SIP Churn Prevention ML Model

Protecting INR 1000 Cr Through Predictive Analytics

Nippon Life India Asset ManagementAVP (Data Science and Intelligence)Asset ManagementAug 2021 - Apr 2023
Model Precision:0.87Model Recall:0.62Retention rate:60% -> 75%

Problem Statement

Systematic Investment Plan (SIP) portfolios at Nippon Life India Asset Management exhibited elevated termination rates, eroding long-term assets under management and diminishing cross-sell potential. Retention teams operated reactively, engaging customers only after cancellation requests were logged. Limited behavioural visibility and fragmented data flows prevented early identification of at-risk customers, leading to revenue leakage and reduced client lifetime value.

Solution

The programme delivered an end-to-end churn prediction and intervention capability, combining supervised machine learning, behavioural segmentation, and coordinated outreach playbooks.

Solution Architecture

  • Data foundation: Consolidated transaction history, SIP contribution cadence, interaction logs, market context, and demographic attributes into a governed feature store.
  • Model design: Gradient boosted ensemble trained on historical churn outcomes with feature engineering across temporal, behavioural, and sentiment dimensions. Monthly refresh cycles with automated drift detection maintained accuracy.
  • Scoring and orchestration: Batch and near-real-time scoring exposed via REST APIs into CRM and marketing automation platforms. Risk tiers (high/medium/low) triggered specific outreach cadences.
  • Engagement workflow: Integrated playbooks for advisor call-downs, personalised email/SMS nudges, and targeted upsell/cross-sell offers aligned to customer risk and portfolio goals.

Governance Controls

  • Explainability dashboards highlighted top predictive factors for compliance and sales leadership.
  • Performance dashboards tracked precision/recall, intervention acceptance, and incremental revenue impact.
  • Feedback loops captured outcome disposition (retained, declined, upgraded) to inform retraining.

Outcomes

Production deployment of the churn model sustained 0.87 precision and 0.62 recall on rolling holdouts, giving retention teams confidence that interventions were directed at the right investors. Coordinated advisor call-downs and automated nudges lifted SIP customer retention from 60% to 75% over the first 12 months, while shrinking intervention cycle times by more than half so outreach landed before cancellation requests surfaced. The programme repositioned retention from reactive case handling to proactive portfolio management, deepening relationships and opening space for targeted upsell plays.

Machine LearningChurn PredictionAsset ManagementCustomer RetentionFinancial Services