5D Library › 订阅制单位经济学、净收入留存与人工智能推理标度律 · English

第三章 现金低谷:为什么单位经济学很好也会把钱烧光

藏起时钟的比率

「LTV / CAC 大于 3」是订阅制企业最常被引用的健康指标:每位客户带回的钱至少是获取它所花成本的三倍。这个比率完全不谈何时。获客成本在签约时支付;毛利逐月到来;而一家增长中的企业,总是在为比已收回成本更多的新客户付钱。两者之间的缺口就是现金低谷,它同时取决于增长、回收期与流失。

同一比率,不同增长

第六段程序固定单位经济学——获客成本 1,200,每月毛利 100,月流失 0.025,所以口号式的 LTV / CAC 为 3.33,回收期 12 个月——只改变新客户增长的速度。

推导

# Chapter 3. The cash trough. Every new customer costs CAC up front and then
# returns gross profit GP a month until it churns (probability c a month). New
# customers per month grow at g a month for 36 months and then stay flat;
# fixed costs F a month.
# Baseline: CAC 1,200; GP 100 a month; c = 0.025 (so the slogan LTV = GP / c =
# 4,000 and LTV / CAC = 3.33); payback = CAC / GP = 12 months; 50 new
# customers in month 1; F = 30,000. Cash starts at 0, so the trough is the
# funding the business needs.
def run(
    g, CAC=1200.0, GP=100.0, c=0.025, F=30000.0, n0=50.0, months=72, stop=36
):
    cash, base, trough, t_trough, t_cf = 0.0, 0.0, 0.0, 0, None
    for t in range(months):
        new = n0 * (1 + g) ** min(t, stop)
        base = base * (1 - c) + new
        cf = base * GP - new * CAC - F
        cash += cf
        if cash < trough:
            trough, t_trough = cash, t + 1
        if t_cf is None and cf > 0 and t > 0:
            t_cf = t + 1
    return trough, t_trough, t_cf, cash


if __name__ == "__main__":
    print(
        "Same unit economics (LTV / CAC = 3.33, 12-month payback), different"
        " growth"
    )
    print("   growth/month  funding needed  trough   first cash-     cash at")
    print(
        "                 (trough)        month    positive month  month 72"
    )
    for g in (0.0, 0.03, 0.06, 0.10):
        tr, tt, tc, end = run(g)
        print(
            "   {:6.0%}        {:12,.0f}     {:>3}       {:>5}       "
            " {:12,.0f}".format(g, -tr, tt, tc if tc else "never", end)
        )
    print(
        "\nReading: identical LTV / CAC, four very different cash"
        " requirements. Faster"
    )
    print(
        "growth means more customers who have been paid for and not yet paid"
        " back, so"
    )
    print(
        "the trough deepens and arrives later. A company with good unit"
        " economics can"
    )
    print("still run out of money if its funding is sized to a slower plan.")
# Same unit economics (LTV / CAC = 3.33, 12-month payback), different growth
#    growth/month  funding needed  trough   first cash-     cash at
#                  (trough)        month    positive month  month 72
#        0%             912,868      23          24           1,380,164
#        3%           1,196,391      28          29           4,228,507
#        6%           2,210,878      38          39          10,431,496
#       10%           8,320,737      41          42          32,183,936
#
# Reading: identical LTV / CAC, four very different cash requirements. Faster
# growth means more customers who have been paid for and not yet paid back, so
# the trough deepens and arrives later. A company with good unit economics can
# still run out of money if its funding is sized to a slower plan.

不增长时,企业需要 912,868 的资金,第 23 个月到达低谷,从第 24 个月起现金为正。月增长 6%、持续三年时,需要 2,210,878,第 39 个月转正;月增长 10% 时需要 8,320,737,是前者的九倍,第 42 个月转正。增长更快的企业最终富得多——10% 时到第 72 个月为 32,183,936——但前提是它熬过低谷。

这就是一种常见失败背后的机制:一家单位经济学极好、按某个增长计划融资的公司,增长得比计划快,结果把钱烧光——正是因为它的客户很好。每加速一个月,低谷就更深,而且来得更晚,晚于资金规模被定下来的时候。

单位经济学相同、四种增长速度下的累计现金:增长越快,低谷越深、来得越晚,而熬过去的企业也越富。

同一比率,不同回收期

第七段程序把增长固定在每月 6%,改变客户的形态:获客成本与流失率同步变化,使企业 A 与 B 的口号比率都保持 3.33,而 C 的比率看上去更好。

推导

# Chapter 3. Same LTV / CAC, different payback. Three businesses, all growing
# new customers at 6 % a month for 36 months, with GP = 100 a month per
# customer and F = 30,000 a month. They differ in CAC and churn:
#   A: CAC 1,200, churn 0.025  -> slogan LTV / CAC 3.33, payback 12 months
#   B: CAC 2,400, churn 0.0125 -> slogan LTV / CAC 3.33, payback 24 months
#   C: CAC 3,600, churn 0.0056 -> slogan LTV / CAC 4.96, payback 36 months
# "Discounted LTV / CAC" uses a 1 % monthly discount rate.
def run(g, CAC, GP, c, F=30000.0, n0=50.0, months=96, stop=36):
    cash, base, trough, t_trough = 0.0, 0.0, 0.0, 0
    for t in range(months):
        new = n0 * (1 + g) ** min(t, stop)
        base = base * (1 - c) + new
        cash += base * GP - new * CAC - F
        if cash < trough:
            trough, t_trough = cash, t + 1
    return trough, t_trough, cash

if __name__ == "__main__":
    d = 0.01
    print(
        "   case  slogan   discounted  payback  funding     trough  cash at"
    )
    print(
        "         LTV/CAC  LTV/CAC              needed      month   month 96"
    )
    for name, CAC, c in (
        ("A", 1200.0, 0.025),
        ("B", 2400.0, 0.0125),
        ("C", 3600.0, 0.0056),
    ):
        ltv = 100.0 / c
        dltv = 100.0 * (1 + d) / (c + d)
        tr, tt, end = run(0.06, CAC, 100.0, c)
        print(
            "   {}     {:6.2f}   {:9.2f}  {:>4.0f} mo  {:10,.0f}  {:>5} "
            " {:11,.0f}".format(
                name, ltv / CAC, dltv / CAC, CAC / 100.0, -tr, tt, end
            )
        )
    print(
        "\nReading: the ratio hides the timing. B has the same slogan ratio"
        " as A but"
    )
    print(
        "needs far more money, and C's better-looking ratio needs the most of"
        " all and"
    )
    print(
        "is still under water at month 96. Once the wait is discounted, C's"
        " ratio is"
    )
    print("the lowest of the three.")
#    case  slogan   discounted  payback  funding     trough  cash at
#          LTV/CAC  LTV/CAC              needed      month   month 96
#    A       3.33        2.40    12 mo   2,210,878     38   28,707,216
#    B       3.33        1.87    24 mo  12,114,170     51   11,693,152
#    C       4.96        1.80    36 mo  27,153,270     62   -9,822,399
#
# Reading: the ratio hides the timing. B has the same slogan ratio as A but
# needs far more money, and C's better-looking ratio needs the most of all and
# is still under water at month 96. Once the wait is discounted, C's ratio is
# the lowest of the three.

A 与 B 的口号比率相同,都是 3.33。B 的客户贵一倍、留存时间也长一倍;它需要 12,114,170 的资金,而 A 只需 2,210,878——五倍多——并且在第 51 个月而不是第 38 个月到达低谷。C 报出的比率最好,为 4.96,却需要最多的钱,27,153,270,到第 96 个月仍有 9,822,399 的缺口。

以每月 1% 折现,排名与口号所暗示的相反:A 的折现比率为 2.40,B 为 1.87,C 为 1.80,是三者中最低的。一个忽略时间的比率,奖励的是回本慢的客户——恰恰是一家增长中的企业最负担不起的客户。

注

用什么代替单一比率。 对于一个增长计划,应报告回收期、在给定折现率下的折现 LTV / CAC,以及按计划增长路径计算的现金低谷及其所在月份。低谷才是决定这个计划能否融资的数字。

假说 · 尚未验证

作者的假说(尚未验证): 在收入仍在增长时失败的风投支持订阅制公司中,最后一次报告的获客成本回收期中位数长于可比的存活公司,而口号式的 LTV / CAC 在两组之间没有差别。什么结果会证明它错:回收期没有差别,或者 LTV / CAC 的差别同样能解释失败。

小结

LTV / CAC 衡量的是客户带回多少,而不是何时带回。因为获客成本先付,一家增长中的企业背负着一个随增长加深的现金低谷——在单位经济学完全相同的情况下,从不增长时的 912,868 到月增长 10% 时的 8,320,737——而且它到来的时间晚于资金规划的时间。口号比率相同的两家企业,所需资金可以相差五倍;比率看上去最好的企业,可能需要最多的钱、折现比率却最低。回收期、折现 LTV / CAC,以及按计划增长路径计算的低谷,应当出现在每一份计划里。

Fan, Y. (2026). Subscription Unit Economics, Net Revenue Retention and AI Inference Scaling Laws: Cohort Survival Mixtures, Negative-Churn Differential Equations, Token Compute Margins and Metcalfe Congestion Limits [订阅制单位经济学、净收入留存与人工智能推理标度律] (Version 1.0.0). 5D Library. https://doi.org/10.5281/zenodo.24090164 · Licensed under Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International (CC BY-NC-ND 4.0).