AI Feature Churn Analytics ft. SAP

Led a team in SAP's case competition (1st place out of 40 teams). I directed the analytical approach, shaped the segmentation strategy, and owned the final recommendation we presented to judges.

SAP had invested $1.1B in AI capabilities for its business software, but the investment was creating a new churn risk. Some users couldn't figure out the AI features and disengaged entirely. Others were using them heavily but found the personalization shallow and unsatisfying. Both groups were churning, but for completely different reasons. A single retention strategy couldn't address both.

We started with bivariate and correlation analysis and found something that reframed the whole problem: age was heavily negatively correlated (-0.79) with AI interaction levels. Low interaction and low satisfaction were both driving churn, but in entirely different populations.

I pushed the team to run cluster analysis rather than treating churn as one monolithic metric. The density distributions across key variables — AI interaction level, satisfaction, personalization effectiveness, response time — revealed clearly separable behavioral groups.

That surfaced three distinct customer segments. Cluster 3: High-Satisfaction Interactors — heavy users who were satisfied and not at risk. Cluster 1: Unsatisfied Customer-Service Seekers — younger, high-engagement users who interacted constantly but found the personalization shallow and unhelpful. Cluster 2: Aging Churners — an older demographic with the highest churn rates and minimal AI engagement. They weren't dissatisfied with the AI. They weren't using it at all.

The segmentation drove the entire strategy. For Aging Churners, we proposed a change management support service with consultants working directly with organizations to drive adoption. For Unsatisfied Seekers, we designed a machine learning personalization engine that would adapt UI, workflows, and process optimization suggestions based on individual user behavior.

We also proposed embedding AI suggestions directly into the SAP user flow — surfacing intelligent recommendations at bottleneck points in real-time process analysis, rather than requiring users to seek them out. We called the combined framework ARC, built around the principle that you can't fix churn you haven't properly segmented.

We also recommended SAP evaluate its AI solutions through an ethical lens covering accountability, privacy, bias, and accessibility, since the adoption barriers we found had real equity implications.

The ARC strategy projected a 5–10% improvement in AI personalization effectiveness and service satisfaction, with targeted churn reduction in the highest-risk segment. Won 1st place out of 40 competing teams at SAP x WiDS.

The segmentation-first approach became the foundation of the entire recommendation. Without splitting the problem into two distinct user groups, every solution we could have proposed would have partially addressed one group while ignoring the other.