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

第五章 恐惧框架的总账:退款、拒付与复购

一本完整的账

文案的效果不只体现在当次转化率上。一句暗示「不买就会失去什么」的文案,若那个「失去」并不是购买所能避免的,买家事后会发现,表现为退款、信用卡拒付(每笔还要付手续费)和投诉。程序以每 1,000 名访客、RM 49 的价格、3% 的基线转化率比较四种写法;转化倍数、退款率、拒付率与投诉成本都是示意参数。

推导

# Chapter 5. p09: The ledger of a fear frame - refunds, chargebacks, risk.
# Per 1,000 visitors at RM 49. A copy variant raises conversion by a
# factor m but, if it implies a loss the product will prevent, raises the
# refund rate r, the chargeback rate b (fee RM 60 each, plus the sale)
# and the expected cost of complaints (probability per sale x cost).
# Net revenue = sales x (49 (1 - r - b) - 60 b - complaint cost).
def net(visitors, conv, m, r, b, comp_p, comp_cost, price=49.0, fee=60.0):
    sales = visitors * conv * m
    return sales, sales * (price * (1 - r - b) - fee * b - comp_p * comp_cost)

if __name__ == "__main__":
    base_conv = 0.03
    rows = (("honest gain frame", 1.00, 0.03, 0.002, 0.000, 0),
            ("loss of access (deadline)", 1.08, 0.04, 0.003, 0.000, 0),
            ("implied loss of fortune", 1.25, 0.12, 0.015, 0.002, 5000),
            ("implied loss, heavy", 1.50, 0.20, 0.030, 0.004, 5000))
    print("   variant                     sales   net RM   vs honest")
    ref = None
    for name, m, r, b, cp, cc in rows:
        s, v = net(1000, base_conv, m, r, b, cp, cc)
        ref = ref or v
        print("   {:26s} {:6.1f}  {:7.2f}   {:+.1%}".format(name, s, v,
                                                         v / ref - 1))
    print("\n   break-even complaint probability, 'implied loss'"
          " (m 1.25):")
    s, v0 = net(1000, base_conv, 1.25, 0.12, 0.015, 0, 0)
    _, vh = net(1000, base_conv, 1.00, 0.03, 0.002, 0, 0)
    print("     {:.4f} per sale at RM 5000 per complaint".format(
        (v0 - vh) / (s * 5000)))
#    variant                     sales   net RM   vs honest
#    honest gain frame            30.0  1419.36   +0.0%
#    loss of access (deadline)    32.4  1513.50   +6.6%
#    implied loss of fortune      37.5  1180.69   -16.8%
#    implied loss, heavy          45.0   716.85   -49.5%
#
#    break-even complaint probability, 'implied loss' (m 1.25):
#      0.0007 per sale at RM 5000 per complaint

诚实的收益框架净收入 RM 1419.36;以「优惠截止」这种关于报告本身的真实损失为框架,转化倍数 1.08、退款率略升,净收入 RM 1513.50(+6.6%);暗示「失去好运」的写法,即使转化倍数达到 1.25,计入退款、拒付和投诉成本后净收入反而低 16.8%;更激进的写法低 49.5%。即使不算任何投诉,只要每笔销售引起投诉的概率超过 0.0007(每次投诉成本按 RM 5,000 计),暗示损失的写法就不如诚实写法。

复购:承诺被经验检验

推导

# Chapter 5. p10: Repeat purchase when claims are checked by experience.
# A buyer's trust t (probability the next claim is true) starts at a
# Beta(a, b) prior with mean 0.6 and is updated after each purchase by
# whether the copy's promise was experienced as kept (probability k).
# Repurchase probability each month = 0.5 x current mean trust. We
# compare copy whose promise is about the product itself (kept 0.9)
# with copy implying outcomes the product cannot secure (kept 0.3).
import random

def ltv(kept, months=12, price=49.0, trials=20000, seed=77010):
    rng = random.Random(seed)
    total = 0.0
    for _ in range(trials):
        a, b = 3.0, 2.0
        rev = price                                 # first purchase
        for _ in range(months - 1):
            if rng.random() < 0.5 * a / (a + b):
                rev += price
                if rng.random() < kept:
                    a += 1
                else:
                    b += 1
        total += rev
    return total / trials

if __name__ == "__main__":
    for name, k in (("promise about the product", 0.9),
                    ("mixed", 0.6), ("implied outcome", 0.3)):
        print("   {:26s} kept {:.1f}: 12-month value RM {:6.2f}".format(
            name, k, ltv(k)))
    # per 1,000 visitors, with conversion and refund rates of p09
    honest = 1000 * 0.03 * 1.00 * ltv(0.9) * (1 - 0.03 - 0.002)
    implied = 1000 * 0.03 * 1.25 * ltv(0.3) * (1 - 0.12 - 0.015)
    print("   per 1,000 visitors after refunds and chargebacks:")
    print("     product promise RM {:,.0f}; implied outcome RM {:,.0f}"
          .format(honest, implied))
#    promise about the product  kept 0.9: 12-month value RM 228.26
#    mixed                      kept 0.6: 12-month value RM 211.17
#    implied outcome            kept 0.3: 12-month value RM 196.19
#    per 1,000 visitors after refunds and chargebacks:
#      product promise RM 6,629; implied outcome RM 6,364

第二个维度是复购。程序假设买家对卖家的信任按每次购买后的经验更新:承诺是关于产品本身(例如「报告包含哪些计算」)的,经验上兑现的概率高(0.9);承诺是关于结果的(「会避免损失」),兑现的概率低(0.3)。一年的人均价值分别为 RM 228.26 与 RM 196.19。按每 1,000 名访客、扣除退款与拒付计算,诚实承诺为 RM 6,629,暗示结果为 RM 6,364:后者多出的首单被更低的复购与更高的退款抵消,而投诉与法律风险还没有计入。

证据

法律边界。 马来西亚《1999 年消费者保护法》禁止在交易中作出误导或欺骗性的陈述 (Laws of Malaysia 1999);预先勾选、隐藏费用、虚假倒计时等「暗模式」已在大规模网站调查中被系统记录 (Mathur et al. 2019)。一个产品若在其自身的出版物中声明「不主张、也不检验任何测算应验」,就不能在销售页上暗示购买可以避免现实损失。

小结

在本书的示意参数下,「真实的损失」(截止、名额)比「暗示的损失」(好运、财运)净收入更高;暗示损失的写法即使提高四分之一的首单,也会被退款、拒付和更低的复购抵消,而法律风险全部落在这一边。

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.

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).