P8. Retention: retention()

34 / 34Machine translation

←

APPENDIX · P8 · METRICS · FOR THE SLIDE20

← Back to the slide20

SELECT sum(r[1]) AS d0, sum(r[2]) AS d1, sum(r[3]) AS d7
FROM (
  SELECT player_id, retention(
      toDate(ts) = '2026-09-01',
      toDate(ts) = '2026-09-02',
      toDate(ts) = '2026-09-08') AS r
  FROM events
  WHERE toDate(ts) IN
    ('2026-09-01', '2026-09-02', '2026-09-08')
  GROUP BY player_id);

What the result should be

d0      d1      d7
55860   31192   31366

D1 = 31192 / 55860 ≈ 56 %
D7 = 31366 / 55860 ≈ 56 %

retention()

Synthetic data

Cohort

r[1] — the first condition is met; r[i] — the first and i-th conditions are both met. A player without a Day 0 is not counted.

The players were selected at random, so D1 ≈ D7. In this particular game, D7 is significantly lower than D1.

Here, d0 refers to those active as of September 1. The registration cohort retrieves the date from the players table in PostgreSQL.

Withholding is determined based on the words used prior to the request. Which day is considered day zero, what counts as a return, and which time zone applies: document this as a flowchart.