5D Library › 企业瓶颈诊断、产品线自蚕食审计与自动化改造回报边界 · English

第五章 固定底座费加增量利润分成:双边道德风险下的契约

为什么要分成,又为什么不能全分

外部架构顾问为企业诊断瓶颈、重构产品线,常见的收费方式是「固定底座费 + 增量利润分成」。分成让顾问有动力做出真实的改善;但改善能否落地,还取决于企业自己的执行投入。双方都可能偷懒,而结果被噪声遮住,谁也无法证明对方偷懒:这是双边道德风险 (Bhattacharyya and Lafontaine 1995)。

设每月的增量毛利(千令吉)为 G=aec+bef+εG = a e_c + b e_f + \varepsilon,ece_c 是顾问的投入,efe_f 是企业自己的执行投入,两者的成本都是 e2/2e^2/2;噪声 ε\varepsilon 的标准差为 σ\sigma。顾问风险厌恶(绝对风险厌恶系数 rr),报酬是 F+βGF + \beta G。顾问会选择 ec=βae_c = \beta a,企业选择 ef=(1−β)be_f = (1-\beta) b。在线性契约与正态噪声下 (Holmström and Milgrom 1987),使双方确定性等价总和最大的分成比例是

β*=a2a2+b2+rσ2.\beta^* = \frac{a^2}{a^2 + b^2 + r\sigma^2}.

分成比例要在三件事之间取舍:顾问投入的效果 aa,企业投入的效果 bb,以及噪声给顾问带来的风险 rσ2r\sigma^2。

推导

# Chapter 5. p09: Fixed base fee plus a share beta of the measured gain,
# with moral hazard on BOTH sides (Bhattacharyya-Lafontaine). Monthly
# gain G = a*e_c + b*e_f + noise (RM thousand), noise sd sigma; the
# consultant's effort e_c and the client's own implementation effort e_f
# each cost e^2/2; the consultant is risk averse (CARA, r per RM 1k).
# The consultant picks e_c = beta*a, the client e_f = (1 - beta)*b, and
# the joint certainty-equivalent surplus is maximised at
#   beta* = a^2 / (a^2 + b^2 + r*sigma^2).
# The base fee F then gives the consultant a certainty equivalent of 25.
A, B, R, CE = 6.0, 10.0, 0.02, 25.0

def surplus(beta, sigma):
    ec, ef = beta * A, (1 - beta) * B
    return A * ec + B * ef - ec ** 2 / 2 - ef ** 2 / 2 \
        - R * beta ** 2 * sigma ** 2 / 2

def base_fee(beta, sigma):
    ec, ef = beta * A, (1 - beta) * B
    gain = A * ec + B * ef
    return CE - beta * gain + ec ** 2 / 2 + R * beta ** 2 * sigma ** 2 / 2

def best(sigma):
    return A ** 2 / (A ** 2 + B ** 2 + R * sigma ** 2)

if __name__ == "__main__":
    print("   noise sd  beta*   surplus at: beta*  0.20   0.50   fee only")
    for sigma in (10.0, 40.0, 80.0):
        b = best(sigma)
        print("   {:7.0f}  {:6.3f}  {:16.1f}  {:5.1f}  {:5.1f}  {:7.1f}"
              .format(sigma, b, surplus(b, sigma), surplus(0.2, sigma),
            surplus(0.5, sigma), surplus(0.0, sigma)))
    s = 40.0
    b = best(s)
    print("   at sigma 40: base fee RM {:.1f}k + {:.1%} of gain;".format(
        base_fee(b, s), b))
    print("   expected gain RM {:.1f}k/month, client keeps RM {:.1f}k".format(
        A * b * A + B * (1 - b) * B,
        A * b * A + B * (1 - b) * B - base_fee(b, s)
        - b * (A * b * A + B * (1 - b) * B) - ((1 - b) * B) ** 2 / 2))
#    noise sd  beta*   surplus at: beta*  0.20   0.50   fee only
#         10   0.261              54.7   54.4   50.8     50.0
#         40   0.214              53.9   53.8   47.0     50.0
#         80   0.136              52.5   51.9   35.0     50.0
#    at sigma 40: base fee RM 8.1k + 21.4% of gain;
#    expected gain RM 86.3k/month, client keeps RM 28.9k

在示意参数下(顾问投入效果 6,企业投入效果 10,风险厌恶 0.02),噪声标准差为 40 时,最优分成是 0.214,总盈余 53.9;分成 0.50 时总盈余降到 47.0,只收固定费时是 50.0。噪声标准差为 80 时,最优分成降到 0.136;噪声为 10 时升到 0.261。在噪声为 40 时,底座费约 RM 8.1k,再加增量毛利的 21.4%;预期增量毛利每月 RM 86.3k,企业扣除顾问费与自身投入之后每月留下 RM 28.9k。

双方总盈余随分成比例的变化,三种噪声水平;圆点为最优分成(p09)。

注

分成过高会伤害企业自己的投入。 分成比例 0.50 的总盈余低于只收固定费,因为企业知道一半的改善要交出去,自己的执行投入就会减少。在企业执行更关键的项目中,顾问分成应当偏低。

隐瞒增量:审计概率与罚则

分成合约的另一个风险是企业少报增量:每少报 dd,就少付 βd\beta d。如果审计的概率是 qq,查出少报时企业要补付并另付 kk 倍罚金,那么只有 q<1/(1+k)q < 1/(1+k) 时少报才划算 (Becker 1968)。核实需要成本,所以审计应当按规则触发,而不是每月都做 (Townsend 1979)。

推导

# Chapter 5. p10: Keeping the client honest about the measured gain.
# A client that under-reports the gain by d saves beta*d of fees. If an
# audit happens with probability q and a found shortfall costs the fee
# owed plus a penalty k times it, hiding pays only when q < 1 / (1 + k).
# Trigger rule: audit when the reported gain falls more than z noise sd
# below the agreed forecast. Monthly gain sd is RM 40k; hiding 30% of an
# expected RM 86.3k gain is RM 25.9k. Summing 12 months shrinks the noise
# by sqrt(12) while a steady shortfall grows twelvefold.
from math import erf, sqrt

def phi(x):
    return 0.5 * (1 + erf(x / sqrt(2)))

SD, HIDE = 40.0, 0.30 * 86.3

if __name__ == "__main__":
    print("   penalty k   audit probability needed 1/(1+k)")
    for k in (0.0, 1.0, 2.0, 3.0):
        print("   {:9.0f}   {:.3f}".format(k, 1 / (1 + k)))
    print("   trigger z  honest audited  cheat audited (monthly)"
          "  (annual)")
    for z in (0.5, 1.0, 1.5, 2.0):
        fa = phi(-z)
        dm = phi(-z + HIDE / SD)
        da = phi(-z + 12 * HIDE / (SD * sqrt(12)))
        print("   {:9.1f}  {:14.3f}  {:13.3f}  {:16.3f}".format(
            z, fa, dm, da))
    print("   hidden shortfall = {:.2f} monthly sd, {:.2f} annual sd".format(
        HIDE / SD, 12 * HIDE / (SD * sqrt(12))))
#    penalty k   audit probability needed 1/(1+k)
#            0   1.000
#            1   0.500
#            2   0.333
#            3   0.250
#    trigger z  honest audited  cheat audited (monthly)  (annual)
#          0.5           0.309          0.559             0.959
#          1.0           0.159          0.362             0.893
#          1.5           0.067          0.197             0.771
#          2.0           0.023          0.088             0.596
#    hidden shortfall = 0.65 monthly sd, 2.24 annual sd

罚金为一倍时,所需审计概率是 0.500;三倍时是 0.250。若规则是「报告的增量低于约定预测超过 zz 个标准差就审计」:每月增量的噪声标准差是 RM 40k,隐瞒三成增量只相当于 0.65 个月度标准差。z=1.0z = 1.0 时,如实报告的企业每月被审计的概率是 0.159,而隐瞒者只有 0.362,不足以威慑。改为按全年合计核对之后,隐瞒相当于 2.24 个年度标准差,同一规则下隐瞒者被审计的概率升到 0.893。

假说 · 尚未验证

作者的假说(尚未验证): 在固定底座费加分成的顾问合约中,写明「按年度合计核对、触发审计、少报补付并罚一倍」的合约,比只写「按月报告」的合约,企业报告的增量与独立核算的增量之差更小。什么会证明它错: 在多家客户的合约对比中,两种条款下报告与核算之差没有差别。

小结

双边道德风险下,最优分成比例是 a2/(a2+b2+rσ2)a^2/(a^2+b^2+r\sigma^2):企业的执行越关键、结果噪声越大,分成越低。在示意参数下约为两成,加一笔小的底座费。隐瞒增量要靠审计与罚则威慑;按月逐笔核对几乎抓不到小幅少报,按年合计核对才有效。

参考文献

Becker, Gary S. 1968. “Crime and Punishment: An Economic Approach.” Journal of Political Economy 76 (2): 169–217.
Bhattacharyya, Sugato, and Francine Lafontaine. 1995. “Double-Sided Moral Hazard and the Nature of Share Contracts.” The RAND Journal of Economics 26 (4): 761–81.
Holmström, Bengt, and Paul Milgrom. 1987. “Aggregation and Linearity in the Provision of Intertemporal Incentives.” Econometrica 55 (2): 303–28.
Townsend, Robert M. 1979. “Optimal Contracts and Competitive Markets with Costly State Verification.” Journal of Economic Theory 21 (2): 265–93.

Fan, Y. (2026). Enterprise Bottleneck Diagnostics, SKU Cannibalization Audits and Automation Return Bounds: Theory-of-Constraints Queueing Networks, Margin Dilution Matrices and Value-Based Advisory Pricing [企业瓶颈诊断、产品线自蚕食审计与自动化改造回报边界] (Version 1.0.0). 5D Library. https://doi.org/10.5281/zenodo.24090217 · Licensed under Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International (CC BY-NC-ND 4.0).