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
第五章 两档牌照的菜单与申请表的甄别
先改正一个出处
讨论高客单产品的「筛选」,常引用罗斯柴尔德与斯蒂格利茨的模型。那个模型讲的是竞争性保险市场:多家保险公司提供保单菜单,风险高低不同的投保人自己选择,均衡可能根本不存在 (Rothschild and Stiglitz 1976)。一家卖方为两类客户设计两档产品,更接近穆萨与罗森的垄断菜单问题:高档产品给高估值的客户,低档产品的质量被压低,好让高估值客户不愿冒充 (Mussa and Rosen 1978)。
菜单把客户分开
推导
# Chapter 5. p09: A two-licence menu as a screening device. Two kinds
# of business client: low fit (value RM 14,000 for the basic licence,
# RM 24,000 for licence plus with advisory) and high fit (RM 20,000 and
# RM 52,000). Serving costs RM 2,000 and RM 9,000. Prices RM 9,800 and
# RM 29,800. Check who picks what, the highest plus price that keeps
# high-fit clients choosing it, and profit per client against the share
# of high-fit clients for three menus. Illustrative values.
VAL = {"low": (14000, 24000), "high": (20000, 52000)}
COST = (2000, 9000)
def choice(kind, pb, pp):
vb, vp = VAL[kind]
opts = [(0, "none"), (vb - pb, "basic"), (vp - pp, "plus")]
return max(opts, key=lambda o: (o[0], o[1] != "none"))[1]
def profit(share, pb, pp):
out = 0.0
for kind, w in (("high", share), ("low", 1 - share)):
c = choice(kind, pb, pp)
if c == "basic":
out += w * (pb - COST[0])
elif c == "plus":
out += w * (pp - COST[1])
return out
if __name__ == "__main__":
for kind in ("low", "high"):
vb, vp = VAL[kind]
print(" {:4s}: surplus basic RM {:,}, plus RM {:,} -> {}".format(
kind, vb - 9800, vp - 29800, choice(kind, 9800, 29800)))
cap = VAL["high"][1] - (VAL["high"][0] - 9800)
print(" high-fit clients keep choosing plus up to RM {:,}".format(cap))
print(" room above the RM 29,800 price: RM {:,}".format(cap - 29800))
print(" share high menu RM plus only RM basic only RM")
for share in (0.1, 0.2, 0.3, 0.5):
print(" {:10.1f} {:7,.0f} {:12,.0f} {:13,.0f}".format(
share, profit(share, 9800, 29800),
profit(share, 10 ** 9, 29800), profit(share, 9800, 10 ** 9)))
# low : surplus basic RM 4,200, plus RM -5,800 -> basic
# high: surplus basic RM 10,200, plus RM 22,200 -> plus
# high-fit clients keep choosing plus up to RM 41,800
# room above the RM 29,800 price: RM 12,000
# share high menu RM plus only RM basic only RM
# 0.1 9,100 2,080 7,800
# 0.2 10,400 4,160 7,800
# 0.3 11,700 6,240 7,800
# 0.5 14,300 10,400 7,800示意模型中,合适度低的客户对基础牌照估值 RM 14,000、对进阶牌照(含顾问服务)RM 24,000;合适度高的客户分别为 RM 20,000 与 RM 52,000。定价 RM 9,800 与 RM 29,800 时,低合适度客户选基础牌照(剩余 RM 4,200,进阶为 RM -5,800),高合适度客户选进阶牌照(剩余 RM 22,200,高于基础的 RM 10,200)。菜单把两类客户分开了。只要进阶牌照不超过 RM 41,800,高合适度客户就仍然选它;现价之上还有 RM 12,000 的空间,这就是高合适度客户得到的信息租金。
在高合适度客户占 0.2 时,菜单每位客户利润 RM 10,400,只卖进阶 RM 4,160,只卖基础 RM 7,800;占 0.5 时分别为 RM 14,300、RM 10,400、RM 7,800。在示意参数下,菜单在每一个比例上都更好。
申请表的摩擦什么时候能甄别
推导
# Chapter 5. p10: Does a longer application form screen for fit? Prior
# share of high-fit applicants 0.20. Each extra question makes a share
# of applicants give up: d_high for high-fit, d_low for low-fit. Friction
# improves the mix only if d_low > d_high (the single-crossing condition
# of screening); with equal drop rates it only loses applicants.
PRIOR = 0.20
def after(k, d_high, d_low, n=1000):
h = n * PRIOR * (1 - d_high) ** k
l = n * (1 - PRIOR) * (1 - d_low) ** k
return h, l
if __name__ == "__main__":
for d_high, d_low in ((0.04, 0.04), (0.03, 0.08), (0.06, 0.08)):
print(" drop per question: high {:.2f}, low {:.2f}".format(
d_high, d_low))
for k in (0, 5, 10):
h, l = after(k, d_high, d_low)
print(" {:2d} questions: {:5.1f} high, {:5.1f} low,"
" high share {:.3f}".format(k, h, l, h / (h + l)))
# drop per question: high 0.04, low 0.04
# 0 questions: 200.0 high, 800.0 low, high share 0.200
# 5 questions: 163.1 high, 652.3 low, high share 0.200
# 10 questions: 133.0 high, 531.9 low, high share 0.200
# drop per question: high 0.03, low 0.08
# 0 questions: 200.0 high, 800.0 low, high share 0.200
# 5 questions: 171.7 high, 527.3 low, high share 0.246
# 10 questions: 147.5 high, 347.5 low, high share 0.298
# drop per question: high 0.06, low 0.08
# 0 questions: 200.0 high, 800.0 low, high share 0.200
# 5 questions: 146.8 high, 527.3 low, high share 0.218
# 10 questions: 107.7 high, 347.5 low, high share 0.237「用一份长申请表把不合适的客户挡在门外」,只有在一个条件下成立:不合适的客户每多一道问题就更容易放弃。这就是甄别理论中的单交叉条件。若两类客户每道问题都有 0.04 的人放弃,10 道问题之后高合适度客户从 200 人降到 133.0 人,低合适度从 800 人降到 531.9 人,比例仍然是 0.200:摩擦只是赶走了客户。若高合适度客户每道题放弃 0.03、低合适度 0.08,10 道题后比例升到 0.298;若前者是 0.06,只升到 0.237,而高合适度客户损失近半(剩 107.7 人)。
注
实务含义。 申请表的问题要挑那些对合适的客户容易、对不合适的客户难的:已批准的预算、明确的期限、决策人本人填写。泛泛的长表单只增加摩擦,不改善客户组成。
假说 · 尚未验证
作者的假说(尚未验证): 在作者的企业牌照申请中,加入「已批准预算」一题后,申请者中最终成交者的比例提高,而成交数不下降。什么会证明它错: 随机实验中,加题组的成交数显著下降,或成交比例没有提高。
小结
两档牌照的菜单靠价格差让客户自己选择,是穆萨-罗森式的甄别,而不是罗斯柴尔德-斯蒂格利茨的保险市场。申请表只有在满足单交叉条件时才有甄别作用。
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).