关于「九成五的漏口:每天 RM 5,000 目标的现金算术」,《弃单召回与升单棘轮》给出的定量结论是什么?

结论:B 段讨论测量本身的噪声。若在 n 位访客上测得转化率 0.05,威尔逊 95 % 区间的宽度随 n 缩小:200 位访客时区间从 0.0274 到 0.0896,宽 0.0622,几乎没有信息;1,000 位时从 0.0381 到 0.0653,宽 0.0272;5,000 位时从 0.0443 到 0.0564,宽 0.0121;20,000 位时从 0.0471 到 0.0531,宽 0.0060。一周的数据若只有一两千位访客,转化率差一两个千分点完全可能只是噪声,不能据此宣布某个改动有效。

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。

English version

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: 0.0060, 0.0121, 0.0272, 0.0274, 0.0381, 0.0443

A. 365 simulated days at an expected RM 5,000
   mean RM 4,984, sd RM 723, cv 0.145
   worst day RM 3,082, best day RM 7,278
   share of days at or above RM 5,000: 0.485
   tenth percentile day: RM 4,102
   to reach RM 5,000 on nine days in ten, plan for about 1.219 x the target
   i.e. an expected RM 6,094 a day, or 2,340 visitors
B. Wilson 95 % interval for a measured conversion of 0.05
   n =    200: 0.0274 to 0.0896  (width 0.0622)
   n =  1,000: 0.0381 to 0.0653  (width 0.0272)
   n =  5,000: 0.0443 to 0.0564  (width 0.0121)
   n = 20,000: 0.0471 to 0.0531  (width 0.0060)
查看完整验算程序
# Chapter 1. p13: why an expected RM 5,000 a day is not RM 5,000 every day,
# and how noisy a measured conversion rate is.
# A. Simulated year (fixed seed): visitors a day 1,920 (p11 scenario 3),
#    each converts with 0.0725; a buyer brings RM 19.90 and with 0.10
#    also the RM 179.10 upgrade; 0.95 kept. Share of days at or above
#    RM 5,000, and the planning level for nine days in ten.
# B. Wilson 95 % interval for a conversion measured at 0.05 on n visitors.
import math
import random

ENTRY, GAP, KEEP = 19.90, 179.10, 0.95


def poisson_small(rng, lam):
    l, k, p = math.exp(-lam), 0, 1.0
    while True:
        p = p * rng.random()
        if p <= l:
            return k
        k = k + 1


def poisson(rng, lam):
    # exact draw: a sum of independent Poisson chunks of mean at most 30
    chunks = int(lam // 30.0)
    counts = [poisson_small(rng, 30.0) for _ in range(chunks)]
    counts.append(poisson_small(rng, lam - 30.0 * chunks))
    return sum(counts)


def day(rng, visitors, conv, up):
    buyers = poisson(rng, visitors * conv)
    cash = [KEEP * (ENTRY + (GAP if rng.random() < up else 0.0)) for _ in range(buyers)]
    return math.fsum(cash)


def wilson(x, n, z=1.959964):
    p = x / n
    den = 1 + z * z / n
    mid = (p + z * z / (2 * n)) / den
    half = z * math.sqrt(p * (1 - p) / n + z * z / (4 * n * n)) / den
    return mid - half, mid + half


if __name__ == "__main__":
    rng = random.Random(20261120)
    days = sorted(day(rng, 1920, 0.0725, 0.10) for _ in range(365))
    mean = math.fsum(days) / len(days)
    sd = math.sqrt(math.fsum((d - mean) ** 2 for d in days) / (len(days) - 1))
    hit = len([d for d in days if d >= 5000.0]) / len(days)
    p10 = days[int(0.10 * len(days))]
    print("A. 365 simulated days at an expected RM 5,000")
    print("   mean RM {:,.0f}, sd RM {:,.0f}, cv {:.3f}".format(mean, sd, sd / mean))
    print("   worst day RM {:,.0f}, best day RM {:,.0f}".format(days[0], days[-1]))
    print("   share of days at or above RM 5,000: {:.3f}".format(hit))
    print("   tenth percentile day: RM {:,.0f}".format(p10))
    lift = 5000.0 / p10
    print("   to reach RM 5,000 on nine days in ten, plan for about {:.3f} x the target".format(lift))
    print("   i.e. an expected RM {:,.0f} a day, or {:,.0f} visitors".format(5000.0 * lift, 1920 * lift))
    print("B. Wilson 95 % interval for a measured conversion of 0.05")
    for n in (200, 1000, 5000, 20000):
        lo, hi = wilson(0.05 * n, n)
        print("   n = {:>6,}: {:.4f} to {:.4f}  (width {:.4f})".format(n, lo, hi, hi - lo))

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 1 章「九成五的漏口:每天 RM 5,000 目标的现金算术」(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}
}

本书其他章节结论