5D Library › 弃单挽回动力学、沉没数据续费棘轮与高客单顾问筛选机制 · English
第四章 数据导出的代价与年付方案的自我选择
把数据还给用户要付出多少
如果续费真的来自数据棘轮,一键导出就会削弱它。欧盟《通用数据保护条例》第二十条规定了数据可携权 (European Parliament and Council of the European Union 2016);马来西亚《2024 年个人数据保护(修正)法令》(Act A1727)于 2024 年 10 月在宪报颁布,其第 9 条在《2010 年个人数据保护法令》中新增第 43A 条「数据可携权」:数据主体可以书面电子通知要求数据控制者把其个人数据直接传送给另一数据控制者,以技术可行与数据格式兼容为限;该条自 2025 年 6 月 1 日起生效 (Parliament of Malaysia 2024)。不论法律是否要求,「把用户的数据扣作人质」都是一种暗模式式的留存 (Mathur et al. 2019)。
推导
# Chapter 4. p07: What one-click data export costs, and what it must
# buy back. If a share e of the saved history can be exported, only
# (1 - e) of the switching cost remains. Expected months paid fall; the
# loss is recovered only if export (shown before sign-up) raises the
# sign-up rate by at least the same ratio. Uses the ratchet of p05.
import importlib.util
from math import exp
def load(name):
spec = importlib.util.spec_from_file_location(name, name + ".py")
m = importlib.util.module_from_spec(spec)
spec.loader.exec_module(m)
return m
p05 = load("p05_sunk_data")
def tenure(e, months=240):
alive, total = 1.0, 1.0
for t in range(months):
z = p05.C + (1 - e) * p05.s(t) / p05.W - p05.X
alive *= 1 / (1 + exp(-z))
total += alive
return total
if __name__ == "__main__":
full = tenure(0.0)
print(" no export: {:.1f} months paid".format(full))
for e in (0.5, 0.8, 1.0):
t = tenure(e)
print(" export share {:.1f}: {:.1f} months ({:+.1%});"
" break-even sign-up lift {:+.1%}".format(
e, t, t / full - 1, full / t - 1))
# no export: 15.9 months paid
# export share 0.5: 7.7 months (-51.8%); break-even sign-up lift +107.4%
# export share 0.8: 5.8 months (-63.7%); break-even sign-up lift +175.5%
# export share 1.0: 5.0 months (-68.4%); break-even sign-up lift +216.4%在第三章的棘轮模型中,没有导出时平均付费 15.9 个月;能导出一半的历史时降到 7.7 个月(-51.8%),能全部导出时 5.0 个月(-68.4%)。要弥补这个损失,导出功能(在注册前就告诉买家)必须让注册率提高 107.4% 到 216.4%,这几乎不可能。但这个结论完全依赖棘轮模型:如果续费上升来自筛选(第三章 p06),导出不会损失任何续费。
注
实务含义。 数据导出的真实代价取决于第三章的问题:留存来自锁定还是来自价值。在区分两者之前就砍掉导出,是用一个未经检验的假设去换用户的信任。
年付方案:谁会选
推导
# Chapter 4. p08: Annual prepay as a self-selection device. Monthly price
# RM 49; annual plan RM 490 ("two months free"). Subscribers differ in
# churn theta ~ Beta(a, b) from the sBG fit of p06 (a = 0.65, b = 2.7).
# A subscriber who knows her theta takes the annual plan when her
# expected monthly payments in the first year exceed RM 490. First-year
# revenue per subscriber: monthly only versus the menu.
import random
A, B, PRICE, ANNUAL = 0.65, 2.7, 49.0, 490.0
def expected_monthly(theta):
return PRICE * sum((1 - theta) ** k for k in range(12))
def year_revenue(menu, aware=1.0, n=200000, seed=79):
rng = random.Random(seed)
total, takers = 0.0, 0
for _ in range(n):
th = rng.betavariate(A, B)
guess = th if rng.random() < aware else rng.betavariate(A, B)
if menu and expected_monthly(guess) > ANNUAL:
total += ANNUAL
takers += 1
else:
total += expected_monthly(th)
return total / n, takers / n
if __name__ == "__main__":
m0, _ = year_revenue(False)
print(" monthly only: RM {:.2f} per subscriber in year 1".format(m0))
for aware in (1.0, 0.5, 0.0):
m1, share = year_revenue(True, aware)
print(" menu, buyers know their churn {:.0%}: RM {:.2f}"
" ({:+.1%}); annual share {:.2f}".format(
aware, m1, m1 / m0 - 1, share))
print(" annual pays off for theta below {:.3f}".format(
min(t / 1000 for t in range(1, 1000)
if expected_monthly(t / 1000) <= ANNUAL)))
# monthly only: RM 316.98 per subscriber in year 1
# menu, buyers know their churn 100%: RM 304.25 (-4.0%); annual share 0.22
# menu, buyers know their churn 50%: RM 330.15 (+4.2%); annual share 0.22
# menu, buyers know their churn 0%: RM 354.91 (+12.0%); annual share 0.22
# annual pays off for theta below 0.034月费 RM 49,年付 RM 490(「送两个月」)。订阅者的流失概率按第三章拟合的 sBG 分布各不相同。只有月付时,每位订阅者第一年平均付 RM 316.98。若订阅者清楚自己会不会长期使用,流失概率低于 0.034 的人会选年付(约 0.22 的人),而这些人本来就会付满 12 个月:加上年付方案后第一年收入反而降到 RM 304.25(-4.0%)。只有当选择与真实流失无关时,年付才把本来会流失的人锁住,收入升到 RM 354.91(+12.0%);知道一半时 RM 330.15(+4.2%)。
假说 · 尚未验证
作者的假说(尚未验证): 作者的订阅产品中,选年付者的月付流失率低于全体平均的一半。什么会证明它错: 按年付者与月付者首月行为匹配之后,两组的潜在流失率没有显著差别。
小结
在棘轮模型中,一键导出会让平均付费月数下降一半以上,但在筛选模型中毫无代价;先查明留存的来源再决定。年付方案是一个自我选择的装置:选它的人越是本来就会留下的人,它越亏钱。
参考文献
Fan, Y. (2026). Cart-Abandonment Recovery Kinetics, Sunk-Data Retention Ratchets and High-Ticket B2B Qualification Mechanisms: Multi-Pulse Follow-Up Timing, Switching-Cost Accumulation and Screening Equilibria [弃单挽回动力学、沉没数据续费棘轮与高客单顾问筛选机制] (Version 1.0.0). 5D Library. https://doi.org/10.5281/zenodo.24090209 · Licensed under Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International (CC BY-NC-ND 4.0).