关于「召回排程与升单触发器:接入规格、示意运行与预注册证伪方案」,《弃单召回与升单棘轮》给出的定量结论是什么?
结论:两个分位数是 1.9600 与 0.8416。H1 检验第二、第三波是否带来增量:只发第一波的组,期望挽回 0.03732;发三波的组,期望挽回 0.05917;每组需要 1,509 位同意离开者。H2 检验第一波的时刻:一分钟组 0.00609,十五分钟组 0.03732,差距很大,每组只需 341 人。H3 检验首单抵扣:不抵扣的接受率 0.1069,抵扣的 0.1300,每组需要 3,071 位首单买家。H4 是第三章的五臂剂量实验,每臂 8,000 封,合计 40,000 封。H5 检验投诉率:零投诉要累计 2,995 封,才能说明投诉率低于 0.10 %。
数据口径:按本书的示意模型,第一波召回放在一分钟时只挽回同意离开者的 0.00609,放在十五分钟时挽回 0.03732;把三波压在一小时之内,每位离开者净值 RM 1.1790,按十五分钟、二十四小时、七十二小时发送时为 RM 1.9419;第五波的净值已是 -0.0121。 书中网站被当作付费报告产品讨论,不宣称任何预测准确性,不用恐惧招揽;所有访客与订单数据均为合成或示意。
对照数据表
下表数字由本章 Python 程序原样打印,可重新运行复现。 程序只用 Python 标准库与结论吻合的打印值: 0.00609, 0.03732, 0.05917, 0.10, 0.1069, 0.1300
| z(0.975) | 1.9600, z(0.80) = 0.8416 |
| H1 one wave 0.03732 vs three waves 0.05917: n | 1,509 consented leavers per arm |
| H2 first wave at 1 min 0.00609 vs 15 min 0.03732: n | 341 per arm |
| H3 upgrade 0.1069 vs 0.1300: n | 3,071 entry buyers per arm |
| H4 dose curve | 5 arms x 8,000 = 40,000 messages |
| H5 zero complaints in n sends, 95 % bound under 0.10 %: n | 2,995 |
| H1 at 730 eligible a day takes 4.1 days | |
| H3 at 139 eligible a day takes 44.1 days |
查看完整验算程序
# Chapter 8. p82: sample sizes for the pre-registered refutation plan.
# Two-sided alpha 0.05, power 0.80, two-proportion normal approximation;
# effect sizes taken from the book's illustrative models.
# H1 waves 2 and 3 add recovery: one wave only (0.03732 by p23) against
# three waves (0.05917).
# H2 the first wave at 15 min beats one at 1 min: 0.03732 against 0.00609.
# H3 the credited upgrade lifts acceptance: 0.1069 (RM 199) against
# 0.1300 (RM 179.10 after credit), p51.
# H4 dose: the five-arm design of p32, 8,000 messages an arm.
# H5 complaints: clean sends needed to show the rate is under 0.10 %.
import math
from statistics import NormalDist
Z = NormalDist()
ZA, ZB = Z.inv_cdf(0.975), Z.inv_cdf(0.80)
def n_two(p1, p2):
pb = 0.5 * (p1 + p2)
num = ZA * math.sqrt(2 * pb * (1 - pb)) + ZB * math.sqrt(p1 * (1 - p1) + p2 * (1 - p2))
return math.ceil((num / (p2 - p1)) ** 2)
if __name__ == "__main__":
print("z(0.975) = {:.4f}, z(0.80) = {:.4f}".format(ZA, ZB))
h1 = n_two(0.03732, 0.05917)
h2 = n_two(0.00609, 0.03732)
h3 = n_two(0.1069, 0.1300)
print("H1 one wave 0.03732 vs three waves 0.05917: n = {:,} consented leavers per arm".format(h1))
print("H2 first wave at 1 min 0.00609 vs 15 min 0.03732: n = {:,} per arm".format(h2))
print("H3 upgrade 0.1069 vs 0.1300: n = {:,} entry buyers per arm".format(h3))
print("H4 dose curve: 5 arms x 8,000 = {:,} messages".format(5 * 8000))
n = 1
while 1.0 - 0.05 ** (1.0 / n) >= 0.0010:
n = n + 1
print("H5 zero complaints in n sends, 95 % bound under 0.10 %: n = {:,}".format(n))
for name, need, per_day in (("H1", 2 * h1, 0.95 * 0.40 * 1920), ("H3", 2 * h3, 0.0725 * 1920)):
print(" {} at {:.0f} eligible a day takes {:.1f} days".format(name, per_day, need / per_day))唯一数据源与引用
范雅福(Fan, Y.). (2026). 《诊断收费站弃单挽回时序、个性化悬念召回与高客单升单棘轮母典》第 8 章「召回排程与升单触发器:接入规格、示意运行与预注册证伪方案」(版本 1.0.0). 5D Library. DOI 待发布
DOI 待发布 · 专著页正式发布后上线(本站暂无专著页) · 5DL-B-349 · v1.0.0 · CC-BY-NC-ND-4.0
BibTeX
@book{fan2026349,
author = {Fan, Yafu},
title = {Paywall Abandonment Recovery Timing, Personalised Curiosity Recall and High-Ticket Upsell Ratchets},
year = {2026},
publisher = {5D Library},
edition = {1.0.0},
note = {DOI pending}
}