Reading Cohort Curves

# Reading Cohort Curves A retention curve plots, for a cohort of users, what % is still active over time. ## A second worked example: two curves, same D1, very different stories A note-taking app and a habit-tracking app both launch cohorts with 40% D1 retention - a strong start. Six weeks later: - **Note-taking app:** 40% (D1) -> 22% (D7) -> 9% (D30) -> 3% (D42), still falling with no sign of leveling off. - **Habit-tracking app:** 40% (D1) -> 19% (D7) -> 14% (D30) -> 13% (D42), flat from D30 onward. Same starting point, opposite diagnosis. The note-taking app is a **leaky bucket** - whatever's driving that D1 spike (a good onboarding demo, maybe) isn't translating into a habit, and the curve gives no evidence it ever will; the fix has to happen upstream, in what happens right after signup. The habit-tracking app's curve is **flat**, not high - only 13% stick - but that 13% is real, durable product-market fit for a niche audience, and the fix there is a growth problem (get more of the right users in), not a retention problem. The number that matters isn't D1. It's whether the curve **stops falling**. A curve that's still dropping at D42 hasn't told you its real retention rate yet - it's told you a lower bound. ## Cohort retention matrix & heatmap in Python ```python import pandas as pd import numpy as np import seaborn as sns import matplotlib.pyplot as plt # Create cohort activity dataframe (Cohort Month x Months Since Signup) cohorts = ["2026-01", "2026-02", "2026-03", "2026-04", "2026-05"] retention_data = [] # Retention profile: Month 0 (100%), M1 (45%), M2 (30%), M3 (25%), M4 (24%) base_rates = [1.0, 0.45, 0.30, 0.25, 0.24] for i, c in enumerate(cohorts): for m in range(len(cohorts) - i): retention_data.append({ "cohort_month": c, "month_number": m, "retention_rate": base_rates[m] + np.random.uniform(-0.02, 0.02) }) df_ret = pd.DataFrame(retention_data) pivot_ret = df_ret.pivot(index="cohort_month", columns="month_number", values="retention_rate") print("--- Cohort Retention Matrix ---") print((pivot_ret * 100).round(1).astype(str) + "%") # Visualization plt.figure(figsize=(8, 4)) sns.heatmap(pivot_ret, annot=True, fmt=".1%", cmap="YlGnBu", cbar=False) plt.title("User Cohort Retention Heatmap") plt.xlabel("Months Since Signup") plt.ylabel("Cohort Month") plt.tight_layout() # plt.show() ``` ## Takeaway Stabilization of the curve indicates product-market fit. An asymptote above 20% proves long-term user retention.