Churn front-loads: the first three months are twice as dangerous as the back half of the year.

Churn hazard by tenure

WITH RECURSIVE months(n) AS (
  SELECT 1 UNION ALL SELECT n + 1 FROM months WHERE n < 12
),
activation AS (
  SELECT s.id, e.at AS t0, s.canceled_at
  FROM subscriptions s
  JOIN subscription_events e ON e.subscription_id = s.id AND e.kind = 'activate'
  WHERE s.billing = 'monthly'
),
bounds AS (SELECT datetime(MAX(at)) AS data_end FROM subscription_events),
exposure AS (
  SELECT m.n AS tenure_month,
         CASE WHEN datetime(a.t0, '+' || m.n || ' months') <= b.data_end
               AND (a.canceled_at IS NULL OR datetime(a.canceled_at) >= datetime(a.t0, '+' || (m.n - 1) || ' months'))
              THEN 1 ELSE 0 END AS at_risk,
         CASE WHEN datetime(a.canceled_at) >= datetime(a.t0, '+' || (m.n - 1) || ' months')
               AND datetime(a.canceled_at) <  datetime(a.t0, '+' || m.n || ' months')
              THEN 1 ELSE 0 END AS canceled
  FROM months m CROSS JOIN activation a CROSS JOIN bounds b
)
SELECT tenure_month,
       SUM(at_risk)  AS subscriptions_at_risk,
       SUM(canceled) AS cancels,
       ROUND(100.0 * SUM(canceled) / SUM(at_risk), 2) AS hazard_pct
FROM exposure
GROUP BY tenure_month
ORDER BY tenure_month;

does not hold on the current window early hazard 5.94667 is below 1.8x late hazard 3.46167