clusteraudiencekit 5.9.0

Enterprise audience intelligence platform. RFM + 6 clustering + streaming + CLV + lifecycle + drift detection + churn + B2B + lookalikes + plugin framework + RBAC + privacy + XGBoost + neural networks + segment intelligence + pattern discovery + temporal analytics + revenue intelligence + B2B governance + price intelligence (105 features). Real-time <1s, 1M+ customers, 31 engines, 384 tests.

ClusterAudienceKit v5.9.0

Enterprise audience intelligence platform — Complete ML stack for customer segmentation at scale. RFM + 6 clustering + streaming + CLV + churn + B2B + lookalikes + XGBoost + neural networks + segment intelligence + pattern discovery + temporal analytics + revenue intelligence + B2B governance + price intelligence (105 features) + privacy + 31 modules.

ClusterAudienceKit is the production-grade segmentation engine for modern martech. Replace your scikit-learn + pandas + lifetimes + Braze/Klaviyo combination with a single, unified platform backed by a Rust engine that handles 1M+ customers in under 500ms with integrated ML models for prediction and pattern discovery.

Install

Pick one:

pip install clusteraudiencekit

OR

uv add clusteraudiencekit

OR

curl -sSfL https://raw.githubusercontent.com/Mullassery/ClusterAudienceKit/main/install.sh | sh

Pre-built wheels for all platforms: INSTALL.md

Get started in 10 lines

from clusteraudiencekit import AudienceSegmenter
import pandas as pd

# Required columns: customer_id, transaction_date, amount
transactions = pd.read_csv('transactions.csv')

segmenter = AudienceSegmenter(method='rfm_kmeans', n_clusters=4)
segmenter.fit(transactions)

segments = segmenter.predict(transactions)
profiles = segmenter.segment_profiles()
print(profiles)
#   segment | size  | avg_recency | avg_frequency | avg_monetary
#   0       | 250k  | 15.3 days   | 8.2 purchases | $450   <- high-value loyalists
#   1       | 180k  | 45.2 days   | 3.1 purchases | $120   <- regular buyers
#   2       | 320k  | 2.1 days    | 2.0 purchases | $80    <- new / recent
#   3       | 250k  | 60.5 days   | 1.0 purchases | $30    <- at-risk / dormant

print(f"Silhouette score: {segmenter.silhouette_score():.3f}")

Why not just use scikit-learn?

You can — until your audience grows. sklearn.metrics.silhouette_score is O(n²): at 100k customers it takes over 2.7 hours. At 1M customers it won't finish. ClusterAudienceKit handles both in under half a second.

Measured timings on Apple M1 (sklearn 1.6.1, pandas 3.0.3):

Customer base	sklearn + pandas	ClusterAudienceKit
1,000	38ms	<9ms
10,000	606ms	<37ms
100,000	>2.7 hours	<130ms
1,000,000	Did not complete	<470ms

Beyond performance, you get the complete production stack:

Capability	sklearn	pandas	lifetimes	Braze	Klaviyo	ClusterAudienceKit v5.0
RFM calculation	—	manual	—	✓	✓	✓
6 clustering algorithms (K-Means, DBSCAN, Hierarchical, GMM, K-Prototypes)	partial	—	—	—	—	✓
Auto K-estimation (Elbow, Gap Statistic, Silhouette)	—	—	—	—	—	✓
Customer lifetime value (CLV)	—	—	✓	—	—	✓
Churn prediction (Logistic + Ensemble)	partial	—	partial	✓	✓	✓
XGBoost gradient boosting	manual	—	—	—	—	✓
Neural networks (MLP + Autoencoder + RNN)	manual	—	—	—	—	✓
AutoML hyperparameter tuning	manual	—	—	—	—	✓
Streaming/incremental updates	—	—	—	—	—	✓
Segment drift detection (K-S, Hellinger, Chi-square)	—	—	—	—	—	✓
Cohort analytics (retention, decay, comparison)	—	—	—	partial	—	✓
Lifecycle tracking (7-stage journeys)	—	—	—	—	—	✓
B2B segmentation + account health	—	—	—	—	—	✓
Lookalike audiences (4 similarity metrics)	—	—	—	—	—	✓
Plugin framework (custom algorithms)	—	—	—	—	—	✓
RBAC + audit logging	—	—	—	—	—	✓
Privacy: Differential Privacy + K-anonymity	—	—	—	—	—	✓
7+ platform integrations (Braze, Klaviyo, HubSpot, Segment, etc.)	—	—	—	1	1	✓
Production dashboard + KPIs	—	—	—	—	—	✓
Quality metrics + profiling	✓ (slow)	—	—	—	—	✓ (fast)
Multi-core by default	partial	—	—	—	—	✓

Full comparison with code examples: docs/comparison.md · Full benchmark methodology: BENCHMARKS.md

Segmentation methods

RFM + KMeans

Scores each customer on Recency, Frequency, and Monetary value, then groups them with KMeans. The standard approach for most Martech teams.

segmenter = AudienceSegmenter(method='rfm_kmeans', n_clusters=4)
segmenter.fit(df)

RFM + K-Prototypes

Extends RFM with categorical attributes — acquisition channel, product category, region — so your segments reflect more than just spend behaviour.

segmenter = AudienceSegmenter(method='rfm_kprototypes', n_clusters=5)
segmenter.fit(df, categorical_columns=['channel', 'region', 'product_category'])

Streaming updates

Update segments incrementally as daily events arrive, without reprocessing your full customer history. Detect and react to campaign-driven drift:

segmenter.fit(historical_data)

for daily_events in event_stream:
    segmenter.update(daily_events)

    stability = segmenter.segment_stability(previous_segments)
    if stability < 0.85:
        segmenter.fit(all_data, refit=True)

    previous_segments = segmenter.predict(customers)

PySpark integration

Use ClusterAudienceKit with Apache Spark DataFrames for large-scale customer segmentation on distributed clusters.

from pyspark.sql import SparkSession
import polars as pl
from clusteraudiencekit import AudienceSegmenter

spark = SparkSession.builder.appName("audience-segmentation").getOrCreate()

# Load customer transaction data from Spark
spark_df = spark.read.parquet("s3://bucket/transactions/")

# Convert to Polars for segmentation (small-scale, in-memory)
polars_df = spark_df.select("customer_id", "purchase_amount", "purchase_date") \
    .toPandas()
polars_df = pl.from_pandas(polars_df)

# Fit segmentation model
segmenter = AudienceSegmenter(method='rfm_kmeans', n_clusters=5)
segmenter.fit(polars_df)

# Get segment assignments
segments = segmenter.predict(polars_df)

# Write segments back to Spark
segments_df = spark.createDataFrame(
    segments.to_pandas(),
    schema=["customer_id", "segment"]
)
segments_df.write.mode("overwrite").parquet("s3://bucket/segments/")

print(f"Segmented {segments_df.count()} customers into {segmenter.n_clusters} segments")

Note: For very large datasets, consider:

• Sampling/filtering in Spark before converting to Polars
• Running segmentation on aggregated RFM scores per customer (reduces memory footprint)
• Caching the Polars DataFrame if running multiple predictions

Configuration

AudienceSegmenter(
    method='rfm_kmeans',        # 'rfm_kmeans' | 'rfm_kprototypes' | 'kmeans_only'
    n_clusters=4,               # number of segments
    recency_window_days=90,     # lookback window in days
    decay_function='linear',    # 'linear' | 'exponential' | 'inverse'
    decay_half_life_days=30,    # half-life for exponential decay
    frequency_threshold=1,      # minimum transactions to include a customer
    monetary_threshold=0.0,     # minimum spend to include a customer
    random_state=42,
    n_jobs=-1,                  # -1 = all cores
)

Documentation

INSTALL.md	pip, uv, and pre-built wheel installation
docs/api-reference.md	All 13 methods
docs/getting-started-simple.md	Guide for non-technical marketing teams
docs/comparison.md	Side-by-side vs sklearn / pandas / lifetimes
BENCHMARKS.md	Benchmark methodology and raw results
docs/troubleshooting.md	Common errors
docs/architecture.md	Design decisions
examples/	Runnable scripts

What's Included in v5.5.0

✅ Phase 1: Core Segmentation

• ✅ Full RFM engine (linear/exponential/inverse decay)
• ✅ 6 clustering algorithms (K-Means, K-Prototypes, DBSCAN, Hierarchical, GMM, custom)
• ✅ Auto K-estimation (Elbow, Gap Statistic, Silhouette)
• ✅ 13 automatic business segments (Champions, Loyal, At Risk, etc.)
• ✅ Behavioral rule engine (SQL-like conditions)
• ✅ Segment profiling + 15+ quality metrics

✅ Phase 2: Production Features

• ✅ Streaming: <500ms real-time updates via event streams with drift detection
• ✅ CLV: Historical, predictive, and probabilistic models (5-year forecasting)
• ✅ Lifecycle: 7-stage journey modeling (Prospect → Churned)
• ✅ Cohort Analytics: Retention curves, decay rates, comparison
• ✅ Drift Detection: K-S, Hellinger, Chi-square tests with severity levels
• ✅ Activation: 7 platform adapters (Braze, Klaviyo, Salesforce, HubSpot, Segment, etc.)
• ✅ Dashboard: Real-time KPIs, trends, segment health, streaming metrics

✅ Phase 3: Advanced Segmentation

• ✅ Churn Prediction: Logistic regression + ensemble random forest with risk scoring
• ✅ B2B Segmentation: Firmographic profiling, account health, expansion opportunities
• ✅ Lookalike Audiences: 4 similarity metrics (Cosine, Euclidean, Manhattan, Jaccard)
• ✅ TAM Calculation: Total addressable market analysis per segment

✅ Phase 4: Enterprise Governance

• ✅ Plugin Framework: Trait-based extensibility for custom algorithms
• ✅ RBAC: 5 roles × 8 actions × 6 resources with granular control
• ✅ Audit Logging: Complete traceability of all actions
• ✅ Privacy: Differential privacy (Laplace, Gaussian), K-anonymity, row suppression

✅ Phase 5.2: Predictive ML

• ✅ XGBoost: Gradient boosting for churn/CLV prediction (hyperparameter tuning)
• ✅ Neural Networks: MLP, Autoencoder, RNN for pattern discovery
  • Dense layers with ReLU/Sigmoid/Tanh activations
  • Backpropagation training via mini-batch SGD
  • Unsupervised feature learning
  • Anomaly detection via reconstruction error

✅ Phase 5.3: Segment Intelligence (10 features)

• ✅ Explainability: Feature importance → segment membership causality (XGBoost wiring)
• ✅ Confidence Score: Membership certainty (distance-based 0-1 scoring)
• ✅ Entropy Analysis: Segment diversity (Shannon entropy + Gini coefficient)
• ✅ Stability Score: Retention tracking + churn resistance
• ✅ Decay Detection: Attrition forecasting with half-life calculation
• ✅ Predictability: Assignment stability with trend detection
• ✅ Differentiation: Segment uniqueness vs. nearest competitor
• ✅ Segment Aging: Member tenure analysis + lifecycle staging
• ✅ Segment Health: Composite 0-100 scoring with alerts

✅ Phase 5.4: Pattern Discovery (21 features)

• ✅ Emerging Audiences: Accelerating segment detection with growth forecasting
• ✅ Hidden Opportunities: Low-engagement high-LTV segment identification
• ✅ Trend-Based Discovery: Time-series trend analysis via linear regression
• ✅ Intent Clusters: Behavioral pattern classification (churn, growth, engagement)
• ✅ Growth Forecasting: Multi-period projection with confidence intervals
• ✅ AI Personas: Auto-generated personas (High-Value, At-Risk, Growth-Oriented, Engaged)
• ✅ Product Affinity: Cross-product relationship discovery with lift calculation
• ✅ Causal Drivers (10 types): Feature → outcome causality + effect size scoring
• ✅ Micro-Communities: Small, tightly-bonded groups (cohesion > 0.7)
• ✅ Customer Tribes: Large, influence-driven groups with core values
• ✅ Lifecycle Discovery: Auto-discovered customer journey stages with transition rates

✅ Phase 5.5: Temporal Analytics (12 features)

• ✅ Temporal Snapshot: Capture segment state at any point in time
• ✅ Historical Reconstruction: Rebuild past segments from event logs with composition changes
• ✅ Segment Size Forecasting: Predict future segment sizes with confidence intervals
• ✅ Composition Forecasting: Predict high-value/churn-risk/new-member ratio changes
• ✅ Membership Forecasting: Predict individual member segment movement probabilities
• ✅ What-If Scenarios: Simulate parameter and rule changes with impact analysis
• ✅ Scenario Comparison: Compare multiple scenarios for revenue and churn impact
• ✅ Sensitivity Analysis: Tornado analysis and parameter elasticity measurement
• ✅ Expansion Planning: Growth planning with resource requirements and ROI
• ✅ Churn Forecasting: Project churn rates over time with intervention opportunities
• ✅ Lifecycle Forecasting: Predict customer stage transitions with Markov chains
• ✅ Trend Momentum: Trend analysis with continuation probability and reversal risk

✅ Phase 5.6: Revenue Intelligence (15 features)

• ✅ Segment Revenue: Revenue per segment with concentration and top-customer metrics
• ✅ Segment ROI: Return on investment calculation with payback period and profitability index
• ✅ Revenue Attribution: Multi-channel and multi-product revenue attribution modeling
• ✅ Revenue Alerts: Real-time anomaly detection with Z-score and variance analysis
• ✅ Revenue Forecasting: Linear regression-based revenue prediction with confidence intervals
• ✅ Customer Acquisition Cost: CAC tracking with payback period and efficiency scoring
• ✅ Margin Analysis: Gross, operating, and net margin breakdown per segment
• ✅ Upsell Opportunities: Product penetration analysis with addressable market sizing
• ✅ Concentration Risk: Herfindahl index, Gini coefficient, and diversification scoring
• ✅ Revenue Efficiency: Revenue per marketing spend, per employee, per customer
• ✅ Revenue Trends: Growth rate analysis with trend direction and momentum detection
• ✅ Cohort Revenue: Track revenue by customer cohort with retention correlation
• ✅ Product Mix Analysis: Product revenue diversification with concentration metrics
• ✅ Growth Rates: CAGR calculation with growth classification (high/moderate/stable/decline)
• ✅ Revenue Health Score: Composite 0-100 score (35% profitability + 25% growth + 25% efficiency + 15% concentration)

✅ Phase 5.7: B2B & Governance (15 features)

• ✅ Account Hierarchy: Track company structures with parent-subsidiary relationships
• ✅ Buying Committees: Identify decision makers with influence and engagement scoring
• ✅ Intent Signals: Aggregate buying intent indicators across multiple sources
• ✅ Account Health: B2B health scoring with engagement, expansion, and churn risk
• ✅ Data Lineage: Track data provenance and transformations across systems
• ✅ Ownership Assignment: Assign primary and secondary owners to segments
• ✅ Advanced What-If: Simulation with constraints and feasibility scoring
• ✅ Segment Genealogy: Track segment evolution with versions and ancestors
• ✅ Feature Provenance: Track feature sources, transformations, and reliability
• ✅ Audit Trails: Complete decision audit with actor, action, and impact tracking
• ✅ Policy Enforcement: Define and enforce segmentation policies with compliance scoring
• ✅ Access Control: Granular ACLs for segment and feature access
• ✅ Segment Contracts: Define SLAs with size, churn, and quality commitments
• ✅ Change Tracking: Track all segment definition changes with member impact
• ✅ Impact Analysis: Analyze change impact on customers, revenue, and contracts

✅ Phase 5.8: Price Intelligence (15 features)

• ✅ Price Elasticity: Measure price sensitivity with demand curve modeling
• ✅ Tier Migration: Predict customer movement across pricing tiers
• ✅ Category Affinity: Identify cross-category purchasing patterns
• ✅ Price Sensitivity: Measure discount response and willingness to pay
• ✅ Discount Optimization: Find optimal discount strategies for conversion lift
• ✅ Revenue Maximization: Identify highest revenue pricing strategy
• ✅ Competitive Pricing: Benchmark against competitors with price gap analysis
• ✅ Price Thresholds: Detect acceptable price ranges and breaking points
• ✅ Demand Forecasting: Forecast demand at different price points
• ✅ Margin Optimization: Maximize profit margins with cost-based pricing
• ✅ Bundle Recommendations: Recommend product bundles with discount strategy
• ✅ Price Ranges: Define minimum and maximum acceptable prices
• ✅ Churn by Price: Analyze churn rates at different pricing tiers
• ✅ Customer Value by Tier: Segment value analysis by pricing tier
• ✅ Price Change Impact: Forecast impact of price changes on volume and revenue

📋 Upcoming (v5.9+)

• Phase 5.2.3: AutoML framework (grid/Bayesian search, ensemble voting)
• Phase 5.9: Graph Intelligence, Real-Time Events, Experimental AI (500+ hrs)

Roadmap: v5.0 → v6.0

Phase 5.2.3 (20 hrs) — AutoML Framework

• Grid search & random search hyperparameter tuning
• Bayesian optimization for model selection
• K-fold cross-validation strategies
• Ensemble voting (XGBoost + Neural Networks)
• Automated feature selection

Phase 5.3 (190 hrs) — Segment Intelligence

• Explainability: Why do customers belong to segments?
• Confidence scoring: How sure are membership decisions?
• Stability metrics: Do segments stay stable over time?
• Decay detection: Which segments are losing relevance?
• Health dashboards: Real-time segment KPIs

Phase 5.4 (250 hrs) — Pattern Discovery + Revenue Intelligence

• Unsupervised audience mining
• AI persona generation
• Trend-based segment discovery
• Causal driver analysis (what causes churn/growth?)
• Revenue attribution & ROI per segment

Phase 5.5 (45 hrs) — Temporal Analytics ✅ COMPLETE

• Time machine: View segments as they existed at any past date
• Forecasting: Predict segment size, composition, and membership movement
• What-if modeling: Simulate parameter and rule changes
• Scenario comparison and planning for expansion
• Sensitivity analysis with tornado charts
• Trend momentum analysis with reversal detection

Phase 5.6 (40 hrs) — Revenue Intelligence ✅ COMPLETE

• Segment revenue tracking with concentration metrics
• ROI calculation with payback period and profitability index
• Multi-channel and multi-product attribution modeling
• Real-time revenue anomaly detection and alerts
• Revenue forecasting with confidence intervals
• CAC tracking and payback period analysis
• Gross/operating/net margin breakdown per segment
• Upsell opportunity identification with addressable market sizing
• Revenue concentration risk analysis (Herfindahl, Gini)
• Revenue efficiency scoring per customer and per marketing spend
• Trend analysis with growth rate classification
• Cohort revenue tracking with retention correlation
• Product mix diversification analysis
• CAGR and growth classification
• Composite health scoring (profitability, growth, efficiency, concentration)

Phase 5.7 (50 hrs) — B2B & Governance ✅ COMPLETE

• Account hierarchy tracking with parent-subsidiary relationships
• Buying committee identification with influence scoring
• Intent signal aggregation with budget and timeline classification
• B2B account health scoring (engagement, expansion, churn)
• Data lineage tracking across systems and transformations
• Segment and feature ownership assignment
• Advanced what-if modeling with constraints and feasibility
• Segment genealogy with version history and ancestors
• Feature provenance with source tracking and reliability scoring
• Comprehensive audit trails for all decisions
• Policy definition and enforcement with compliance scoring
• Granular access controls and ACLs
• Segment contracts with SLAs and compliance tracking
• Change tracking with member and revenue impact
• Impact analysis for changes on customers and revenue

Phase 5.8 (50 hrs) — Price Intelligence ✅ COMPLETE

• Price elasticity analysis with demand curve modeling
• Tier migration prediction with growth/churn assessment
• Category affinity and cross-sell opportunity scoring
• Discount optimization with conversion lift forecasting
• Revenue maximization and competitive benchmarking
• Price threshold detection and range definition
• Demand forecasting at multiple price points
• Margin optimization with cost-based pricing
• Bundle recommendations with discount strategy
• Churn analysis by pricing tier
• Customer value segmentation by tier
• Price change impact modeling

Phase 5.9+ (500+ hrs) — Advanced Engines

• Graph intelligence (relationships, households, networks)
• Real-time events (live alerts, anomaly detection, triggers)
• Experimental AI (self-healing segments, autonomous discovery)

Community

• GitHub Issues — Report bugs and request features
• GitHub Discussions — Questions and best practices
• Code of Conduct — Be respectful and constructive

Contributing

Pull requests welcome! See CONTRIBUTING.md for development setup and guidelines.

For security issues, see SECURITY.md.

Author

Georgi Mammen Mullassery — github.com/Mullassery

License

MIT

🔒 Security & Error Handling

ClusterAudienceKit v2.0 includes production-grade security:

• Input Validation: Pydantic models validate all requests
• Resource Limits: DoS protection (10M customers, 1000 clusters, 100 features)
• Memory-Safe Rust: No buffer overflows or data races
• Detailed Error Messages: Clear recovery steps for all failures
• Audit Logging: Track all activation exports and API calls
• Rate Limiting: Streaming buffer management and batch timeouts

v2.0 Features Deep Dive

Real-Time Streaming (<1s latency)

from clusteraudiencekit import StreamingSegmenter

segmenter = StreamingSegmenter(config={
    'batch_size': 100,
    'batch_timeout_ms': 5000,
    'decay_factor': 0.95
})

# Process events as they arrive
for event in event_stream:
    update = segmenter.process_event(event)
    if update.segment_changed:
        platform_manager.activate(update.customer_id, update.new_segment)

Drift Detection & Alerts

from clusteraudiencekit import DriftDetector

detector = DriftDetector()
drift = detector.detect_feature_drift(
    'recency',
    baseline_values,
    current_values,
    method='kolmogorov_smirnov'
)

if drift.severity >= DriftSeverity.High:
    alert_manager.notify(f"Critical drift in {drift.feature_name}")
    segmenter.fit(refit=True)  # Auto-refit on critical drift

Enterprise Platform Activation

from clusteraudiencekit import ActivationOrchestrator

orchestrator = ActivationOrchestrator(config={
    'batch_size': 1000,
    'max_retries': 3,
    'timeout_ms': 30000
})

# Register platforms
orchestrator.register_platform(braze_credential)
orchestrator.register_platform(klaviyo_credential)
orchestrator.register_platform(salesforce_credential)

# Activate to multiple platforms simultaneously
results = orchestrator.process_batch(messages)
success_rate = orchestrator.success_rate()  # Monitor performance

Cohort Analytics

from clusteraudiencekit import CohortAnalytics

# Track retention over time
cohort = CohortAnalytics.create_cohort(
    cohort_id='2026-Q3',
    period=CohortPeriod.Monthly,
    customers=customer_list
)

# Add retention snapshots
CohortAnalytics.add_retention_point(cohort, age_in_months=1, retained_count=950)
CohortAnalytics.add_retention_point(cohort, age_in_months=2, retained_count=900)

# Get insights
decay_rate = CohortAnalytics.retention_decay_rate(cohort)
print(f"Monthly decay: {decay_rate:.3f}")

Production Dashboard

from clusteraudiencekit import DashboardProvider

dashboard = DashboardProvider.generate_dashboard(
    summary=summary_metrics,
    segments=segment_cards,
    kpis=kpi_list,
    streaming=streaming_metrics,
    drift_alerts=drift_summary,
    time_range=TimeRange.Last7Days
)

# Export for frontend
data = DashboardProvider.export_summary(dashboard)

🆕 What's New in v2.0 (Production Ready)

Seven Production Systems in One Import

1. RFM + Clustering — Core segmentation with 4 algorithms
2. Streaming Segmentation — Real-time updates <1 second
3. Customer Lifetime Value — Historical + predictive + probabilistic
4. Lifecycle Tracking — 7-stage customer journey
5. Drift Detection — Statistical monitoring with alerts
6. Cohort Analytics — Retention curves and comparisons
7. Enterprise Activation — Push to 7 platforms instantly

Performance Metrics

Operation	100k Customers	1M Customers
RFM calculation	45ms	180ms
K-Means clustering	85ms	350ms
Silhouette score	120ms	450ms
Drift detection	65ms	250ms
Streaming update	5ms	8ms
Batch activation	500ms	2000ms

All benchmarks on Apple M1 Pro, single core