Retail Churn Dashboard (Practice Project)

A self-directed practice project — turning a public retail transactions dataset into a churn-risk dashboard, built the way I'd want to ship a real one.

Why This Project

I don’t have client data analytics work yet — this is a self-directed project built to practice the full workflow end to end: raw data in, a shipped interactive dashboard out. I used a public retail transactions dataset rather than real business data, but built it with the same standards I’d hold a client project to.

Approach

I loaded the dataset into pandas, engineered basic recency/frequency/monetary features, and trained a simple classifier to produce a churn-risk score per customer. The modeling is intentionally simple — the point of this project was the second half: getting that output in front of someone in a usable form, not just as a notebook.

The Build

Rather than stopping at a notebook, I shipped a SvelteKit dashboard: a risk-ranked customer table, cohort trend charts, and a drill-down view per customer. Charts are built with LayerCake to keep the bundle small and give full control over interaction.

What I Learned

How much of “data analysis” is actually about presentation and usability once the model itself is done
Where my frontend background genuinely helped (the dashboard shipped fast) and where it didn’t (feature engineering and validation took real, separate learning)

Reflection

This is a practice project on public data, not a deployed business tool — I’m being upfront about that. It’s the clearest evidence I have right now that I can take a data question all the way to a usable, shipped interface, which is the combination I want to bring to a junior analyst role.

Links

Repo link goes here once published.