CUBE

CUBE (a, b) produces every subset of its columns, (a, b), (a), (b), (): 2^n groupings, the classic OLAP cube. Does a coupon change basket size, and does it differ by channel?

SELECT coalesce(channel, '(all)') AS channel, coalesce(promo, '(all)') AS promo,
       count(DISTINCT order_id) AS orders, round(avg(gross_amount), 2) AS avg_line
FROM (SELECT *, CASE WHEN coupon IS NULL THEN 'none' ELSE 'coupon' END AS promo
      FROM mart.sales WHERE status <> 'cancelled') s
GROUP BY CUBE (channel, promo)
ORDER BY 1, 2;

 channel | promo  | orders | avg_line
---------+--------+--------+----------
 (all)   | (all)  |  94110 |    25.43
 (all)   | coupon |   7579 |    25.36
 (all)   | none   |  86531 |    25.44
 android | (all)  |  32869 |    25.39
 ...
 web     | coupon |   1520 |    26.22
 web     | none   |  17503 |    25.43

Twelve rows answer both questions: coupons did not raise the average line in this sample data, except slightly on the web. coalesce labels rolled-up rows safely only because neither column has NULLs. Cubes grow fast (five dimensions give 32 groupings), so list just the combinations you need with GROUPING SETS.