What the score buys you over calling at random
cumulative gains, holdout set · illustrative
0
20%
50% of list contacted
100%
0
50%
100%
top 20% of the list → 58% of conversions
random gets 20% — the model is 2.9× baseline here
← random at the same depth
House rule: no score ships without beating the trivial baseline — so the baseline is drawn first, on every gains chart, every time.