5D Library › 回溯信念锁定、序贯痛点唤醒与马尔可夫升单自动机 · English
第四章 痛点唤醒的收益与代价,以及两道护栏
「序贯痛点唤醒」是什么
销售页面上的「痛点唤醒」,是在提出产品之前先放大一种恐惧或痛苦:「你今年的财运缺口如果不补……」「这种体感说明你的代谢已经……」。它能提高当下的转化,这一点并无争议。本章要问的是另外三件事:它带来的退款与流失有多少,它在扣除这些之后还剩多少利润,以及它把谁带进了买家之中。
欧盟《不公平商业行为指令》把利用骚扰、强迫或不当影响、明显损害消费者决策自由的行为列为「攻击性商业行为」(第 8、9 条)(European Parliament and Council of the European Union 2005);经合组织把在线界面中操纵消费者选择的设计归为「黑暗商业模式」(OECD 2022)。本章的模型不讨论法律适用,只计算经济后果,但读者应当知道这条边界的存在。
收益与代价
设唤醒强度 在 0 与 1 之间。普通访客的转化率为 ;10% 的访客正处于急性困境(失业、疾病、失恋),他们对唤醒的反应是普通人的三倍。首单 99;退款与拒付的比例随 的平方上升,每次另付 15 的手续费;买家一年内的复购价值为 240,留下的概率随 下降(后悔导致流失)。
推导
# Chapter 4. [p07] Pain-point agitation: what it buys and what it costs.
# Agitation intensity a in [0, 1] (how hard a sales page stirs up a
# fear or a pain). Per visitor: conversion c(a) = 0.04 + 0.08 a for
# ordinary visitors; 10% of visitors are in acute distress and convert
# at 0.04 + 0.24 a. First purchase 99; refund or chargeback share
# phi(a) = 0.05 + 0.35 a^2, each costing a 15 fee on top of the refund;
# buyers stay for repeat purchases worth 240 over a year with
# probability 0.6 - 0.3 a (regret drives churn). Reported per 1,000
# visitors: net revenue, and the share of buyers who were in distress.
def per_thousand(a):
c_ord = 0.04 + 0.08 * a
c_dis = 0.04 + 0.24 * a
buyers = 1000 * (0.9 * c_ord + 0.1 * c_dis)
phi = 0.05 + 0.35 * a * a
first = buyers * ((1 - phi) * 99 - phi * 15)
repeat = buyers * (1 - phi) * (0.6 - 0.3 * a) * 240
distress = 100 * c_dis / buyers
return first, repeat, first + repeat, buyers, distress
if __name__ == "__main__":
print(" a buyers first net repeat net total net"
" distressed share")
rows = []
for k in range(0, 11):
a = k / 10
f, r, t, b, d = per_thousand(a)
rows.append((t, a))
print(" {:3.1f} {:8.1f} {:10.0f} {:11.0f} {:10.0f} {:17.3f}".format(
a, b, f, r, t, d))
first_best = max(range(0, 101),
key=lambda k: per_thousand(k / 100)[0]) / 100
total_best = max(range(0, 101),
key=lambda k: per_thousand(k / 100)[2]) / 100
print(" best a for first-purchase revenue only: {:.2f}".format(
first_best))
print(" best a counting refunds and repeat business: {:.2f}".format(
total_best))
print("\nReading: refunds and churn pull the optimum down, but not")
print("to zero: on its own the market still rewards fairly strong")
print("agitation, and the harder it is, the larger the share of")
print("buyers who are in acute distress. Limits have to be set.")
# a buyers first net repeat net total net distressed share
# 0.0 40.0 3732 5472 9204 0.100
# 0.1 49.6 4608 6422 11030 0.129
# 0.2 59.2 5429 7181 12610 0.149
# 0.3 68.8 6172 7735 13907 0.163
# 0.4 78.4 6814 8074 14889 0.173
# 0.5 88.0 7333 8197 15530 0.182
# 0.6 97.6 7704 8107 15811 0.189
# 0.7 107.2 7906 7811 15717 0.194
# 0.8 116.8 7915 7326 15241 0.199
# 0.9 126.4 7708 6672 14380 0.203
# 1.0 136.0 7262 5875 13138 0.206
# best a for first-purchase revenue only: 0.75
# best a counting refunds and repeat business: 0.63
#
# Reading: refunds and churn pull the optimum down, but not
# to zero: on its own the market still rewards fairly strong
# agitation, and the harder it is, the larger the share of
# buyers who are in acute distress. Limits have to be set.只看首单,最优强度是 0.75;计入退款与复购,最优强度降到 0.63,但没有降到零:在这个模型中,市场本身仍然奖励相当强的唤醒。与此同时,买家中处于困境者的比例从无唤醒时的 0.100 一路升到满强度时的 0.206。唤醒越强,生意就越依赖最脆弱的人。
两道护栏
程序比较两道护栏:其一是 7 天冷静期,任何买家都可以无理由退回首单,唤醒造成的后悔会随着情绪平复而变成退款;其二是困境筛查,对处于急性困境的访客不展示任何唤醒内容。
推导
# Chapter 4. [p08] Two guardrails: a cooling-off refund and a distress
# screen. Model of p07. (i) Cooling-off: every buyer may return the
# first purchase within 7 days, no questions asked; regret surfaces as
# the agitation fades, so the share returning is 0.05 + 0.50 a (no
# fee). (ii) Distress screen: visitors in acute distress are shown
# no sales page with agitation (they convert at 0.04 whatever a).
# The best agitation level and net revenue per 1,000 visitors under
# each regime.
def net(a, cooling, screen):
c_ord = 0.04 + 0.08 * a
c_dis = 0.04 if screen else 0.04 + 0.24 * a
buyers = 1000 * (0.9 * c_ord + 0.1 * c_dis)
if cooling:
back = 0.05 + 0.50 * a
first = buyers * (1 - back) * 99
else:
back = 0.05 + 0.35 * a * a
first = buyers * ((1 - back) * 99 - back * 15)
repeat = buyers * (1 - back) * (0.6 - 0.3 * a) * 240
return first + repeat, 100 * c_dis / buyers
if __name__ == "__main__":
print(" regime best a net/1,000 distressed")
for name, c, s in (("no guardrail (p07)", False, False),
("cooling-off", True, False),
("distress screen", False, True),
("cooling-off + screen", True, True)):
k = max(range(0, 101), key=lambda k: net(k / 100, c, s)[0])
v, d = net(k / 100, c, s)
print(" {:26s} {:7.2f} {:11.0f} {:12.3f}".format(
name, k / 100, v, d))
v0, d0 = net(0.0, True, True)
print(" cooling-off + screen at a = 0: net {:.0f}".format(v0))
print("\nReading: each guardrail lowers the best level of agitation;")
print("together they cut the share of buyers in distress to about")
print("a twentieth, at a cost in net revenue. That cost is the price")
print("of not selling to people in distress.")
# regime best a net/1,000 distressed
# no guardrail (p07) 0.63 15822 0.190
# cooling-off 0.52 12754 0.183
# distress screen 0.57 13488 0.049
# cooling-off + screen 0.43 11059 0.056
# cooling-off + screen at a = 0: net 9234
#
# Reading: each guardrail lowers the best level of agitation;
# together they cut the share of buyers in distress to about
# a twentieth, at a cost in net revenue. That cost is the price
# of not selling to people in distress.两道护栏各自降低了最优唤醒强度:冷静期降到 0.52,困境筛查降到 0.57,两者合用降到 0.43。困境筛查把买家中困境者的比例从 0.190 降到 0.049,两者合用为 0.056。代价是净收入:从 15822 降到 11059(每 1,000 位访客)。这一差额就是「不向处于困境的人推销」的价格。
假说 · 尚未验证
作者的诠释(尚未验证): 对一家测算公司而言,冷静期与困境筛查不是成本,而是长期经营的条件:没有它们,平台投诉、支付渠道风控与监管迟早会以更高的代价强加同样的限制。什么会证明这一判断错:若长期数据显示,不设护栏的同类服务在三年内既没有遭遇显著的投诉与渠道限制,用户的复购也没有下降,那么护栏的商业理由不成立,只剩下伦理理由。作者认为伦理理由本身已经足够。
小结
痛点唤醒在只看首单时显得越强越好;计入退款与流失后最优强度下降,但不会自行降到零,而且越强的唤醒使买家中处于困境者的比例越高。冷静期与困境筛查两道护栏合用,把这一比例降到 0.056,代价约为净收入的三成。
参考文献
Fan, Y. (2026). Retrodictive Belief Locking, Sequential Pain-Point Agitation and Markovian Upsell Automata: Bayesian Trust Avalanches, Sunk-Cost Commitment Ratchets and Cross-Domain Transition Matrices [回溯信念锁定、序贯痛点唤醒与马尔可夫升单自动机] (Version 1.0.0). 5D Library. https://doi.org/10.5281/zenodo.24090186 · Licensed under Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International (CC BY-NC-ND 4.0).