Telco Enhances Customer Retention Management with Advanced Predictive Model and Uncovers Insights into Churn Triggers

Spread the love
Case Study
Telco Enhances Customer Retention Management with Advanced Predictive Model and Uncovers Insights into Churn Triggers

Industry: Telecoms

Challenge: Predictive Modelling, Churn Management and customer retention

Solution: Machine learning and predictive analytics

Technology: Python, Oracle

Customer Satisfaction Survey Concept, Businessman Using Computer Laptop Select Smiley Face Icon With Yellow Five Stars To Evaluate Product And Service.

Situation and Challenge

Our telecommunications client, a subsidiary of a global telecom operator, sought to improve the accuracy of their existing customer churn models. These models had become suboptimal, leading to an overstatement of the churn rate and a resulting exaggeration of acquisition campaign performance.

As the industry leader in a highly competitive and commoditised market, the client recognised the need for a better predictive model to accurately identify real churners or segments of churners. This would enable more effective retention strategies for different customer segments with different triggers and motivations for churning.

Particularly challenging for the operator was to fully understand the impact of high rotational churn – a phenomenon where customers churn and immediately rejoin again. This is common in prepaid mobile phone services, with existing customers taking up new subscriptions to avail of special offers only available to new customers.

High rotational churn is leading to a loss of revenue and increased costs associated with acquiring new customers. To address this complex customer behaviour, we set out to build a new model to help identify and exclude highly probable rotational churn events.

Insight and Action​

Our analysis of detailed call records revealed a high incidence of rotational churn, also known as spinning. Market research evidence further corroborated our findings, showing that 20-30% of supposedly churned customers had stayed with the company but with a different contract or subscription after registering a new SIM card. Subsequently, it was discovered that legacy models had been trained on events that were influenced by both real and rotational churn.

To improve the accuracy of churn predictions, we developed a fingerprint-based methodology to uniquely identify rotational churners by pairing usage patterns and other interaction profiles with similarities between the churned and newly activated SIMs.

Our usage fingerprints were created by combining various data variables and derived variables sourced from demographic profiles, Call Detail Records (CDR), SIM and subscriber IDs, etc. We formulated various scenarios according to business objectives, considering factors such as:

  • The relevant historical period needed to construct robust fingerprints
  • The overlapping timeframe when two SIMs are used together (such as when one SIM is being churned and the other activated)
  • The weight of the fingerprint components based on frequency, intensity, and similarity scores to pair interactions between subscribers, as well as market definitions and customer segments
Results and Impact

Using a similarity-based weighted nearest-neighbour (k-NN) learning algorithm, our predictive model had remarkable results – a 95% accuracy rate in identifying rotational churners.

By further retraining our models based on real churn events, the performance and accuracy of churn prediction increased. This enabled the business to target retention efforts on at-risk, high-value customers, resulting in an impressive 8% improvement in retention rates and a 20% reduction in marketing and retention campaign spend.

Furthermore, the operator now had a deeper understanding of the triggers and reasons behind churn. They can now adapt promotional efforts accordingly and ensure our ROI remains high.

Let's Help You Transform for Growth and Accelerate Impact

    Contact Img
    Case Studies

    Related Content.

    Data Security System Shield Protection Verification

***These Are Our Own 3D Generic Designs. They Do Not Infringe On Any Copyrighted Designs.***

    Transport Company Improves Cybersecurity Management with Automated Threat Intelligence Reporting

    Female Special Agent Works On A Laptop In The Background Special Agent In Charge Talks To A Military Man In The Monitoring Room. In The Background Busy System Control Center With Monitors Showing Data Flow.

    Improving Fraud Detection and Prevention with Intelligent Risk-Based Fraud Detection System

    Movie Theater Ticketing Counter

    Predictive Modelling Helps a Ticketing Company Create Targeted and Personalised Offers to Increase Average Transaction Value