5D Library › 说服文案代数、损失框架不对称性与结账微界面工程 · English

第六章 转化漏斗的上限与最优价格

漏斗是乘积

从「看到免费报告」到「付款」,转化率是各环节通过率的乘积。程序取三个示意环节:读到优惠 0.40,打开结账 0.15,完成付款 0.55。

推导

# Chapter 6. p11: Where conversion can and cannot be gained.
# A funnel: report viewed -> offer read -> checkout opened -> paid.
# Overall conversion is the product of stage rates. A change that lifts
# one stage by a relative x lifts the total by exactly x; the ceiling for
# a stage is reaching 1. We list the maximum total lift available from
# each stage and the lift from realistic 10% stage improvements.
STAGES = (("offer read", 0.40), ("checkout opened", 0.15),
          ("paid", 0.55))

def total(rates):
    out = 1.0
    for r in rates:
        out *= r
    return out

if __name__ == "__main__":
    rates = [r for _, r in STAGES]
    base = total(rates)
    print("   baseline conversion per report viewer: {:.4f}".format(base))
    print("   stage             rate   ceiling lift   +10% relative")
    for i, (name, r) in enumerate(STAGES):
        up = rates[:]
        up[i] = 1.0
        ten = rates[:]
        ten[i] = min(1.0, r * 1.10)
        print("   {:16s}  {:.2f}   {:10.2f} x  {:12.4f}".format(
            name, r, total(up) / base, total(ten)))
    best = rates[:]
    best = [min(1.0, r * 1.10) for r in best]
    print("   all three +10%: {:.4f} ({:+.1%})".format(
        total(best), total(best) / base - 1))
#    baseline conversion per report viewer: 0.0330
#    stage             rate   ceiling lift   +10% relative
#    offer read        0.40         2.50 x        0.0363
#    checkout opened   0.15         6.67 x        0.0363
#    paid              0.55         1.82 x        0.0363
#    all three +10%: 0.0439 (+33.1%)

总转化率为 0.0330。任何一个环节的相对提升会原样传到总体:把任一环节提高一成,总体都是 0.0363。上限则不同:「读到优惠」最多能把总体提高到 2.50 倍,「打开结账」6.67 倍,「完成付款」1.82 倍。也就是说,最大的空间在「打开结账」这一环(读者对报价的兴趣),而不是第三章的结账界面;后者的空间存在,但封顶较低。三个环节各提高一成,总体提高 33.1%。

最优价格与锯齿

推导

# Chapter 6. p12: The revenue-maximising price, with and without the
# left-digit effect. WTP lognormal (median RM 30, log-s.d. 1.0, as in
# p02); perceived price as in p03 with theta 0.3. Revenue per visitor =
# price x P(WTP > perceived price). Grid of whole-ringgit prices.
from math import log, erf, sqrt

def lead(p):
    s = str(int(p))
    return int(s[0]) * 10 ** (len(s) - 1)

def perceived(p, theta):
    L = lead(p)
    return L + (1 - theta) * (p - L)

def share(p, med=30.0, sd=1.0):
    return 0.5 * (1 - erf((log(p) - log(med)) / sd / sqrt(2)))

if __name__ == "__main__":
    for theta in (0.0, 0.3):
        best = max(range(5, 301),
                   key=lambda p: p * share(perceived(p, theta)))
        r = best * share(perceived(best, theta))
        print("   theta {:.1f}: best price RM {} -> RM {:.2f} per visitor"
              .format(theta, best, r))
    print("   theta 0.3, revenue per visitor near the optimum:")
    for p in (38, 39, 40, 41, 48, 49, 50):
        print("     RM {:3d}: {:.3f}".format(p, p * share(perceived(p, 0.3))))
#    theta 0.0: best price RM 41 -> RM 15.47 per visitor
#    theta 0.3: best price RM 39 -> RM 16.55 per visitor
#    theta 0.3, revenue per visitor near the optimum:
#      RM  38: 16.418
#      RM  39: 16.552
#      RM  40: 15.472
#      RM  41: 15.587
#      RM  48: 16.210
#      RM  49: 16.276
#      RM  50: 15.237

在第一章的估值模型中,没有左位数字效应时,每位访客收入最高的价格是 RM 41(RM 15.47)。加入 θ=0.3\theta = 0.3 的左位数字效应后,最优价格移到 RM 39(RM 16.55),而 RM 40 只有 RM 15.472:越过整十,收入曲线出现一个「悬崖」。RM 49 的 16.276 也明显高于 RM 50 的 15.237。收入对价格的曲线因此呈锯齿状,每个整十之前是一个局部高点。

每位访客收入与价格:左位数字效应使曲线呈锯齿状(p12)。

假说 · 尚未验证

作者的假说(尚未验证): 在作者的测算报告中,RM 39 与 RM 49 两个尾数价格的收入差异小于 RM 40 与 RM 39 之间的差异。什么会证明它错: 若按第八章的方法进行三臂随机测试,RM 40 与 RM 39 的收入差不显著,而 RM 49 与 RM 39 的差显著,这一假说就不成立。

小结

总转化率是各环节的乘积;提升任何环节一成,总体提升一成;上限最大的是「打开结账」。左位数字效应使最优价格落在整十之下,收入曲线呈锯齿状。

Fan, Y. (2026). Persuasive Copywriting Algebra, Loss-Framing Asymmetry and Checkout Micro-Architecture: Prospect-Theory Valuation Ratios, Numerical Specificity Credibility and Cognitive-Fluency Conversion Bounds [说服文案代数、损失框架不对称性与结账微界面工程] (Version 1.0.0). 5D Library. https://doi.org/10.5281/zenodo.24090205 · Licensed under Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International (CC BY-NC-ND 4.0).