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 aa 在 0 与 1 之间。普通访客的转化率为 0.04+0.08a0.04 + 0.08 a;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 aa 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 aa 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.
唤醒强度与每 1,000 位访客的净收入,以及买家中处于困境者的比例。

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

European Parliament and Council of the European Union. 2005. Directive 2005/29/EC of 11 May 2005 Concerning Unfair Business-to-Consumer Commercial Practices in the Internal Market (Unfair Commercial Practices Directive). Official Journal of the European Union, l 149, 11 June 2005.
OECD. 2022. Dark Commercial Patterns. OECD Digital Economy Papers No. 336. OECD Publishing.

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).