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

第六章 顾问时间的分配与订金的信号

二十个通话时段,六十份申请

推导

# Chapter 6. p11: Twenty advisory calls a week, sixty applications.
# Each application has a fit score; the true chance of closing is
# logistic(score), and a closed deal is worth RM 9,800 or RM 29,800
# (30% choose plus). A call costs RM 350 of advisor time. Compare
# first-come-first-served with calling the 20 highest scores, when the
# score is informative (noise 0.5) and when it is mostly noise (2.0).
import random
from math import exp

SLOTS, APPS, WEEKS, CALL = 20, 60, 2000, 350.0
DEAL = 0.3 * 29800 + 0.7 * 9800

def week(rng, noise, ranked):
    apps = []
    for _ in range(APPS):
        z = rng.gauss(-2.2, 1.0)
        p = 1 / (1 + exp(-z))
        apps.append((z + rng.gauss(0, noise), p))
    if ranked:
        apps.sort(key=lambda a: -a[0])
    called = apps[:SLOTS]
    return sum(p for _, p in called) * DEAL - SLOTS * CALL

if __name__ == "__main__":
    print("   deal value RM {:,.0f}; call cost RM {:.0f}".format(DEAL, CALL))
    for noise in (0.5, 2.0):
        res = []
        for ranked in (False, True):
            rng = random.Random(79)
            res.append(sum(week(rng, noise, ranked)
                           for _ in range(WEEKS)) / WEEKS)
        print("   score noise {:.1f}: first come RM {:,.0f} a week".format(
            noise, res[0]))
        print("     top scores RM {:,.0f} a week ({:+.0%})".format(
            res[1], res[1] / res[0] - 1))
#    deal value RM 15,800; call cost RM 350
#    score noise 0.5: first come RM 34,912 a week
#      top scores RM 69,833 a week (+100%)
#    score noise 2.0: first come RM 34,912 a week
#      top scores RM 51,628 a week (+48%)

示意模型中,一位顾问每周有 20 个通话时段,收到 60 份申请;一笔成交平均值 RM 15,800(三成选进阶牌照),每次通话的时间成本 RM 350。按申请先后打电话,每周净收入 RM 34,912。若申请时有一个评分,而评分与真实成交概率相关(噪声 0.5),只打给评分最高的 20 位,每周 RM 69,833(+100%);即使评分噪声很大(2.0),仍有 RM 51,628(+48%)。评分不必准确,只要比随机好。

订金什么时候是信号

可退的订金(成交时抵扣、未成交时退还)常被当作筛选工具。在信号理论中,一个信号要能区分两类人,必须让它对两类人的成本不同 (Spence 1973)。可退订金对申请者的成本只是填写的麻烦与资金的短暂占用,两类客户未必不同。

推导

# Chapter 6. p12: A refundable booking deposit as a signal. Applicants
# are high fit (share 0.20, close with 0.40 after a call) or low fit
# (close with 0.05). A deposit of RM 500, credited to the licence and
# refunded if no deal, costs an applicant only effort and liquidity:
# k_high and k_low in RM. Applicant i books if her expected gain from
# the call exceeds that cost, with gains spread uniformly up to G_high
# and G_low. The deposit screens only if it costs low-fit applicants
# relatively more, or their gains are smaller.
PRIOR, CLOSE_H, CLOSE_L = 0.20, 0.40, 0.05

def book_rate(cost, gain_max):
    return max(0.0, 1 - cost / gain_max)

def after(k_h, k_l, g_h, g_l, n=1000):
    h = n * PRIOR * book_rate(k_h, g_h)
    l = n * (1 - PRIOR) * book_rate(k_l, g_l)
    closes = h * CLOSE_H + l * CLOSE_L
    return h, l, closes / (h + l)

if __name__ == "__main__":
    cases = (("no deposit", 0, 0, 1000, 1000),
             ("same cost, same gains", 150, 150, 1000, 1000),
             ("same cost, low gains smaller", 150, 150, 1000, 300),
             ("cost higher for low fit", 100, 250, 1000, 1000))
    for name, kh, kl, gh, gl in cases:
        h, l, rate = after(kh, kl, gh, gl)
        print("   {:28s}: {:5.1f} high, {:5.1f} low;"
              " close per call {:.3f}".format(name, h, l, rate))
#    no deposit                  : 200.0 high, 800.0 low; close per call 0.120
#    same cost, same gains       : 170.0 high, 680.0 low; close per call 0.120
#    same cost, low gains smaller: 170.0 high, 400.0 low; close per call 0.154
#    cost higher for low fit     : 180.0 high, 600.0 low; close per call 0.131

没有订金时,200 位高合适度与 800 位低合适度申请者都会预约,每次通话成交 0.120。若订金对两类人的成本相同、通话对两类人的预期收益也相同,两类人按同样比例退出(剩 170 与 680),每次通话成交仍是 0.120:订金只减少了预约。只有当低合适度客户从通话中得到的预期收益更小(剩 400 人),或订金对他们更费事(剩 600 人)时,每次通话成交才升到 0.154 或 0.131。

注

伦理边界。 不可退的「申请费」或「诚意金」在许多情况下会被视为不公平的条款;用虚假的名额或倒计时制造紧迫感,属于误导性的暗模式 (Mathur et al. 2019; Laws of Malaysia 1999)。可退、可抵扣、条款写清楚,是订金的底线。

假说 · 尚未验证

作者的假说(尚未验证): 在作者的顾问服务中,按申请评分安排通话,比按先后安排通话,每个通话时段的成交额至少高 30%。什么会证明它错: 随机分配周次的实验中,两种安排每个时段的成交额差距小于 30%。

小结

把有限的顾问时间给评分最高的申请者,即使评分噪声很大,也比先来先得好得多。可退订金只有在对不同客户的成本或收益不同时才是信号;否则它只是减少预约。

参考文献

Laws of Malaysia. 1999. Consumer Protection Act 1999 (Act 599). Part II: Misleading and deceptive conduct, false representation and unfair practice.
Mathur, Arunesh, Gunes Acar, Michael J. Friedman, et al. 2019. “Dark Patterns at Scale: Findings from a Crawl of 11K Shopping Websites.” Proceedings of the ACM on Human-Computer Interaction 3 (CSCW): 81:1–32.
Spence, Michael. 1973. “Job Market Signaling.” The Quarterly Journal of Economics 87 (3): 355–74.

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).