关于「实证闭环:熟客台账与定价与成交工具规格」,《熟客唤醒与凑单闪购》给出的定量结论是什么?
结论:熟客台账分四张表,周一导出、周内执行、周末回填、每季检验。蒸馏小工具把第三章 p31、第四章 p41 与第六章 p62 的公式合在一起,输入示意默认值时给出:面额 RM 6 的券只发间隔比 0.85 到 11.37 的熟客,免邮门槛 RM 170,10 件尾单两小时闪购定价 RM 85,每月闪购最多 1 场。三项预注册检验分别需要每组 3,100 位熟客、9,588 位访客与 3,551 位熟客,结果不论正负都要记录。
数据口径:示意参数下,面额 RM 6 的券只在间隔比 0.85 到 11.37 之间净值为正,刚买客与久睡客都是负值;不设对照组的核销记账会把增量高估 12.2 倍。 书中参数为示意,不对应 HOMIE-X 任何真实客户、订单或供货商,书中价格一律为示意;全部计算由书中程序生成。
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: 170, 3100, 3551, 9588
z_a + z_b = 2.8016 T1 coupon holdout: delta RM 0.451, sd 6.00 / 6.67 buyers per arm 3,100 T2 threshold 170 vs none: delta RM 0.772, sd 18.09 / 20.04 visitors per arm 9,588 T3 flash erosion 0.30 vs 0.27: regulars per arm 3,551 T1 with a 2,000-buyer test (1,000 per arm): power 0.356
查看完整验算程序
# Chapter 8. p82: sample sizes for the three pre-registered tests.
# Two-sided alpha 0.05, power 0.80; n per arm = (z_a + z_b)^2 (v1 + v2)
# / delta^2. Inputs come from the book's own programs (illustrative):
# T1 coupon holdout in the band (p33 cell: P_alive 0.515, anyway 0.199,
# lift 0.120, RHO 0.01; margin RM 22, coupon RM 6): delta = the true
# incremental margin per buyer, variances of a two-point outcome.
# T2 free-shipping threshold RM 170 against no free shipping (p41
# model): delta and per-visitor variance by quadrature.
# T3 flash erosion: 30-day full-price repurchase among regulars 0.30
# without flashes against 0.27 with weekly flashes (two proportions).
import math
from statistics import NormalDist
ZA = NormalDist().inv_cdf(0.975)
ZB = NormalDist().inv_cdf(0.80)
K = (ZA + ZB) ** 2
def n_arm(v1, v2, delta):
return int(math.ceil(K * (v1 + v2) / delta ** 2))
def coupon_test():
pa, b, u, rho, m, c = 0.515, 0.199, 0.120, 0.01, 22.0, 6.0
pt = pa * (b + u) + (1.0 - pa) * rho
ph = pa * b
vt = (m - c) ** 2 * pt * (1.0 - pt)
vh = m ** 2 * ph * (1.0 - ph)
delta = pt * (m - c) - ph * m
return delta, vt, vh
MED, SD, MU, MUF, S, F, ALPHA, DELTA, OVER = (
85.0, 0.55, 0.30, 0.15, 12.0, 12.0, 0.25, 25.0, 4.0)
Z = [i / 100.0 for i in range(-600, 601)]
WZ = [math.exp(-z * z / 2.0) / math.sqrt(2.0 * math.pi) / 100.0 for z in Z]
def moments(t):
"""Mean and variance of margin per visitor under threshold t."""
m1, m2 = [], []
for w, z in zip(WZ, Z):
b = MED * math.exp(SD * z)
if b >= t:
outs = [(1.0, MU * b - S)]
else:
d = t - b
tau = math.exp(-d / DELTA)
stay = (1.0 - tau) * (1.0 - ALPHA)
outs = [(tau, MU * b + MUF * (d + OVER) - S),
(stay, MU * b + F - S), (1.0 - tau - stay, 0.0)]
for pr, v in outs:
m1.append(w * pr * v)
m2.append(w * pr * v * v)
mean = math.fsum(m1)
return mean, math.fsum(m2) - mean * mean
if __name__ == "__main__":
print("z_a + z_b = {:.4f}".format(ZA + ZB))
d, vt, vh = coupon_test()
print("T1 coupon holdout: delta RM {:.3f}, sd {:.2f} / {:.2f}".format(
d, math.sqrt(vt), math.sqrt(vh)))
print(" buyers per arm {:,}".format(n_arm(vt, vh, d)))
m_t, v_t = moments(170.0)
m_0, v_0 = moments(400.0)
print("T2 threshold 170 vs none: delta RM {:.3f}, sd {:.2f} / {:.2f}"
.format(m_t - m_0, math.sqrt(v_t), math.sqrt(v_0)))
print(" visitors per arm {:,}".format(n_arm(v_t, v_0, m_t - m_0)))
p0, p1 = 0.30, 0.27
print("T3 flash erosion 0.30 vs 0.27: regulars per arm {:,}".format(
n_arm(p0 * (1 - p0), p1 * (1 - p1), p0 - p1)))
print("T1 with a 2,000-buyer test (1,000 per arm): power {:.3f}".format(
NormalDist().cdf(d / math.sqrt((vt + vh) / 1000.0) - ZA)))唯一数据源与引用
范雅福(Fan, Y.). (2026). 《私域代购熟客复购唤醒、满额凑单毛利配比与限时清仓脉冲闪购母典》第 8 章「实证闭环:熟客台账与定价与成交工具规格」(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-353 · v1.0.0 · CC-BY-NC-ND-4.0
BibTeX
@book{fan2026353,
author = {Fan, Yafu},
title = {Repeat-Buyer Reactivation in Private-Channel Commerce, Threshold Bundle Margin Design and Flash-Clearance Dynamics},
year = {2026},
publisher = {5D Library},
edition = {1.0.0},
note = {DOI pending}
}