Calculate Customer Health Scores

Designs, implements, and optimizes customer health scoring systems that predict customer behavior, flag at-risk accounts, and drive proactive customer success. Scores combine four dimensions - product adoption (feature usage, login frequency, engagement depth), relationship health (support ticket volume and sentiment, NPS, stakeholder engagement), business outcomes (goal achievement, ROI realization, expansion indicators), and behavioral patterns (usage trends, seasonal variation, cohort comparisons) - as weighted composite scores normalized to a 0-100 scale, weighting critical metrics higher, applying decay functions to time-sensitive data, and ranking customers by percentile within their cohort rather than on raw values alone. It implements both a basic weighted model (usage, engagement, support, and outcome sub-scores combined by fixed weights) and an advanced time-decay model that applies exponential decay to historical metrics and computes a trend-momentum factor comparing a customer's most recent week against the prior week. Adoption metrics are computed via percentile-ranked SQL over a 30-day event window (feature breadth, engagement frequency, engagement depth, core-feature adoption), and support health is scored from ticket volume, resolution time, CSAT, and escalation rate. Scores are banded into risk categories - Healthy, At Risk, High Risk, Critical - each mapped to a recommended action, with priority further adjusted for high-value accounts.

Use it to build or refine a customer health scoring model - defining metric weights, choosing between a static weighted model and a time-decay/momentum model, wiring up percentile-based adoption and support scoring, and setting the risk bands and recommended actions a CSM team acts on.

Input is raw customer usage, engagement, support, and outcome data. Output is a normalized 0-100 health score, a risk category and priority (adjusted for account value), a recommended action per category, and validation tooling - validate_health_score_accuracy - that measures the score's predictive accuracy against actual churn/retention outcomes using accuracy, precision, and recall at different score thresholds.

Implemented in Python with numpy for time-decay and momentum calculations and scikit-learn (accuracy_score, precision_recall_curve) for validating score accuracy against real outcomes, plus SQL PERCENT_RANK for percentile-based adoption scoring directly against a product-events table.

For customer success and CS-ops teams building or maintaining a health-scoring system. It also covers implementation practices (data-freshness checks, statistical outlier detection, validating calculations against known outcomes, gradual rollout of scoring changes), reporting (executive dashboards, cohort-based comparisons, alerts when a score moves more than 20 points), and where to extend it further: automated weight optimization against churn correlation, anomaly detection, time-series predictive scoring, NLP-based support-ticket sentiment analysis, and segmentation by industry, lifecycle stage, account tier, or geography.

def calculate_health_score(customer_data):
    # Normalize metrics to 0-100 scale
    usage_score = normalize_usage_metrics(customer_data['usage'])
    engagement_score = normalize_engagement_metrics(customer_data['engagement'])
    support_score = normalize_support_metrics(customer_data['support'])
    outcome_score = normalize_outcome_metrics(customer_data['outcomes'])
    
    # Apply weights (should sum to 1.0)
    weights = {
        'usage': 0.35,
        'engagement': 0.25,
        'support': 0.20,
        'outcomes': 0.20
    }
    
    health_score = (
        usage_score * weights['usage'] +
        engagement_score * weights['engagement'] +
        support_score * weights['support'] +
        outcome_score * weights['outcomes']
    )
    
    return min(100, max(0, health_score))