.

Beyond churn prediction, SNA also has two other compelling use cases in the telecom sector. First, it’s incredibly effective in targeting new customers by identifying potential leads through network connections. By understanding the social ties between users, telecom companies can find customers who are more likely to respond to marketing offers or promotions. Second, SNA can optimize network resource management by analyzing usage patterns and customer behaviors, helping telecom companies better predict peak demand times and improve network reliability. These insights can drive smarter decisions, enhance customer experience, and ultimately boost profitability.

The takeaway? By using advanced AI and machine learning to analyze customer behavior and network connections, telecom companies can drastically improve their ability to predict and prevent churn, ultimately boosting their bottom line