CUSTOMER RETENTION INTELLIGENCE
live browser ML
Predict churn. Explain it. Price the decision.
A premium customer-retention AI product that turns machine-learning output into something a founder, analyst, student, ML engineer, or business team can actually understand and use. Prediction, explanation, counterfactuals, decision economics, calibration, drift, fairness, MLOps, learning, and monetization—all in one product.
01 No Python server
02 Real browser inference
03 Explainable output
04 Mobile-first
LIVE CUSTOMER RISK
Decision Intelligence
example risk
74.8%
HIGH PRIORITY
Top signalMonth-to-month
ModelLogistic Regression
ConfidenceStrong
Recommended next step
Retention review + service-friction check
Explainable AI
Cost-aware threshold
Drift + fairness
THE PRODUCT, NOT JUST THE MODEL
A complete AI decision system
Most ML demos stop at a probability. This one continues into explanation, economics, reliability, governance, production engineering, and education.
01 / CORE
82%
churn probability
Live Churn Intelligence
Score a customer instantly, apply a business threshold, and translate model output into a clear decision.
inferencerisk banddecision policy
02 / XAI
Why did the AI say that?
Plain-English reasons first. Technical feature contributions when you want them.
Contract↑
Tenure↓
Support↑
03 / WHAT-IF
Counterfactual Optimizer
Automatically ranks the model changes that reduce churn risk the most.
2-year contract−21%
Add support−9%
Card payment−5%
DECISION ECONOMICS
Retention ROI Engine
Turn churn probability into expected benefit, intervention cost, net value, and expected ROI.
expected valueROI
COST-SENSITIVE ML
Threshold Cost Optimizer
Choose false-positive and false-negative costs and find a better operating threshold than blindly using 0.50.
policyfalse-positive cost
RELIABILITY
Confidence & Calibration
Inspect decision entropy and whether predicted probabilities behave like real probabilities.
entropyreliability
MLOPS
Data Drift Simulator
Shift production-like distributions and calculate PSI to understand model-health monitoring.
PSImonitoring
GOVERNANCE
Fairness Audit
Compare true-positive, false-positive, precision, and flag-rate behavior across demo groups.
TPRFPR
DATA ENGINEERING
Input Quality Gate
Validate schema, categories, ranges, missing values, and suspicious values before inference.
schemavalidation
BUSINESS INTELLIGENCE
Cohort Intelligence
Find high-risk customer segments by contract, internet service, tenure, and average monthly charges.
OPERATIONS
Batch Customer Scoring
Upload CSV, score customers locally in the browser, prioritize risk, and export the scored file.
ENGINEERING
Production & MLOps
ETL, feature pipelines, model registry, API serving, observability, drift, retraining, and rollback.
ONE PRODUCT. FOUR AUDIENCES.
Same AI, different explanation
The experience changes depending on who is using it.
Business owner
“Who is likely to leave, why, what should we do, and is the intervention financially worth it?”
Student
“Teach me the full project from data to model evaluation, mathematics, deployment, and interview explanation.”
Data scientist
“Show me metrics, calibration, cohorts, contributions, threshold trade-offs, and data behavior.”
ML engineer
“Show me validation, model packaging, observability, drift, versioning, MLOps, and system architecture.”
MONETIZATION-READY PRODUCT
Sell the project at multiple value levels
The strongest business model is not one cheap ZIP. Use the live demo as your free acquisition layer, then sell learning access, full source code, commercial rights, and customization separately.
LEARNING PACK$29
For learners
- Dataset
- Project guide
- Notebook
- Model explanation
- Tutorial roadmap
PORTFOLIO PRO$79
Full project source
- Everything in Learning
- Full premium website source
- Prediction engine
- Explainability labs
- Interview prep
- Deployment guide
COMMERCIAL / UNIVERSITYFrom $249
For organizations
- Commercial deployment discussion
- Brand customization
- Institution/company use
- Feature tailoring
- Support options
Pricing is included as a suggested selling structure and can be edited in index.html.
LIVE DEMO → PRODUCT → SERVICE
Use the demo to create higher-value customers
A visitor can test the AI for free, then buy the source package or contact you for customization, training, or business deployment.
AI EXPERIENCE
AI Copilot
Ask questions in normal language. Switch perspective to get a CEO explanation, beginner lesson, data-science view or ML-engineering view.
I’m your project copilot. Ask: “Why is this customer high risk?”, “What should the business do?”, “Explain ROC-AUC”, “How would this go to production?”, or “What should I learn from this project?”
Suggested questions
LIVE INFERENCE
Predict customer churn
Enter a customer profile. The trained model runs locally in the browser, then the decision layer explains risk, confidence and next action.
AI
Prediction ready when you are
Complete the profile or load the example customer.
EXPLAINABLE AI
Why this prediction?
See human-readable reasons first, then inspect technical feature contributions.
Plain-English explanation
Run a prediction first.
Feature contribution
Positive contribution raises churn log-odds; negative contribution lowers it.
Important: predictive contribution is not causal proof. A feature can help prediction without being the true cause of churn.
ADVANCED WHAT-IF AI
Counterfactual Optimizer
Instead of manually trying changes, the system tests multiple realistic alternatives and ranks which ones reduce predicted churn risk the most.
Current customer
Run a prediction first
Best risk-reduction options
Counterfactual recommendations will appear here.
Counterfactuals are not causal guarantees. They show how this model’s prediction changes when inputs change. Real interventions should be tested experimentally.
DECISION ECONOMICS
Retention ROI Engine
A churn score alone does not tell you whether intervention is worth it. Convert risk into expected business value.
Economics
Run a prediction, then calculate ROI.
Decision rule
Expected benefit = P(churn) × success rate × retained value
Expected net value = expected benefit − intervention cost
A high churn probability is not automatically a reason to offer a discount. Economics matters.
COST-SENSITIVE MACHINE LEARNING
Threshold Cost Optimizer
Choose business costs and let the system find the operating threshold with the lowest expected classification cost on the reference dataset.
Cost across thresholds
RELIABILITY
Confidence & Calibration Lab
Two underrated questions: “How uncertain is this decision?” and “Do predicted probabilities behave like probabilities?”
Latest prediction entropy
—
Normalized Bernoulli entropy is highest near 50% probability and lowest near 0% or 100%.
Decision confidence
—
This demo uses 1 − entropy as a decision-confidence indicator. It is not a full estimate of epistemic uncertainty.
Calibration
Check chart
A calibrated 70% prediction group should churn roughly 70% of the time over many examples.
Reliability curve
SEGMENT INTELLIGENCE
Cohort Intelligence
Model performance is individual. Business strategy often works at the segment level. Compare churn patterns across contract + internet cohorts.
Cohort table
Reference synthetic dataset
MLOPS MONITORING
Data Drift Simulator
Production data changes. Simulate feature drift and calculate Population Stability Index (PSI), a common monitoring signal.
Interpretation
< 0.10Low0.10–0.25Moderate> 0.25Significant
PSI thresholds are conventions, not universal laws. Always combine them with domain context and model-performance monitoring.
RESPONSIBLE AI
Fairness Audit
Compare group-level error behaviour on this synthetic demo. This is a teaching tool for governance, not a real-world fairness certification.
Group comparison
Important: fairness analysis depends on legal context, affected populations, measurement quality and the actual decision being made. No single metric defines fairness.
DATA ENGINEERING
Data Quality Validator
Before prediction, validate schema, allowed categories, numeric ranges and suspicious values. Production ML fails as often from bad data as from bad algorithms.
Validate current profile
Checks performed
✓ Required fields✓ Numeric type✓ Numeric range✓ Allowed categorical values✓ Missing values✓ Suspicious total-vs-tenure pattern✓ Schema completeness
OPERATIONAL ML
Batch Scoring
Upload a CSV. Customers are scored locally in the browser, ranked by risk and downloadable as a new CSV.
EXPLORATORY DATA ANALYSIS
Data Explorer
Understand the training distribution before trusting the model.
Rows—
Churn—
Avg tenure—
Avg monthly—
Churn by contract
Churn by payment method
ALGORITHM COMPARISON
Model Arena
Compare multiple candidate models on the same held-out data. “Best” depends on your metric and operating constraints.
Best ROC-AUC on this synthetic dataset—
Logistic Regression
Fast, interpretable, probability-based baseline.
Random Forest
Nonlinear ensemble of randomized decision trees.
Gradient Boosting
Sequential trees that correct earlier errors.
KNN
Instance-based learning using nearby examples.
MATHEMATICAL FOUNDATIONS
Interactive Math Lab
Understand the mathematics without turning the experience into a textbook.
Probability from a linear score
z = w₀ + w₁x₁ + … + wₙxₙ
σ(z) = 1 / (1 + e−z)
0.5000
Binary cross-entropy
L = −[y log(p) + (1−y) log(1−p)]
Confident wrong predictions receive a large penalty. Training adjusts model parameters to reduce average loss.
Precision, recall and F1
Precision = TP / (TP + FP)Recall = TP / (TP + FN)F1 = 2PR / (P + R)Accuracy = (TP + TN) / N
Binary entropy
H(p) = −p log₂(p) − (1−p) log₂(1−p)
Entropy is highest at p=0.5 and lowest near p=0 or p=1. It is useful as a simple decision-uncertainty signal.
Population Stability Index
PSI = Σ (actual% − expected%) × ln(actual% / expected%)
PSI compares the distribution of production data with a reference distribution.
CAREER ROADMAP
ML Engineer Learning Path
Use this one project to learn the complete machine-learning lifecycle.
ACADEMY
Tutorial Academy
Deep lessons with purpose, mistakes and interview framing.
SYSTEM DESIGN & MLOPS
Production Architecture
How a browser demo becomes a secure, monitored enterprise ML system.
CRM / Billingsource
→Validation + ETLquality
→Feature Pipelineconsistency
→Model Registryversioning
→Prediction APIinference
→Retention Apphuman action
Monitoring
Latency, errors, drift, calibration, prediction distribution and delayed labels.
Model Registry
Version model artifacts, metadata and approval state.
Champion / Challenger
Test new candidates before replacing production.
Rollback
Return safely to a prior version when issues appear.
Security
Authentication, authorization, input validation and PII minimization.
Business Experiment
Measure whether the intervention improves retention, not only whether the model predicts churn.
EXPORT
AI Decision Report
Generate a shareable report from the latest prediction, explanation, confidence and recommended business action.
Report settings
Run a prediction and generate a report.
CAREER READINESS
Interview Preparation
Learn to explain engineering decisions, not only code.
60-second explanation
“I built an end-to-end customer churn decision-intelligence system. I created a reproducible preprocessing pipeline, trained and compared multiple classifiers, evaluated them with multiple metrics, deployed browser-side inference, added explainability, counterfactual analysis, cost-aware thresholding, calibration, drift simulation, fairness checks, batch scoring and business ROI. I also documented how the system would evolve into a monitored production architecture.”