Churn Predictor - Gainsight
About Gainsight: Gainsight is a software company that helps businesses become more customer-centric through its Customer Success Management (CSM) platform. Gainsight's platform offers customer success, community, education, and adoption technology, including AI technology that aims to transform customer success.
Gainsight's platform can help businesses:
-
Drive growth: Use customer-led, product-led, and community-led strategies to scale efficiently and create alignment
-
Increase product adoption: Use advanced segmentation to drive personalized engagement
-
Prevent churn: Use customer health insights to drive renewals and reduce churn
-
Turn customer data into opportunities: Use data-freshness measured in seconds to avoid loss of productivity
Gainsight's mission is to be Human-First, which means cultivating innovative ideas, nourishing a deep connection to communities, introducing industry-defining products, and never saying no to childlike joy. Some say that Gainsight isn't for everyone, but for the right people, it can be a special place.
1. Objective: As part of my role at Gainsight, I led a project focused on developing a Churn Predictor aimed at enhancing our Customer Success Management platform. The primary goal was to enable our clients to proactively identify and retain at-risk customers by predicting churn likelihood with high accuracy.
2. Value Proposition: The Churn Predictor was designed with a clear value proposition: if our clients could foresee the possibility of their customers churning, they could take preemptive action using established playbooks to retain those customers. This predictive capability would directly contribute to reducing churn rates and increasing customer lifetime value.
3. Approach and Solution: To build the Churn Predictor, we leveraged a comprehensive set of data sources, including:
-
User Adoption Data: Tracking how actively customers were using the product.
-
User Engagement Data: Monitoring customer interaction levels with various features.
-
Competitor Data: Analyzing competitor activities, pricing strategies, and market positioning.
-
Geographical Data: Understanding regional influences on customer behavior.
We employed advanced statistical models to process this data and developed a supervised Machine Learning model to calculate the probability of churn for each customer. The model was designed to analyze patterns in the data and predict churn with a high degree of accuracy.
4. Challenges in Version 1 (V1): Upon launching the first version of the Churn Predictor, we encountered a significant issue. Despite the model's predictions, a few high-profile cases emerged where customers churned even though their predicted churn probability was very low. While these instances were rare, their impact was considerable, leading to concerns among our clients about the model's reliability.
​
5. Root Cause Analysis: To understand the root cause of these anomalies, we conducted extensive research. This included:
-
Market Research: Analyzing broader market trends and customer behavior patterns.
-
Stakeholder Interviews: Engaging with account managers and Customer Success Managers (CSMs) to gather insights.
-
Data Analysis: Revisiting the data inputs and model outputs to identify any discrepancies.
Through this process, we identified a common factor in the cases where the model failed: leadership changes within our clients' customer organizations. In these scenarios, new leaders often brought in different product preferences or strategic directions, which the initial model had not accounted for.
​
6. Solution Enhancement: To address this gap, we enhanced the model by incorporating linguistic machine learning techniques. We began collecting and analyzing real-time data from press releases, executive bios, and other relevant sources provided by data vendors. This allowed us to detect leadership changes within customer organizations and assess the potential impact on churn probability.
When the model identified a leadership change involving a key decision-maker, it adjusted the churn probability accordingly, increasing it to reflect the heightened risk. This enhancement significantly improved the model's accuracy and addressed the concerns raised by our clients.
​
7. Outcome and Impact: The refined Churn Predictor was well-received by the Gainsight client community. The added capability to factor in leadership changes and other external influences not only increased the accuracy of the predictions but also reinforced client confidence in our platform. This project demonstrated the importance of continuously iterating on AI-driven products to ensure alignment with user needs and market dynamics.
​
Conclusion:
The development and enhancement of the Churn Predictor at Gainsight highlight the critical role of adaptive, data-driven approaches in product management. By addressing real-world challenges through innovative solutions, we were able to deliver a tool that provided substantial value to our clients and set a new standard in Customer Success Management.
​
More about Churn Predictor here.