PartnerPulse - AI powered Partner churn prediction

Stage 1a: Batch Churn Prediction PartnerPulse begins by analyzing the full partner portfolio (~5,000 partners), engineering 60+ behavioral, financial, and sentiment-based features from the last 6 months. These include commission trends, referral activity, login behavior, support tickets, sentiment signals, and competitor exposure. A bulk ML model (HistGradientBoostingRegressor) predicts churn risk, assigns LOW/MEDIUM/HIGH risk classes, and computes SHAP explanations for the top 50 highest-risk partners—making each prediction transparent and actionable. Stage 1b: Competitive Intelligence (Parallel) In parallel, the system assesses external market pressure by measuring multi-homing rates, share of voice, and sentiment gaps. It also scrapes key competitors (PocketOptions and OlympTrade) to capture both product positioning and affiliate program structures (commissions, CPA rates, tiers). This produces a structured, portfolio-wide competitive landscape. Stage 2: Churn Diagnosis Predictions and competitive insights are combined to diagnose why each partner is at risk. SHAP drivers are mapped to five root causes: Revenue Decline, Engagement Drop, Competitor Pressure, Support Dissatisfaction, and Platform Mismatch. Each partner receives a primary and secondary root cause with weighted attribution. Stage 3: Cohort Grouping & Action Partners are grouped into cohorts by root cause, partner type, and platform. Each cohort is assigned tailored retention strategies - ranging from commission restructuring and re-engagement campaigns to competitive counter-offers or support escalation which is refined by cohort size and risk severity. Stage 4: Internal Communication Insights are operationalized through automated executive briefings, real-time Slack alerts via OpenClaw, and a consolidated CEO-level summary-ensuring churn risk translates into timely, targeted action.