Why first-time buyersnever come back

A purchase analysis tool that finds what stops SuperYou customers reordering, and what wins the second order.

What the data says

Five things decide the second order

Buy 1 Get 1 buyers almost never return

Start them on the variety pack

First taste decides it

Every day of delivery past three costs repeat orders

One tool, four questions

Built for the retention team

Each panel is a tab in the tool. Open it to filter, compare and read the detail behind every chart.

Who comes back?

Repeat rate by first product, offer, price, channel, delivery time, rating, flavour, city tier and age.

Who's about to leave?

A machine-learning model scores every recent first-time buyer on their chance of not ordering again.

Why do they leave?

Reviews from one-time buyers, tagged by their main reason and sized.

What brings them back?

A randomised day-21 test of four nudges shows each one's real effect on the repeat rate.

Machine learning

The model sees it coming

Trained on customers' first orders and tested on it never saw. It ranks who is likely to leave, so outreach goes to the right people first.

AUC, logistic regression (chosen)
AUC, random forest
AUC, offer-only rule

One nudge doesn't fit all

Use of AI

Where AI does the work

Machine learning

Churn prediction

Logistic regression and a random forest are trained on first-order data and compared on held-out customers. The better model scores every recent buyer, and permutation importance shows what drives the score.

Claude, AI assistant

Built with Claude

Claude (through Claude Code) wrote the data simulation, the model pipeline, the dashboard and this site from the project brief.

Planned, not built

Live Claude features

Review theme discovery, a win-back plan writer, plain-English questions on the data and personal win-back messages were designed, but were not implemented due to API issues.

See every number for yourself

Six tabs: overview, drivers, churn model, reasons, win-back plan and method. Every chart is computed from the same 5,000 customers.

Open the tool