Kudakwashe Kadema
Data Scientist Machine Learning Engineer Software Engineer

Telecom Retention Case Study

I built a full-stack retention engine: automated quality controls, exploratory analytics, statistical testing, clustering, and an executive dashboard that turns insights into action.

Launch Interactive Dashboard Automated EDA Report Executive Insight PDF

Summary

I built this case study to demonstrate a boardroom-ready retention toolkit. Every deliverable flows from a signed-off data contract into profiling, data quality automation, statistical inference, segmentation, and a living dashboard. You can stress-test the approach directly from this hub.

Data Foundations

Auditable and Trusted

I documented lineage, validated schema with Pandera, and enforced quality gates before any modelling or storytelling took place.

Analytics Engine

Story-Driven EDA

The notebooks move from descriptive analytics to inference, quantifying support load and app adoption as real levers on churn risk.

Activation

Actionable Personas

I fused segmentation, propensity scores, and KPI storytelling into an interactive Streamlit experience for sales, care, and strategy teams.

Key Assets

Segmented Dataset

Customer-level file enriched with churn scores, cluster personas, and recommended retention actions.

Download segmented.csv

Cluster Summary

Aggregate view per persona with ARPU, engagement, support demand, and churn probability insights.

Download segment_summary.csv

Source Repository

Full codebase including cleaning pipelines, statistical notebooks, cluster scripts, Streamlit app, and documentation.

View on GitHub

Flagship Notebooks

These notebooks carry the narrative from exploration to activation. I published nbviewer links so reviewers can walk through the storytelling without cloning the repository.

Segmentation Blueprint (05)

I sequence clustering, persona profiling, and activation scoring that feeds the executive dashboard and pilot playbooks.

View on nbviewer View on GitHub

Interested in the data contracts, cleaning pipelines, or dashboard engineering notebooks? Explore the full collection on GitHub .

End-to-End Narrative

From the first data audit to executive recommendations, I stitched every phase together so decision-makers can trace insights back to governed, well-documented steps.

01 | Align I mapped business questions to data availability, wrote the data card, and locked quality expectations in a Pandera contract.
02 | Explore I profiled distributions, quantified missingness, and surfaced churn drivers with statistical testing and scenario-ready visuals.
03 | Activate I productionised the clean dataset, scored clusters, and delivered a Streamlit control tower with direct downloads for analysts and executives.