Learn Customer Retention with Phoebe

Two tracks from "our retention is 27%" to knowing what that means and what to do about it

A stakeholder track that teaches why retention is the money metric, how to read a cohort heatmap in thirty seconds, which levers actually move the number - with the famous case studies folklore-checked - and exactly what to demand from your data team. And an analyst track that builds the whole toolkit in Python on real and synthetic data: D1/7/14/30 retention defined precisely (three different "D7"s exist, and tools disagree on purpose), the registration-cohort triangle and its heatmap, the viz gallery, RFM segmentation, survival analysis, CLV models, and churn prediction that actually helps. Every number is cohort-based, forward-looking, and frozen once complete - a completed month's retention never restates, only the current window updates. Ecommerce and game examples throughout. A live in-browser cohort simulator lets you flatten the curve lever by lever and discover what doubling ad spend cannot buy.

12sessions
2tracks
8runnable notebooks
3live widgets

The stakeholder track 🤝 · for leaders, PMs & anyone who reads dashboards · 4 x 45 min · no code to write

Value mode: why a five-point retention move is worth more than a marketing budget, how to read the triangle and the curve without touching Python, which levers are documented to work (Duolingo, Supercell, Amazon Prime) and which famous numbers are folklore, and the four questions that turn any retention readout into a decision.

💰 Stakeholder 1 · easy

Retention is the money metric

The leaky bucket vs compounding, Reichheld's real 1990 numbers, the honest version of "5x cheaper", and why retention precedes monetization everywhere.

▶ Stakeholders start here 🔥 Stakeholder 2 · easy

Read the chart, read the business

The cohort triangle's three scan directions, the three curve shapes and what each means for the company, and benchmarks with their honest vintage.

Triangle + curves 🎮 Stakeholder 3 · medium

Levers that move retention

Duolingo's CURR playbook, Supercell's kill gates, Prime and Sephora loyalty economics, winback - every case labeled documented or folklore.

Folklore-checked 🎯 Stakeholder 4 · medium

Asking the right questions of your DA

Which cohort, what changed, how much money - plus the mix-shift trap where the blended number rises while every segment falls. Drive it yourself.

▶ Mix-shift trap live

The analyst track 🛠️ · for data analysts · 8 x 45 min · pandas + seaborn + lifelines + XGBoost, notebook per session

Build mode: define D1/7/14/30 retention precisely enough that your SQL matches the tool (a1), build the registration-cohort table and heatmap in five lines of pandas (a2), master the chart family and drive the flatten-the-curve simulator (a3), segment with RFM and the power-user curve (a4), go censoring-correct with survival analysis (a5), model CLV with buy-till-you-die (a6), predict churn without leaking or lying (a7), and ship the stakeholder-ready retention report (a8). Synthetic data with known answers first, then the gloriously messy UCI Online Retail II.

📏 Analyst 1 · easy

Retention metrics, precisely

Three different "D7"s exist: classic, rolling, bracket. The frozen-history law, day boundaries, stickiness ratios, and the denominator decisions.

▶ Analysts start here▶ Variant switcher 🔥 Analyst 2 · medium

The cohort table and heatmap

Users who joined month 0, percent back in months 1-10. The canonical pandas recipe, the four heatmap choices, and the three scan directions.

The core session 📈 Analyst 3 · medium

The retention viz gallery

Curve shapes as diagnosis, the truncated-cohort trap, layer-cake, growth accounting, L28 - and the flatten-the-curve simulator with its ad-spend trap.

▶ Live simulator 🧩 Analyst 4 · medium

Segmentation: who is retained

RFM scoring and personas, behavioral vs acquisition cohorts, stickiness by segment, and the wholesalers who eat your averages.

Champions to hibernating ⏳ Analyst 5 · hands-on

Survival analysis

Right-censoring done right: Kaplan-Meier curves, log-rank tests, Cox hazard ratios with lifelines - and when a plain cohort table is enough.

lifelines 💎 Analyst 6 · hands-on

CLV and buy-till-you-die

BG/NBD and P(alive) for customers who never say goodbye, Gamma-Gamma for value, where the models break, and the honest LTV formulas.

P(alive) 🤖 Analyst 7 · hands-on

Churn prediction that helps

Label and feature windows without leakage, XGBoost + SHAP for drivers, thresholds priced by intervention economics - and why prediction is not prevention.

Leakage spot-the-bug 📦 Analyst 8 · capstone

Capstone: the retention report

Full pipeline on Online Retail II, ending in the one-pager a stakeholder can act on: which cohort, what changed, how much money.

Ship it
easy - everyone medium hands-on capstone

Retention is where analytics meets money. It pairs with learn-metric-decomposition-with-phoebe (retention is a driver in every tree), builds on learn-python-data-analysis-with-phoebe (the pandas muscle), and hands off to learn-experimentation-with-phoebe when you want to prove a lever caused the lift.

Built from the Amplitude and Mixpanel retention documentation, GameAnalytics 2025 benchmarks (11,600 games), Sequoia and Reforge on curve shapes, Fader-Hardie buy-till-you-die papers, the lifelines and lifetimes libraries, Jonathan Hsu's growth accounting, and the documented Duolingo, Supercell, and Dollar Shave Club cases - taught with the folklore flagged as folklore.

Choose your path 🗺️

Two doors, one subject. Stakeholders never write code; analysts never sit through a chart they have not just built themselves.

🤝 The stakeholder path (read & decide) s1→ s2→ s3→ s4
🛠️ The analyst path (build & ship) a1→ a2→ a3→ a4→ a5→ a6→ a7→ a8
🚀 The full journey (decide AND build) s1-s4→ a1-a8

The knowledge map 🧠

Both tracks at a glance - hover a session to spotlight its concepts, click any node to jump in.