5D Library › 结账顺手加购弹性、一键升单动力学与跨领域推荐矩阵 · English
第四章 客单价乘数与组合定价
乘数的分解
平均客单价乘数 = 1 + 各加购的(接受率 × 价格比)之和 + 升单接受率 × 补差价比。由此可以反过来问:一个宣称的提升需要多高的接受率?
推导
# Chapter 4. p07: What it takes to lift average order value by 22-48%.
# AOV multiplier = 1 + sum over bumps of (take x price ratio) + (upsell
# acceptance x step-up ratio). We solve for the take rate a single bump
# needs at a given price ratio, and list combinations of p02 and p03.
def needed_take(lift, ratio):
return lift / ratio
if __name__ == "__main__":
print(" target lift ratio 0.25 ratio 0.45 ratio 0.55")
for lift in (0.22, 0.30, 0.48):
row = [needed_take(lift, r) for r in (0.25, 0.45, 0.55)]
print(" {:11.0%} ".format(lift) + " ".join(
"{:10.3f}".format(x) if x <= 1 else "{:>10s}".format("impossible")
for x in row))
combos = (("one bump (p01)", 0.499 * 0.451),
("two bumps (p02)", 0.501 * 44 / 98 + 0.505 * 29 / 98),
("two bumps + upsell (p03)",
0.501 * 44 / 98 + 0.505 * 29 / 98 + 0.13 * 300 / 98))
for name, x in combos:
print(" {:26s} AOV multiplier {:.3f}".format(name, 1 + x))
# target lift ratio 0.25 ratio 0.45 ratio 0.55
# 22% 0.880 0.489 0.400
# 30% impossible 0.667 0.545
# 48% impossible impossible 0.873
# one bump (p01) AOV multiplier 1.225
# two bumps (p02) AOV multiplier 1.374
# two bumps + upsell (p03) AOV multiplier 1.772单个加购要把客单价提高 22%,在价格比 0.45 时需要 0.489 的接受率,在价格比 0.25 时需要 0.880;提高 48% 在价格比 0.25 或 0.45 时都不可能(需要的接受率超过 1),在价格比 0.55 时需要 0.873。把前两章的结果合起来,单个加购的乘数为 1.225,两个加购为 1.374,再加付款后的升单为 1.772。「22% 到 48%」这样的说法因此需要说清楚是由哪几个环节组成的。
组合价还是结账加购
把同样两件商品放在商品页上作为组合价(主商品单买 RM 98,两件合买 RM 128),与在结账时加购(主 RM 98,加购 RM 44)相比,哪个更好?这是亚当斯与耶伦的混合捆绑问题 (Adams and Yellen 1976)。
推导
# Chapter 4. p08: A discounted bundle on the product page, or a bump at
# checkout? Values of main item and add-on (lognormal, medians 120 and
# 45, log-s.d. 0.8, correlation rho). Strategy 1: main RM 98 + bump
# RM 44 at checkout. Strategy 2: mixed bundling - main RM 98 alone or
# both for RM 128. Revenue per visitor (Adams and Yellen 1976 logic).
import random
from math import exp, sqrt
def sim(rho, n=200000, seed=78008):
rng = random.Random(seed)
bump = bundle = 0.0
for _ in range(n):
z0 = rng.gauss(0, 1)
z1 = rho * z0 + sqrt(1 - rho * rho) * rng.gauss(0, 1)
vm, va = 120 * exp(0.8 * z0), 45 * exp(0.8 * z1)
if vm > 98:
bump += 98 + (44 if va > 44 else 0)
s_alone, s_both = vm - 98, vm + va - 128
if max(s_alone, s_both) > 0:
bundle += 128 if s_both >= s_alone else 98
return bump / n, bundle / n
if __name__ == "__main__":
print(" rho bump at checkout mixed bundle")
for rho in (-0.5, 0.0, 0.5, 0.9):
a, b = sim(rho)
print(" {:4.1f} {:16.2f} {:12.2f}".format(rho, a, b))
# rho bump at checkout mixed bundle
# -0.5 68.92 95.83
# 0.0 72.48 88.85
# 0.5 76.03 84.16
# 0.9 79.90 80.97在示意模型中,组合价每位访客的收入总是高于结账加购:两件商品估值不相关时为 RM 88.85 对 RM 72.48;负相关(−0.5)时为 RM 95.83 对 RM 68.92;高度正相关(0.9)时两者接近,RM 80.97 对 RM 79.90。原因是组合价也能卖给那些单看主商品不值 RM 98、但两件合起来值 RM 128 的访客,而结账加购只能卖给已经决定买主商品的人。
假说 · 尚未验证
作者的假说(尚未验证): 在作者的报告类商品中,商品页组合价与结账加购同时提供时,总收入高于只提供其中之一。什么会证明它错: 若三臂随机测试中「两者都提供」的收入不显著高于较好的单一方案,这一假说就不成立。
小结
客单价乘数是各环节「接受率 × 价格比」之和;48% 的提升单靠一个加购几乎不可能。在示意模型中,商品页组合价比结账加购多卖给一群「两件合起来才值得」的访客,收入更高。
参考文献
Fan, Y. (2026). Order-Bump Friction Elasticity, One-Click Upsell Kinetics and Cross-Domain Recommendation Matrices: Zero-Reentry Impulse Attachment, Conditional Mutual Information and Cart-Value Multipliers [结账顺手加购弹性、一键升单动力学与跨领域推荐矩阵] (Version 1.0.0). 5D Library. https://doi.org/10.5281/zenodo.24090207 · Licensed under Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International (CC BY-NC-ND 4.0).