Contratos Homogéneos para Agentes Heterogéneos - Alineación de La Composición y Compensación de La Fuerza de Ventas

Homogeneous Contracts for Heterogeneous Agents: Aligning Sales Force Composition
and Compensation
Author(s): ØYSTEIN DALJORD, SANJOG MISRA and HARIKESH S. NAIR
Source: Journal of Marketing Research , APRIL 2016, Vol. 53, No. 2 (APRIL 2016), pp.
161-182
Published by: Sage Publications, Inc. on behalf of American Marketing Association
Stable URL: http://www.jstor.com/stable/43832424
JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide
range of content in a trusted digital archive. We use information technology and tools to increase productivity and
facilitate new forms of scholarship. For more information about JSTOR, please contact support@jstor.org.
Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at
https://about.jstor.org/terms
Sage Publications, Inc. and American Marketing Association are collaborating with JSTOR to
digitize, preserve and extend access to Journal of Marketing Research
This content downloaded from

ff:ffff:ffff:ffff:ffff:ffff on Thu, 01 Jan 1976 12:34:56 UTC
All use subject to https://about.jstor.org/terms
0YSTEIN DALJORD, SANJOG MISRA, and HARIKESH S. NAIR*
Observed contracts in the real world are often veiy simple, which partly reflects
the constraints faced by contracting firms in making the contracts more complex.
In this article, the authors focus on one such rigidity: the constraints faced by firms
in fine-tuning contracts to the full distribution of heterogeneity of their employees.
The authors explore the implication of these constraints for the provision of
incentives within the firm. The study's application is to sales force compensation,
wherein a firm maintains a sales force to market its products. Consistent with
ubiquitous real-world business practice, the study assumes that a firm is restricted
to fully or partially set uniform commissions across its agent pool. The authors
show that this restriction implies an interaction between the composition of agent
types in the contract and the compensation policy used to motivate them, leading
to a "contractual externality" in the firm and generating gains to sorting. This article
explains how this contractual externality arises; discusses a practical approach to
endogenizing agents and incentives at a firm in its presence; and presents an
empirical application to sales force compensation contracts at a U.S. Fortune 500
company that explores these considerations and assesses the gains from a
sales force architecture that sorts agents into divisions to balance firmwide
incentives. Empirically, the authors find that the restriction to homogeneous
plans significantly reduces a firm's payoff, relative to a fully heterogeneous
plan, when the firm is unable to optimize the composition of its agents.
However, a firm's payoff under a homogeneous plan comes very close to that
under a fully heterogeneous plan when the firm can optimize both composition
and compensation. Thus, in the empirical setting of this study, the ability to choose
agents mitigates the loss in incentives from the restriction to uniform contracts.
The authors conjecture this result may hold more broadly.
Keywords: sales force compensation, incentives, hiring, agency theory,

dynamics
Homogeneous Contracts for Heterogeneous

Agents: Aligning Sales Force Composition
and Compensation
In many interesting market contexts, firms face rigidities or prevented by regulation from conditioning their premiums on
constraints in fine-tuning their contracts to reflect the full consumer characteristics like race and credit score. Royalty
distribution of heterogeneity of the agents they are contracting rates in business-format franchising in the United States
with. For example, automobile insurance companies are often are typically constrained by norm to be the same across all
*0ystein Daljord is Assistant Professor of Marketing, Booth School of Business, University of Chicago (e-mail: daljord@chicagobooth.edu). Sanjog Misra is Charles
H. Kellstadt Professor of Marketing and Neubauer Family Faculty Fellow, Booth School of Business, University of Chicago (e-mail: sanjog.misra@chicagobooth.edu).
Haiikesh S. Nair is Professor of Marketing, Graduate School of Business, Stanford University (e-mail: haiikesh.nair@stanford.edu). The authors thank Guy Arie, Nick
Bloom, Francine Lafontaine, Sridhar Moorthy, Paul Oyer, Michael Raith, Kathiyn Shaw, Chuck Weinberg, and Jeff Zwiebel; their discussant at the Quantitative
Maiketing and Economics (QME) conference, Curtis Taylor, and Lanier Benkard and Ed Lazear in particular, for their useful comments and suggestions. The authors also
thank seminar participants at Arizona Economics, Harvard Economics, Michigan Economics, MIT Sloan, Stanford Graduate School of Business, UBC Sauder, UC Davis
Graduate School of Management, UT Austin McCoombs, UToronto Rotman, and the 2012 IOFest, Marketing Dynamics, World Congress of Econometrics, and QME
conferences for useful feedback. Jean-Pierre Dubé served as associate editor for this article.
© 2016, American Marketing Association Journal of Marketing Research

ISSN: 0022-2437 (print) Vol. Lin (April 2016), 161-182
1547-7193 (electronic) 161 DOI: 10.1509/jmr.l4.0018

200.3.152.96 on Fri, 19 Aug 2022 03:56:44 UTC
162 JOURNAL OF MARKETING RESEARCH, APRIL 2016
Our second point

franchisees in a given chain.1 Wholesale is that uniformity
contracts implies that the
between
manufacturers and downstream retailers
presence of ain
salesthe
agent United States
in the firm imposes an externality
on the
typically involve similar wholesale other agents
prices to inall
the pool through its effect on the
downstream
shape ofarea
retailers within a given geographic the common
dueelement of the incentive contract. For
to Robinson-
Patman considerations. In sales forces, thea firm's
example, suppose context of
agents are the
homogeneous in all
respects except or
empirical example in this article, incentive theircommission
risk aversion, and rates
there are three
on output are invariably set the agents
same A, B, and C whoall
across couldsales
be employed, with C the most
agents
risk averse.
within a firm. For example, a firm Agent C needs
choosing a more insurance (i.e., a higher
salary-plus-
proportionscheme
commission sales force compensation of income through salary) than
typically A and B, and
sets
retaining
the same commission rate for every salesC requires
agent a lower
on common commission rate. It
its payroll,
in spite of the fact that exploring could
the then be thatforce
sales A and B are worse off with C in the firm
heterogeneity
(i.e., they receivecould
and setting an agent-specific commission lower commissions). Thus, the presence
theoretically
of the highly risk-averse
improve the incentives at the individual agent imposes an externality
level. Although rea- on the
other sales agents
sons are varied, full or partial uniformity of in incentives
the firm through isthe well
endogeneity of
documented as a ubiquitous feature contractof choice.
real-world sales force
compensation (Lo, Ghosh, and LafontaineThis externality2011; can be substantial
Mantrala, when agent types are
Sinha, and Zoltners 1994; Rajumultidimensional
and Srinivasan 1996;
(e.g., when agents Rao
are heterogeneous in
1990; Zoltners, Sinha, and Zoltners 2001).
risk aversion, productivity, and costs of expending sales
The focus of this article is on effort).
the implications
Consider the example in forthe previous
the paragraph. It
principal that follow from this may be restriction to to
optimal for the principal uniform
rank the three agents on
contract terms across agents. the We basisdo not
of their risk take a to
aversion and strong
drop the "low-type" C
from the agent
stance on the source of the uniformity pool. instead
but Thus, if risk focus
aversion were
on the only
the fact that any such uniformity source inof a heterogeneity,
contract andimplies
we enlarged that
the contracting
agents and contract terms mustproblem be chosento allow thejointly. In the
principal to choose both the optimal
sales force context, for example, this creates
composition contractual
and compensation, it might be that "low types"
externalities in the firm, that is, like C interactions between
would impose few externalities the
because they would
composition of agent types in the contract
be endogenously droppedand fromthe com-
the firm. However, consider
pensation policy used to motivate them,
what happens andare
when types generates
multidimensional. Suppose
gains to sorting. This article explains how
that in addition this
to risk contractual
aversion, agents are heterogeneous in
externality arises; discusses a practical approach
their productivity to
(in the sense of endo-
converting effort into
genizing agents and incentives atoutput)
a firm and that
inC, the
itsmost risk-averse, isand
presence; also the most
presents an empirical applicationproductive.
to sales Then,force compensation
the principal faces a trade-off: dropping C
from the poolthat
contracts at a U.S. Fortune 500 company enablesexplores
the principal these
to set more high-
considerations and assesses the gains
powered from
incentives to Aaand
sales force
B but also entails a large
architecture that sorts agents into divisions
loss in output becauseto
C is balance firm-
the most productive. In this trade-
wide incentives. off, the principal's optimal strategy could be to retain C in
As a motivating example, consider a firm that has the agent pool and to offer all agents the lower common
chosen a salary-plus-commission scheme. Suppose all commission induced by C' s presence. Thus, multidimen-
agents have the same productivity, but there is heteroge- sional types increase the chance that the externalities we
neity in risk aversion among the agents. The risk-averse have discussed persist in the optimally chosen contract.
agents prefer that more of their pay arises from fixed salary, More generally, multidimensionality of the type space also
but the less risk-averse prefer more commissions. When points to the need for a theory to describe who should be
commissions are restricted to be uniform across agents, retained and who should be let go from the sales force,
including the more risk-averse types in the firm, it implies because agents cannot be ranked as desirable or undesirable
that the firm cannot offer high commissions. Dropping the on the basis of any single variable.
most risk-averse agents from the firm might then enable the Our main question explores the codependence between
firm to profitably raise commissions for the rest of agents. composition and compensation. We ask to what extent
Knowing that, the firm should choose agents and com- composition and compensation complement each other in
missions jointly. This is our first point: the restriction to realistic sales force settings. We use an agency-theoretic
uniformity implies that sales force composition and com- setup in which the principal chooses both the set of agents
pensation are codependent. To address this issue, the to retain in the firm and the optimal contract with which to
contracting problem must be enlarged to allow the principal incentivize the retained agents. We use our model to
to choose the distribution of types in its firm along with the simulate how the contract form changes when the distri-
optimal contract form given that type distribution. bution of ability changes, which helps measure the size of
the externalities and assess the value of policies to mitigate
Quoting Lafontaine and Blair (2009, pp. 395-96), "Economic theory
them. A realistic assessment of these issues is dependent on
suggests that franchisors should tailor their franchise contract terms for each the distribution of heterogeneity in the agent pool and thus
unit and franchisee in a chain. In practice, however, contracts are remarkably is inherently an empirical question.
uniform across franchisees at a point in time within chains [emphasis ours]... We leverage access to a rich data set that contains the
a business-format franchisor most often uses a single business-format
franchising contract - a single royalty rate and franchise fee combination -
joint distribution of output and contracts for all sales agents
for all of its firanchised operations that join the chain at a given point.... Thus, at a Fortune 500 contact lens company in the United States.
uniformity, especially for monetary terms, is the norm." We build on our analysis developed in Misra and Nair

200.3.152.96 on Fri, 19 Aug 2022 03:56:44 UTC
Aligning Sales Force Composition and Compensation 163
(2011), which uses these data to identify primitive agent this could be broadly relevant in other settings and could
parameters (cost of effort, risk aversion, productivity) and help rationalize the prevalence of homogeneous contracts
to estimate the multidimensional distribution of heter- in many sales force settings in spite of the potential profit
ogeneity in these parameters across agents at the firm. The loss that results from reduced incentives.
data show that agents are paid according to a nonlinear, We then simulate a variety of sales force architectures
quarterly incentive plan consisting of a salary plus a linear
in which the firm sorts its sales agents into divisions. We
commission that is earned if realized sales are above a require each division to offer a uniform commission to all
contracted quota but below a prespecified ceiling. The agents within its purview but allow commissions to vary
nonlinearity of the incentive contract creates dynamicsacrossin divisions. We then simultaneously solve for the
optimal commissions and the optimal allocation of agents
agents' actions, causing agents to optimally vary their effort
profiles as they move closer to or exceed their quotas. Theto divisions. In the context of our empirical example, we
find that a small number of divisions generates profits to
joint distribution of output and the distance to the quota thus
identify "hidden" effort in this moral hazard setting. theInprincipal that come very close to those under fully
Misra and Nair (2011), this identification strategy is heterogeneous
in- contracts. If the firm is allowed to choose
corporated into a structural model of agents' optimization its composition as well, this profit gain can be achieved
with even fewer divisions. The main takeaway is that
behavior to recover the primitives underpinning agent
simple contracts, combined with the ability to choose
types. Here, we use these estimates as an input into a model
of simultaneous contract form and agent composition agents, seem to do remarkably well, compared with more
choice for the principal. complex contracts, at least in the context of our empirical
example.
Solving this model involves computing a large-scale
combinatorial optimization problem, in which the firm
Our analysis is related to a literature that emphasizes
the "selection" effect of incentives, for example, Lazear's
chooses one of 2N possible sales force configurations from
among a pool of N potential agents, and solving for(2000a)
the famous analysis of Safelite Glass Corporation's
incentive
optimal common incentive contract for the chosen pool. To plan for windshield installers, in which he
find the optimal composition of agents, we can alwaysdemonstrates that higher-ability agents remained with the
enumerate the profits for all 2N combinations when N company
is after it switched from offering a straight salary
small. If N is large, simple enumeration algorithmtoisa piece rate; or the analysis of Bandiera et al. (2007),
practically infeasible (N is approximately 60 for our focal
who study managers at a fruit-picking company that started
firm but could number in the hundreds or thousands in
hiring more high-ability workers after it switched to a
contract under which workers' pay depended on their
other applications). Because agents are allowed a multi-
performance.
dimensional type space, it is generally not possible to find a Lazear and Bandiera et al. present models of
simple cut-off rule whereby agents above some parameter how the way various types of agents are sorted or retained
threshold are retained. Exploiting a characterization of atthea firm changes in response to an exogenously specified
optimal solution for a class of composition-compensation piece rate. In our setup, the piece rate itself changes as the
problems, we derive an algorithm that allows us to search set of agents at the firm changes, because the firm jointly
chooses the contract and the agents. The endogenous
the exploding composition-compensation 'pace by re-
adjustment of the contract as the types change is key to our
ducing it to a standard optimization program for continuous
functions on compact sets. The reduction makes the story.
ex- The closest model to ours that we know of in the
ecution time of the algorithm independent of the compo- literature is that of Lazear (2000b), who shows that firms
may choose incentives to attract agents of high ability.
sition space itself. In examples we compute, we can search
Unlike in our setup, Lazear considers unidimensional
a space of 25,000 composition-compensation pairs in fractions
of a second. A power-set search of a space of that size is in an environment with no asymmetric information
agents
otherwise prohibitive. or uncertainty, so many of the contractual forces identified
We use the algorithm to simulate counterfactual con- in this study are not a feature of his analysis. Lazear also
tracts and agent pools at the estimated parameters. makes We a broader point that piece rates have an advantage of
helping manage heterogeneity within a firm. Our analysis
explore to what extent a change in composition of the agent
pool affects the nature of optimal compensation for thosehere has parallels to this insight.
agents, and we quantify the profit impact of jointly opti- The literature on relative performance schemes and on
mizing over composition and compensation. We find that teams (Hamilton, Nickerson, and Owen 2003; Holmstrom
allowing the firm to optimize the composition of its agent1982; Kandel and Lazear 1992; Misra, Pinker, and Shumsky
types has bite in our empirical setting. When the firm 2004)
is has identified other contexts wherein one agent's
characteristics or actions substantively affect another
restricted to homogeneous contracts and no optimization
agent's welfare through an interaction with incentives. The
over types, we estimate its payoff is significantly lower than
that under fully heterogeneous contracts. However, contractual
the externality we identify persists when agents
payoff under homogeneous contracts when the firmhave can exclusive territories and there are no across-agent
optimize both composition and compensation comes very complementarity or substitution effects in output, and it
close to that under fully heterogeneous contracts. is We relevant even when contracts are absolute and not relative.
demonstrate these results are robust to parameter It is thus distinct from the mechanisms identified in the
un-
certainty that the principal may have about its estimatesliterature.
of A related literature on contract design in which
the agents' types. Overall, we find the ability to chooseone principal contracts with many agents focuses on the
agents significantly helps balance the loss in incentivesconditions under which relative incentive schemes arise
imposed by the restriction to homogeneity. We conjecture endogenously as optimal, and not on the question of the joint

200.3.152.96 on Fri, 19 Aug 2022 03:56:44 UTC
choice of agents and incentives, result

which in endogeneity
is our biases in attempts
focus to measure the causal
here.
effect of contracts
Broadly speaking, the relative incentive on outcomes.
scheme Our model, which provides
literature
focuses on the value of contractsa rationale for why agentout
in filtering and contract
common characteristics are
shocks to demand and output codetermined
and on the across firms, has similar implications
advantages of for
contracting on the ordinal aspect empirical work that uses
of outputs across-agent
when outputdata. The
ismodel im-
hard to measure (e.g., Green and plies that the variation
Stokey 1983; inKalra
contract terms
and across
Shi firms is en-
2001; Lazear and Rosen 1981; Lim, Ahearne, and Ham dogenous to worker characteristics at those firms. Whereas
2009; Mookherjee 1984). Common shocks and noise in the Ackerberg and Botticini (2002) stress the omitted-variables
output measure are not compelling features of our empirical problem, the endogeneity implied by our model derives from
setting, which involves sales of contact lenses to optom- the simultaneity of contracts and agents.
etrists, in which seasonality and comovement in demand is A second stream pertains to work that has measured
limited and sales (output) are precisely tracked. A small complementarities in human resources (HR) practices
theoretical literature also emphasizes why a principal might within firms, testing the theory developed in Milgrom and
choose a particular type of agent in order to signal com- Roberts (1990) and Holmstrom and Milgrom (1994),
mitment to a given policy (e.g., shareholders might choose a among others. This theory postulates that HR activities
"visionary" CEO with a reputation for change management (e.g., worker training) and incentive provision are com-
to commit to implementing change within the firm; e.g., plementary activities. A large body of empirical work has
Rotemberg and Saloner 2000). Our point that a principal measured the extent of these complementarities using
might choose agents for incentive reasons is distinct from correlations of these activities and worker productivity in
the focus in this literature, which emphasizes commitment across-firm data (e.g., Ichniowski, Shaw, and Prennushi
1997). Our model predicts that workers and incentives (or
as the principal's rationale for its choice of agents. A related
theoretical literature stream has also noted that contracts
HR practices more generally) are optimally jointly chosen.
may signal information that affects the set of potentialWhen better workers also have corresponding better
employees or franchisees a principal may contract withproductivity, the simultaneity of worker choice and HR
(Desai and Srinivasan 1995; Godes and Mayzlin 2012),practices implies that the incidences of HR practices are
without focusing explicitly on the principal's choice endogenous
of in productivity regressions, which confounds the
agents. measurement of such complementarities using across-firm
Findings related to our results here - that it may be data. A related implication of our model is that endogenously
optimal for the principal to drop some agents and to group adjusting common clauses of firmwide contracts can generate
together agents of differing types into divisions in order to across-agent dependencies in output that generate indirect
achieve appropriate separation - are also reflected in a complementarities. If these are not accounted for, they may
small theoretical literature on multidimensional screening, confound measurement of other sources of direct com-
canonical examples of which are discussed in the context of plementarities, such as peer effects, that researchers are
nonlinear pricing in Armstrong (1996) and Rochet and interested in measuring using within-firm personnel data.
Choné (1998). For example, Armstrong (1996) shows that More research and better data are required to address these
the optimal price schedule for a multiproduct firm that faces kinds of difficult econometric concerns in empirical work.
consumers with (unknown) multidimensional types may We now discuss our model setup and present the rest of the
involve excluding some consumers from its products in analysis.
order to extract more revenue from the high-value con-
sumers. Rochet and Choné (1998) show that such optimal
MODEL
contracts typically involve some degree of "bunching," so
that consumers of different types choose the same bundle of A firm wishes to optimize the composition and com-
products. Although there are these parallels, note that this pensation of its sales force. The firm is assumed to know all
literature focuses on adverse selection as the manifestation the agents' relevant characteristics with certainty, but it is
of asymmetric information. In contrast, our focus is on thenot able to observe their efforts with certainty. Conditional
moral hazard problem, which has a more complicatedon the group of agents, the problem is similar to the classic
structure because the unobservable "type" of the agent (i.e.,hidden-action problem of Holmstrom (1979). Principal
hidden effort) changes with the contract. certainty about the agents' characteristics may be a rea-
Our model predicts that agents and incentives across firmssonable assumption for the retention problem, in which the
are simultaneously determined and has implications for twofirm has known the agents for a long time (as in our ap-
related streams of empirical work. One stream measures theplication), but it is far more questionable as a point of
effect of incentives on workers and tests implications of departure for hiring. We therefore restrict attention here to
contract theory using data on observed contracts and agent the retention problem, that is, whom the firm should retain
characteristics across firms (for a review, see Prendergast when there are contractual externalities that depend on the
1999). In an important contribution to the econometrics insales force composition. To simplify the application, we
this area, Ackerberg and Botticini (2002) note that whenabstract away from uncertainty about the agents' types to
agents are endogenously matched with contracts, the cor- avoid issues of learning and adverse selection. Learning
relation observed in data between outcomes and contract about agent type is not of first-order importance in our
characteristics should be interpreted with caution. A potentialapplication because most agents in our data have been with
for confounds arises from unobserved agent characteristics the firm for a long time (mean tenure 9 years). However, this
that may potentially be correlated with both outcomes andmay be an important dynamic for new workers. We discuss
contract forms. The resulting omitted- variables problem may these issues in more detail later in the article.

200.3.152.96 on Fri, 19 Aug 2022 03:56:44 UTC
Reflecting the empirical application, the firm has divided any marketing or sales force effort. Given this, an equiv-
its potential market into N geographic territories, and the alent interpretation of the principal's decision to offer a
maximum demand at the firm is for N sales agents.2 There contract that induces an agent to quit is that the principal
are N heterogeneous sales agents indexed by i = 1, N has decided to vacate the territory managed by the agent.
currently employed or employable. Let Mn denote the Obtaining only the base level of sales from the territory but
power set spanned by N (i.e., all possible sub-sales forces offering an improved contract to the other agents is more
that could be generated by N) and the set of com- beneficial to the firm than retaining the agent and incurring
pensation contracts possible for a specific sub-sales force the added pay and contractual externalities induced by his
M. Let Si denote agent i's output, W(Si) his or her wages or her presence.
conditional on output, and .ř*(Si|ei) the cumulative dis- We can reformulate the problem in an equivalent bi-level
tribution function (CDF) of output conditional on effort setup by allowing the principal to choose the optimal
choice e¡. Effort e¡ is privately observed by the agent and contract in a first step and then solving pointwise for the
not by the principal, whereas output Si is observed by both optimal configuration of the chosen contract. The program
the agent and the principal and thus is contractible. As is described above is equivalent to the case in which the
common in the agency literature, we assume that the agent principal maximizes:
chooses effort before sales are realized and that the agent
and the principal share the same beliefs about the condi- n= W€W"
W€W" Jmax
ieMwJ f X [Si-VKSOldFÍSile,),
tional distribution of output (^(Siļei)) (common knowl-
with
edge about outcomes). Because sales are stochastic,
the principal cannot back out the hidden effort from re-
alized output, which generates the standard moral hazard Mw = argmax f £ ßi - W(Si)]d*-(Si|e¡),
M € Mn J i € M
problem.
The principal maximizes as follows: subject to the IC and IR constraints as before. Contractual
externalities arise because some elements of the contract
W(Si) are common across agents, which makes the problem
(1) M €Mn, WeWe
M €Mn, n=JJi€A1
Í £ [Si-mSOWSileO, nonseparable across agents. Because Mw G Mn is point-
I
where the control, (M, W), is the set of active agents and wise the optimal sub-sales force plan for each considered
their compensations. The maximization is subject to the contract W, the solution to this revised problem returns the
incentive compatibility (IC) constraints that the effort chosen solution to the original program. Representing the program
by each agent i is optimal, this way helps clarify our numerical solution algorithm.
APPLICATION SETTING
(2) ei = arg max ju[W(Si),C(e;jLii)] d^*(Si|e) Vi eM, To illustrate the main forces at work clearly and to oper-
ationalize the setup described in the previous section for our
and the individual rationality (IR) constraints that eachQactive
agent i receives at least expected reservation utility Uj from
empirical setting, we now discuss the parametric assump-
staying with the firm and working under the suggested tions we impose. We employ a version of the well-known
contract, Holmstrom and Milgrom (1987) model for two reasons: First,
the model has a closed-form solution. That is useful from both
an illustrational and a computational point of view. Second, the
(3) Ju[W(Si))C(e;ni)]dJF(Si|ei) >Ü° Vi e M. optimal contracts are linear (salary plus commission), which is
empirically relevant.
This setup endogenizes the principal's choice of the agent
Although we use linear contracts as the illustrational
pool in the following way. The principal knows each
vehicle, the qualitative aspects of the setup hold more
agent's type (including reservation utility) and designs a
contract such that the IR constraints in Equation 3 are generally for any multilateral contracting problem with
satisfied only for the set of agents in M and violated for all
interagent externalities induced by common contractual
others. This contract provides the chosen set of agents in
components. Each agent i is described completely by a
M enough utility for them to stay; the rest are better off tuple {hi.ki.d^rijO^Uf}. The elements of the tuple will
become clear in what follows. We assume that sales are
pursuing an outside option. Thus, the contract endogenously
induces the preferred agents to stay and the others to quit.3
generated by the following functional:
To complete the model, we also need to specify what (4) Si =hi + kiei + Oi£i.
happens to demand from a territory managed by an agent if
that agent leaves. We assume that sales equivalent to the This functional has been used in the literatur
intercept in the output function (discussed in the next sec- and Srinivasan 1993) and interprets h as the
tion) continue to accrue to the firm even if no agent operates the absence of selling effort (i.e., E[Si|e¡ = 0]
in that territory. This encapsulates the notion that a base level marginal productivity of effort, and of as the un
of sales will be generated to the firm even in the absence of sales production process. As is usual, we assum
observes Si and knows the agent's type (i.e., {hi
agents. The density .F(Si|ei) is induced by th
2More generally, the need for a maximum of N agents can be thought of as
Under linear contracts, compensation is W
implying that total profit for the firm is concave in N.
3This need not be implemented by explicitly "firing" an agent. Not offering
where oq is the salary, which can be agent spec
raises as part of a restructuring exercise, or providing only reduced pay, could commission rate, which is common across ag
induce the outcome. is assumed to have a constant absolute risk a

200.3.152.96 on Fri, 19 Aug 2022 03:56:44 UTC
function Ui(Wi) = -exp{-riWi}, Equation

defined overthewealth
10 encapsulates Wļ,
effect of each agent type on
the contract:
which in turn is linear in output and the optimal
quadratic (ai, ß) dependsin
(convex) on the
the distribution
cost of effort, that is, Wļ = of
oc¡characteristics
+ ßSj -of|e?. the entire
The agent pool. Equation 10
agent
chooses effort to maximize expected utility,
demonstrates the contractual where the by ho-
externality implied
mogeneityto
expectation is taken over the shocks restrictions:
sales: when an agent joins or
E€[U¡(W¡)] = leaves the
sales force,
- Jexp[-ri(0Ci + ßSi - |ef)] d J7 fa). The he or she affects everyone
implied else by changing the
certainty
equivalent (CE) for the agent is optimal ß. Equation 10 also helps us build intuition about
how the distribution of types in the firms shifts the common
(5) C£i=a i + ß(hj + kid) - commission
^ ef -rate. With everything else held fixed, an increase
|ß2o?>
in the level of risk aversion in the sales force increases y, which
maximizing which implies that the optimal
reduces effort
the commission rate aschoice
expected. Asfornormalized
the agent is ei(ß) = ß|. productivity, k?/di, increases in the sales force, y falls, and
The Principal's Problem optimal commissions increase as expected. Both results are
intuitive. Assessing the net effect on commissions as both
The principal treats agents as exchangeable and cares
risk aversion and normalized productivities change is more
only about expected profits, which it maximizes subject to
difficult because it depends on the extent of affiliation
the IC and IR constraints to find the optimal agent-specific
between
salaries and common commission rate these parameters
(cci,ß), in the sales force (e.g., whether
N more risk-averse agents are more productive or less).
max E[n] = E (S¡ - ßSi - ai), such that To see how the profit function depends overall on agents'
(O..P) i=l types, we can write Equation 8 evaluated at the optimal
commission ß as
IC:ei(ß) = ß^ Vi = 1, ...,Nand
di N / N 2'
IR: C£' > U? Vi = 1, N. (11) E[n(ß)]=X(hi-U

In the preceding, Uf is the CE of the agent's outside-option
Noting that d¡ is the agent's cost
utility. The principal's problem can be simplified by in-
1 /di as a measure of the agent's e
corporating the IC constraint,
1/di expend the same effort at
(6) be interpreted as decomposing th
N N optimally chosen incentive level
E[n] =first X
comprises (E(S¡
the total baselin
i=l i=l
when each is employed but expen
Furthermore, The second is the if
optimalthe
commi
average of each agent's efficienc
(7) Oi(ß) correspond
= to each
U? agent's produc
-
the insurance component of the
and, substituting in
second part reflects incentives. E
Nr profits are separable across agent
(8) E[n]=X r (1 - ß)[hi + kiei(ß)] - Uf In the absence of endogenizing ß
i=i L
agent i in the pool has no bearin
another agent.
+ |ß[hi+kiei(ß)]-|ei(ß)2-|ß2o?| Substituting for the optimal ß
N
write the total payoff to the pri
incentives as
= S [hi + kiei(ß) - Uf - 1 ei(ß)2 - |ß2o? .
Differentiating with respect to ß, we get
N
(9) = X [kiej(ß)
ÖP i=l
- dei(ß)diei(ß) - r,ßaf] ,
where e[(ß) = = jļf. Solving Equation 9 gives the optimal

common commission,' Equation 12 shows that at the optimal ß , the payoff across
agents is no longer separable across types. Equation 12 defines
00) the firm's optimization problem over the N agent types given
optimal choice of incentives for each subconfiguration.
where
To build intuition, suppose the firm has the option of
retaining three agents who have low, medium, and high risk
aversion. If forced to retain all three on the payroll on a
yNü
^i=ldi
salary-plus-commission scheme, the firm cannot have a very
high-powered incentive scheme with a high commission
is an aggregate measure of the distribution of types within the rate because the highly risk-averse agent has to be provided
firm. The salary, a¡, can be obtained by substitution. significant insurance. Firing the highly risk-averse agent

200.3.152.96 on Fri, 19 Aug 2022 03:56:44 UTC
will allow the firm to optimally offer a higher commission heterogeneous contracts with individual commissions of
rate to the remaining two agents. Depending on the pro- 56% and 18%, they bring in a profit of .20. Under a uniform
ductivity and cost parameters of the three agents, we can commission of 45%, the profit is reduced to .02 because
construct examples in which the payoff to the firm with two Agent 1 gets underpowered incentives and Agent 2 gets
agents at the higher commission is greater than with three overpowered incentives.
agents at the lower commission. The firm makes lower sales with only Agents 2 and 3
than with all three agents (2.67 vs. 5.99), but compen-
A Simple Three-Agent Example sation payout is also lower, and the net effect is a higher
profit. By changing parameters, we can generate other
In a three-agent example, we show that the principal may
examples in which the low-commission agent can be
not find the most prolific sales agent attractive if that agent
dropped from the pool and give the remaining agents
sufficiently skews the incentives of the other agents. We set
high-powered incentives, which is the mirror image of this
the parameters to (h, k, a) = 1 for all agents, and we give the
setting. These examples illustrate the complication in-
remaining parameters in Table 1. Agent 1 has the lowest
duced by multidimensional types: the desirability of an
cost of effort d, Agent 2 has the lowest risk aversion, and
agent cannot be ordered on any one dimension by a simple
Agent 3 has the lowest outside utility U°. Because of the cut-off rule.
multidimensional type space, it is not immediately clear
Note that the profit under the optimal uniform com-
which agents are the most profitable.
mission plan (1 .33) comes close to that under the individual
The first three rows in Table 2 present the individually
optimal plan (1.49), despite the restriction to common
optimal contracts for all three agents. With no contractual
commissions and the heterogeneity in individually optimal
externalities, the firm will retain all agents and provide
commissions. The three-agent examples illustrate how
each agent a commission tailored to his or her type and
careful choice of sales force composition can compensate
set a salary that leaves each agent his or her reservation
for the reduced incentives implied by the common contract
utility.
terms, a point that also shows up in the "Empirical Ap-
In the fully heterogeneous plan summarized in the last
plication" section.
row of Table 2, there are no externalities. Agent 1 gets the
The simple three-agent example shows how sales force
largest commission (56%) and Agent 2 the smallest
composition and compensation interact in influencing
(18%), and the firm makes a profit of 1.49. Now, consider
firm profits. Although the example has only three agents,
what happens when the firm is restricted to a common
it provides a glimpse into the workings of this interaction
commission but individual salaries for each agent. The
and in particular shows that commonly used performance
results are given in the upper rows of Table 1 . Solving for
measures such as sales (or even profit contribution) might
each configuration, we find that the firm should optimally
not be useful in determining which agents should stay.
drop Agent 1 from the pool (expected payoff of 1 .33 with
Indeed, in the examples discussed in this section, Agent 1
an optimal common commission rate to Agents 2 and 3 of
had the highest productivity in terms of sales (3.78 vs.
25%). Including Agent 1 in the composition requires the 1.12 and 1.57 for Agents 2 and 3, respectively) and the
firm to set a higher commission rate (39%), which is too
second-highest profit contribution (.14 vs .06 and 1.29
high for the other agents, reducing the firm's payoff to
for Agents 2 and 3). However, keeping Agent 1 in the
1.24.
pool distorted the incentives to the others. We conjecture
Looking at the first two rows of the two agent composi-
that similar patterns apply more generally in real-world
tions, we see that Agent l's presence in the pool exerts skew sales forces. To examine this conjecture, we use data
the incentives of Agents 2 and 3 away from their individually
from a real sales force. Before doing so, however, we
optimal commissions. If Agent 1 were employed individu-
introduce our algorithm for computing the optimal sales
ally, he or she would be paid a high-powered commission rate
force composition.
of 56%, while Agents 2 and 3, if employed individually,
would be given commissions of only 18% and 29%, re- THE SOLUTION ALGORITHM
spectively. In this example, the firm is better off dropping the
The composition-compensation problem above is a
high-powered sales agent from the pool. Without Agent 1 , the
mixed-integer program: the commission is a continuous
firm can set an intermediate level of commission (25%) that
variable, and the composition choice is a binary-integer
provides Agents 2 and 3 better incentives. For a more dra-
program. The combinatorial nature of integer programs
matic effect, consider Agents 1 and 2 individually. Under
generally requires computationally demanding algo-
rithms. The examples in the previous section used a
complete enumeration of possible sales force composi-
Table 1
tions to examine the profitability of the possible com-
AGENT TYPES IN THE THREE-AGENT EXAMPLE
positions. Because the space of possible compositions
grows exponentially in the number of agents, complete
Parameter Agent 1 Agent 2 Agent 3 enumeration is not feasible for applications to real-world
d ili sales forces, in which there may be hundreds or thousands
5 2 2 of agents. We provide an algorithm that collapses the
r 4 3 5 mixed-integer program to a standard optimization problem
of searching a continuous function on a compact set.
U° 9 10
4
The complexity of the resulting optimization program is
linear in N. It therefore scales and leads to substantial

200.3.152.96 on Fri, 19 Aug 2022 03:56:44 UTC
Table 2
COMPOSITION, PROFITS, COMMSSIONS, SALES, AND COMPENSATION IN THE THREE-AGENT EXAMPLE
Composition a Profits Commission Rate Sales Compensation
Uniform commissions M E[n(ß)] ß E[XS¡] E[Wm]

One agent (1,0,0) .14 .56 3.78 3.64
(0,1,0) .06 .18 1.12 1.06
(0,0,1) 1.29 .29 1.57 .29
Two agents (1,1,0) .02 .45 4.54 4.52
(1,0,1) 1.28 .44 5.06 3.78
(0,1,1) 1.33 .25 2ą61 1.33
Three agents (1,1,1) 1.24 .39 5.99 4.74
Individually optimized contracts (1, 1, 1) 1.49 (.56, .18, .29) 6.47 4.99
aThe three digits in each set correspond to Agents 1, 2, and 3, respectively; 1 = agent present in c
Notes: Boldface indicates the optimal composition.
improvements in computational time over changing

without alternative ge-
the incentives for the other agents, t
neric algorithms for mixed-integer programs,
that even
agent can clearly befor
profitably employed at a c
relatively small N. mission optimized for the composition including him
her. If so,isthe
The key idea is that, when the commission excluded
held fixed,agent belongs to the optim
composition,
the individual profitability of any agent and we haveof
is independent a contradiction. We can t
construct the conditional
the sales force composition. The principal's problemvalue function
ofthat concentrates
whether to keep an agent at a givenout
commission
the composition therefore
problem conditional on ß as
depends only on whether the agent is profitable at that
commission and not on the composition itself. We can
then solve the problem with an inner-outer loop pro-
(14) 7t(ß)= X IniSOJtiiß),
i€(l
cedure. In the inner loop, we find the optimal composition

for any commission by retainingwhich only agents
sums who
the are
profit contribut
profitable at that commission. In the outer loop,in
retained wethe
search
optimal composi
for the optimal commission over the cause
unit interval. We now
the composition problem
formalize the idea.
integer optimization, one may su
Equation 10 gives the optimal uniform commission ß forfunction is discontinuous, preclu
any given composition profile. Given ß, one can back out We now prove the following us
the firm optimal salaries from the IR constraint, agent by
tional value function:
agent, as follows:
P2: The conditional value function rc(ß) is continuous in ß.
(13) a(ß):C£(a, ß) = U°, Proof. Both the expectation of the sales process in
Equation 4 and the concentrated salaries in Equation 13 are
where we note that the optimal salaries a(ß) are continuous in
continuous functions of ß, so it follows that the function
ß. The optimal contracts are now summarized by ß. Define
7üi(ß) is continuous in ß for all íg(1,...,N). The profit
the profit contribution of an agent i' at contract ß as
contribution of any agent i to 7c(ß) is the upper envelope of
7Ci(ß) =E[Si(ß) - (cxi(ß) - ßSi)] foralli G M.two continuous functions of ß, that is, max{0,7Ci(ß)},
which is itself continuous. Finally, for any ß, rc(ß) is the
sum of continuous
We then show that at the optimum, the profit contribution of functions and is therefore itself also
any included agent is positive. continuous. ■
Algorithm.
Pi: At the optimal composition-compensation pairs (M*, ß(A4*))> The results above give us the solution
algorithm:
1€ 7Ci(ß(A1*)) > OforalliGjM* and
1. For a given ß, calculate 7Ci(ß) for all i G (1, N).
2. 7Ci(ß(.M*)) < Ofor alii G (1, ...,N)/Ať' 2. Search the conditional value function (Equation 14) over
[0, 1] for the optimal ß*.
Proof. For the first part, suppose not, and that 7ii'(ß(A4*))
At ß*, the
< 0 for some agent i' G M*. Then the firm could increase optimal composition is i:7ii(ß*) > 0,iG (1, ...,N).
profits
The algorithm
by firing agent i' and fix the compensation W(ß(.M*)) for thereduces the joint problem to a search of a
remaining agents in A/i*'i/. Because none of thecontinuous
remaining function over a compact set, which is a
standard
agents would face different incentives, their profit optimization problem. Utilizing this character-
contri-
butions would stay constant and net profit would ization enables us to reduce the problem to a unidimen-
improve.
But that contradicts the optimality of the composition-
sional search over ßG [0, 1], rather than a combinatorial
compensation pair It follows that 7ij(ß(.M*)) search over the space of 2 compositions. For N moderately
> 0 for all i G M*. Part 2 follows by analogous logic.
large, ■
this approach can yield a significant improvement
The intuition is straightforward: if an agent excluded
in terms of computational time. Note that 7c(ß) is piece-
from the optimal composition can be profitably wise differentiable and not generally globally concave.
employed

200.3.152.96 on Fri, 19 Aug 2022 03:56:44 UTC
We provide some illustrations of its properties for various Misra and Nair (2011) used these data to estimate the
sizes of N in the Appendix. underlying parameters of agents' preferences and envi-
ronments, using a structural dynamic model of forward-
EMPIRICAL APPLICATION looking agents. For our simulations, we use some parameters
Our data come from the direct selling arm of the salesfromforce that study as well as some that are calibrated. We
provide
division of a large contact lens manufacturer in the United States a short overview of the model and estimation in
the following sections, noting differences from their
(we cannot reveal the name of the manufacturer for confiden-
analysis
tiality reasons). These data were used in Misra and Nair (201 1), in passing.
and some of the description in this section partially borrows
from that study. Contact lenses are primarily sold through The Model for Sales Agents
prescriptions to consumers from certified optometrists.The Im-compensation scheme involves a salary (Xt, paid in
portantly, industry observers and casual empiricism suggests month t, as well as a commission on sales ßt. The sales on
that there is little or no seasonality in the underlying demand for the commission is accrued are reset every R months.
which
the product. The manufacturer employs a direct sales force The
in the
commission ßt is earned when total sales throughout
United States to advertise and sell its product to optometrists
the sales cycle Qt exceed a quota at and fall below a ceiling
(also referred to as "clients"), who are the source of demand
bt. No commissions are earned beyond bt. Let It denote the
origination. The data consist of records of direct orders made since the beginning of the sales cycle, and let qt
months
from each doctor's office through an online ordering system,
denote an agent's sales in month t. Let %t be an indicator for
and they have the advantage of tracking the timing and origin of
whether an agent stays with the firm. When %t = 0, it in-
sales precisely. Sales agents for this manufacturer are assigned
dicates the agent has left the company and is pursuing an
their own nonoverlapping geographic territories and are paid option. Assume that once an agent leaves the firm,
outside
according to a nonlinear period-dependent compensation he or she cannot be hired back (i.e., %t = 0 is an absorbing
schedule. Pricing issues play an insignificant role in output
state). The total sales, Qt, the current quota, at, the months
because the salesperson has no control over the pricing decision,
since the beginning of the cycle, It, and the agent's em-
and price levels remained fairly stable during the period for
ployment status, %t, are the state variables for the agent's
which we have data. The compensation schedule for the problem.
agents We collect these in a vector st = {Qt, at, It, %t} and
involves salaries, quotas, and ceilings. Commissions are collect
earned the observed parameters of the agent's compensa-
on any sales that exceed the quota but are below the ceiling.
tionThescheme in a vector = {a, ß}. We will use the data in
salary is paid monthly, and the commission, if any, is paidcombination
out at with a model of agent behavior to back out the
the end of the quarter. The sales on which the output-basedparameters indexing agents' types. The results in this study
compensation is earned are reset every quarter. Additionally,
are the
obtained taking these parameters as given.
quota may be updated at the end of every quarter depending onindex i for agent is suppressed in what follows. At
The
the agent's performance ("ratcheting"). Our data include thethebeginning of each period, we assume the agent ob-
history of compensation profiles and payments for every sales his or her state and chooses to exert effort et. On the
serves
agent, as well as monthly sales at the client level for each of theseof this effort, sales qt are realized at the end of the
basis
sales agents for a period of approximately 3 years (38 months).
period. Sales qt is assumed to be a stochastic, increasing
The firm in question has more than 15,000 differentfunction stock of effort e and a demand shock et, where
keeping units (SKUs). The product portfolio reflects the qt large
= q(et, e). The agent's utility is derived from his or her
diversity in patient profiles (e.g., age, incidence of astigmatism,
compensation, which is determined by the incentive
nearsightedness, farsightedness), patient product needs (e.g., We write the agent's monthly wealth from the
scheme.
daily, disposable), and contact lens characteristics (e.g.,
firm as Wt = W(st,et,et;|Li, V) and the cost function as
hydrogel, silicone-hydrogel). The product portfolio of the firm
de? /2, where d is to be estimated. We assume that agents
features new product introductions and line extensions that
areand
reflect the large investments in research and development risk-averse and that, conditional on %t = 1, their per-
testing in the industry. The role of the sales agent is partlyutility function is
period
informative, in providing the doctor with updated information de2
(15) ut = u (Qt, at, It, Xt = 1) = E [Wt] - r x var [Wt] -
about new products available in the product line, and in
suggesting SKUs that would best match the needs of each
Here, rby
patient profile. The sales agent also plays a persuasive role is a parameter that indexes the agent's risk aversion,
and
showcasing the quality of the firm's SKUs relative to those ofthe expectation and variance of wealth is taken with
competitors. Although each agent's frequency of visiting respect to the demand shocks et- In the case of a salary plus
piece
doctors is monitored by the firm, the extent to which the rate of the type considered earlier, Equation 15
agent
promotes the product once inside the doctor's office cannot be to exactly the form denoted in Equation 5 for the
collapses
CE. Wecan
monitored or contracted upon. In addition, although visits can thus interpret Equation 15 as the nonlinear-
be tracked, whether a face-to-face interaction with a contract
doctor analog to the CE of the agent under a linear
commission.
occurs during a visit is within the agent's control (e.g., an The agent's payoff from leaving the focal
firm
unmotivated agent can simply register with the receptionist, and pursuing an outside option is normalized to U°,
which counts as a visit but is low on effort).4 that is, u, = u(Qt, a,, It, %t = 0) = U°.
In this model, sales are assumed to be generated as a
function of the agent's effort, which is chosen by the agent
4The firm does not believe that sales visits are the right measure of effort.
to maximize
Even though sales calls are observed, the firm specifies compensation based his or her present discounted payoff subject to
on sales, not calls. the transition of the state variables. The first state variable,

200.3.152.96 on Fri, 19 Aug 2022 03:56:44 UTC
total sales, is augmented by theGiven this specification,

realized salesthe structural
each parameters
month, that de-
scribe an agent
except at the end of the quarter, when the Q, are estimated in
agent two steps. with a
begins
fresh sales schedule; that is,
Estimation
(16) n m Q,+ o " _ í Qt + Qt ifFirst,

It we< recognize
R that once effort, et is estimated, we
(16) n m Q,+ o 1 " _ ļ o if It = R ' can treat hidden actions as known. The theory implies St
For the second state variable, quota, we for
is the state vector estimate a semi-
the agent's optimal dynamic effort
parametric transition function that relates the updated
choice. We can use the theory, combined quotawith dynamic
to the current quota and the performance programming of thefor
to solve agent relative
the optimal policy function
to that quota in the current quarter: e*(st; Œ), given a guess of the parameters Q. Because et is
f at if It < R known, we can then use et = e*(st;Q) as a second-stage
(17) at+i = { v,K estimating equation to recover Q. Misra and Nair (201 1)
12,k=i0knat,Qt v,K + qt) + vt+1 if It = R implement this approach on an agent-by-agent basis to
recover Q for each agent separately. The approach ex-
In Equation 17, the new quota is allowed to depend flexibly on
ploits the panel data available at the client level for each
a t and Qt + qt, via a Korder polynomial basis indexed by
agent to avoid imposing cross-agent restrictions, thereby
parameters 0k to capture, in a reduced-form way, the man-
obtaining a semiparametric distribution of the types in
ager's policy for updating agents' quotas. The term vt + i is an
the firm.
independent and identically distributed random variable that is
The question remains of how the effort policy, et = e(st),
unobserved by the agent in month t. The distribution of vt + 1 is
can be obtained. The intuition used in Misra and Nair
denoted £/v(.) and will be estimated from the data. The
(2011) is to exploit the nonlineari ty of the contract
transition of the third state variable, months since the begin-
combined with panel data for identification. The non-
ning of the quarter, is deterministic, augmented by 1 with the
linearity implies the history of output within a compen-
passage of calendar time within the quarter. Finally, the agent's
sation horizon and is relevant for the current effort
employment status in month t + 1 depends on whether the
decision because it affects the shadow cost of working
agent decides to leave the firm in period t. Given the state
today. Thus, effort is time-varying and dynamically ad-
transitions, we can write an agent's problem as choosing
justed. The relationship between current output and his-
effort to maximize the present discounted value of utility
tory is observed in the data. This relationship will pin
each period, where future utilities are discounted by the
down hidden effort. Intuitively, the path of output within
factor p. We collect all the parameters that describe
the compensation cycle is informative of effort. We refer
the agent's preferences and transitions in a vector
the reader to Misra and Nair (201 1) for further details of
Q = {|X,d,r,<?e(-)>!?v(-)'0k,k=i,...,K}- In month It < R, the estimation and identification.5 For the counterfactuals in
agent's present discounted utility under the optimal effort
policy can be represented by a value function that satisfies the
this article, we need estimates of {h,k,d,r,U°,i/e(.)}.
following Bellman equation:
Here, we assume that QeĻ) ~ N(0,a2). So, we need
0 = {h, k, d, r, U°, a}. We use the same parameters as Misra
(18) V(Qt,at,It,xt;£2,vI') and Nair for these, estimating a by imposing the normality
assumption of the recovered demand-side errors from the
i u(Qt, at, It, e; £2, ¥) ]
model. The parameter k is not estimated in Misra and Nair.
^ n 1 + p/ev(Q' + 1 = e)l' a> + 1 r we exploit additional data not used in that study on
Here,
X..,€(0,l),e>0 ^ n 1 ļ =at>It+1)Xt+i;a4/)f(et)det + 1 + 1 J the number of calls made by each agent i to a client j in
Similarly, in month It = R, the Bellman equation determining
month t, which we denote as kyt. We observe kyt and
effort is obtain a rough approximation for ki as k¡ « ^XtXjkijt- The
incorporation of k in the model does not change any of the
(19) VtQt.at.R.x,;^) other parameters estimated in Misra and Nair; it changes
only their interpretation. We use these parameters for all
' u(Q„a„R,xt,e;i2,V) the simulations reported next.6
= max J +PJvJeV(Q-+'=°'at+'
RESULTS
>o =a(Q„q(e„e),a„Vt+l),l,xt+1;£2,T) f"
k X f(et)<ļ>(vt+ i)detdvt+ i We first discuss the results from the calibration of the
agent type parameters. We report these results in Table 3.
Conditional on the agent's staying with the firm, the optimal
effort in period t, et = e(st;Q,lF), We use a set of 58the
maximizes agents in our analysis who are all
value
located in one division of the firm's overall sales force.
function:
The numbers we report have been scaled to preserve
(20) eistifì,^) = argmax{V(st jß,^)}.
e > 0
5See also Steenburgh (2008) and Larkin (2010), who note that effort is
affected by how far away the agent is from his or her quota.
The agent stays with the firm if the value from employment is
6Note that given that the data are from only one firm and there is no hire/fire
positive: variation, only an upper bound on each agent's outside option is identified.
We use these upper bounds as the estimate for Uf in our results reported next.
%t+1 = lifmax{V(st;Q^)}>0. Point estimation of each agent's outside option will require data on workers
who leave the firm and their pay elsewhere, as well as specifying a fuller
This completes the specification of the model that specifies model of labor market sorting. In simulations we conducted, the firm's
the agent's behavior under the plan that generated the data. modest uncertainty around U? did not significantly alter the results reported.

200.3.152.96 on Fri, 19 Aug 2022 03:56:44 UTC
171
Aligning Sales Force Composition and Compensation
Table 3 scatterplot shows pairs of agent types across the agent pool.
ESTIMATES OF AGENT TYPES For instance, plot (4,1) in Figure 1 shows a scatterplot of risk
aversion (r) versus cost of effort (d) across all agents in the
Parameter pool. Plot (1,4) is symmetric and shows a scatterplot of cost
of effort (d) versus risk aversion (r). The seven agents who
Statistic h k r d cr U°
are not included in the optimal composition are represented
Mean .9618 1.0591 .0466 .0436 .4081 .0811 by open symbols, highlighted in red. For example, we see that
Median .9962 1.0802 .0314 .0489 .3114 .0864 one of the dropped agents, represented as an open circle, has a
Minimum .5763 .2642 .0014 .0049 .0624 .0710
high risk aversion (first row), an average level of sales-
Maximum 1.4510 1.8110 .3328 .1011 1.5860 .1032
territory variance (second row), an average level of pro-
ductivity (third row), a low cost of effort (fourth row), a low
outside option (fifth row), and a low base level of sales (last
confidentiality; however, the scaling isrow). This agentuniformly
applied has a low cost of effort; however, his/her
and is comparable across agents. For purposes of intuition,
high risk aversion, low outside option, and fit relative to the
the reader should consider h and U° distribution
to be in of millions of across the rest of the
these characteristics
dollars. So, roughly speaking, the median outside
agents implies option
that this agent is not included in the preferred
in the data is approximately $86,400, and the
composition. average
Figure 1 illustrates the importance of multi-
agent's sales in the absence of effort isdimensional close toheterogeneity
a millionin the composition-compensation
dollars. trade-off that faces the principal, and it emphasizes the im-
The rest of the article is organized as follows: We portance of allowing for rich heterogeneity in empirical in-
condition on the parameters above and solve for the optimal centive settings.
composition and compensation for the firm's sales force In Figure 2, we plot the location of these salespeople
using the algorithm described previously. We then discuss on the empirical marginal densities of profitability and
two different scenarios. First, we simulate the fully het- sales across sales agents. What is clear from Figure 2 is
erogeneous plan, whereby each agent receives a compen- that there is no a priori predictable pattern in the location
sation plan (salary plus commission) tailored specifically of these agents. In some cases, the agents lie at a tail end
for him or her. We also simulate the partially heterogeneous of the densities, although this does not hold generally.
contract, whereby the commission rate is common across Furthermore, the dropped agents are not uniformly at the
agents but the salaries can vary across individuals. In all the bottom of the heap in terms of expected sales or profit
results presented, we assume that when an agent is excluded contribution under the fully heterogeneous plan. For
from the sales force, the territory provides revenues equal to example, Agent 33, one of the dropped agents, has ex-
xh, with X = 0.95 and h the intercept in the output equation. pected sales of $1.70 million under the fully heteroge-
This assumption reflects the fact that even if a territory were neous plan, which would place him/her in the top decile
vacated, sales would still accrue on account of the brand or of agents in terms of sales. In addition, he/she is also in
because the firm might use some other (less efficient) the top decile across agents in terms of profitability.
selling approach such as advertising. We also explored However, in his/her case, the variance of sales is the
alternative assumptions (e.g., t = 0, x = 1); these results are highest in the firm, and this creates a large distortion in
available from the authors on request. Qualitatively, the the contract because the effect it induces on the optimal
results obtained were similar to those presented here. In the commission rate ß. Not including this agent allows the
next section, we organize our discussion by presenting firm to improve the contract terms of other agents,
details of the optimal sales force composition chosen by thereby increasing profits. Other agents are similarly
the firm under these plans. We then present details of effort, dropped because of some other externality that in-
sales, and profits. fluences the compensation contract. Figures 1 and 2
accentuate the difficulties of ranking agents as desir-
Composition able or not desirable on the basis of a single type-based
We start with the fully heterogeneous plan as a bench- metric and the need for a theory of value to assess sales
mark. We find all agents have positive profit contributions agents.
when plans can be fully tailored to their types. Conse-
Compensation
quently, the optimal configuration under the fully hetero-
We now discuss the optimal compensation implied for
geneous plan is to retain all agents (the "status quo"). This
is not surprising, as noted in our previous three-agent the firm under the optimal composition. We compare the
simulation. fully heterogeneous plan with the partially heteroge-
Simulating the partially heterogeneous compensation neous plans with and without optimizing composition. In
plans, we find the optimal composition in this sales force Figure 3, we plot the density of optimal commission rates
would involve letting go of seven salespeople. It is note- under the fully heterogeneous plan along with those for the
worthy to investigate the characteristics of the agents who are partially heterogeneous plans. The dotted vertical lines are
dropped and to relate them to those of the agent pool as a drawn at the common commission rate for the partially
whole. In Figure 1, we plot the joint distribution of the heterogeneous plans, with the blue dotted line indicating
primitive agent types {h, k, d, r, U°, a} for all agents at the the rate when composition is optimized and the black dotted
firm. The marginal densities of each parameter across agents line indicating the rate when composition is not optimized.
are presented on the diagonal. Each point in the various Looking at Figure 3, we see that the commission rates vary
scatterplots along the off-diagonals represents an agent. Each significantly across the sales agents under the fully

200.3.152.96 on Fri, 19 Aug 2022 03:56:44 UTC
Figure 1
JOINT DISTRIBUTION OF CHARACTERISTICS OF AGENTS WHO ARE RETAINED AND DROPPED FROM FIRM UNDER PARTIALLY
HETEROGENEOUS PLANS
optimal composition) exert an externality that brings the

heterogeneous plans, going as high as 2.5% for some agents
(median commission of approximately 1.2%). Under the commission rate down. By eliminating the "bad"
overall
partially heterogeneous plans, the optimal commission rates the firm is able to increase incentives. To what extent
agents,
does
are lower. Of note is that the ability to fine-tune sales this improve effort, sales, and profitability? We discuss
force
this question
composition has significant implications in this setting. In next.
particular, when the firm is constrained from fine-tuning the
Effort and Outcomes
sales force, it sets an optimal common commission of ap-
proximately .5%. When it can fine-tune the sales force,The theprofits for the firm under the fully heterogeneous
firm optimally sets a higher commission rate of approxi-plan are estimated to be $60.56 million. We decompose
mately .9%. When the firm is constrained by theprofits
com- with and without partially heterogeneous plans
pensation structure, the extreme agents (eliminatedand with and without optimizing composition. As noted
in the

200.3.152.96 on Fri, 19 Aug 2022 03:56:44 UTC
Figure 2
PROFITABILITY AND SALES OF ELIMINATED SALES AGENTS
A: Contribution Under Status Quo
in / '
£
(/>
o " /
" / '
'
c / '
& LO
c / _
/ ' '
'
P -
-.5 .0 .5 1.0 1.5
B: E(Sales) Un
in / '
*2-
<0 /
/ '
'
c / '
s c / / ' '
IT) _ j/ '
O -
I , I
.0 .5 1.0 1.5 2.0 2.5
previously, in
empirical CDF of effort our
and sales under the three scenarios. da
contributions, The status quo plan is the oneandthat keeps the currentcon
sales
is identical force composition
to butthe changes compensation. Both the
statu
fully heterogeneous. sales and effort distributions under the fully heterogeneous C
heterogeneous plan fall to the right of the plan wi
partially heterogeneous plans.
tical to those However, Kolmogorov-Smirnov
under tests show that the dis- th
in the first row of Table 4. tribution of sales and effort under the optimized composi-
In contrast to the fully heterogeneous compensation tion and compensation scenario is not statistically different
structure, there is a significant difference in profit levelsfrom that under the fully heterogeneous plan. This is strik-
when compensation plans cannot be customized. Lookinging because it suggests that by simply altering composition
at Table 4, we see that partially heterogeneous plans that in conjunction with compensation, a firm can reap large
include the ability to fine-tune composition come very closedividends in motivating effort, even under the constraints of
to the fully heterogeneous plan in terms of profitability partial heterogeneity in contractual terms. This is also why
($59.18 million vs. $60.56 million). But partially hetero- the overall profits under the optimal composition plan with
geneous plans that do not include the ability to fine-tune thecommon commissions are so close to those under hetero-
sales force cause a distortion in incentives and result in a geneous plans.
profit shortfall of $3.4 million, bringing the total profits We now assess the extent to which profits at the indi-
down to $55.78 million. vidual sales agent level, under the partially heterogeneous
To decompose the source of profitability differences plan that includes the ability to choose the composition
across the different scenarios, in Figure 4, we depict the of agents, approximate the profitability under the fully

200.3.152.96 on Fri, 19 Aug 2022 03:56:44 UTC
Figure 3
OPTIMAL COMMISSION RATES UNDER FULLY HETEROGENEOUS, FULLY HOMOGENEOUS AND PARTIALLY HETEROGENEOUS PLANS
---■ Optimal partial-het plan ļ f sv

Status quo partial-het plan ; / '
80 - / '
60 / I '
oT
■e /
/ 1
' '
'
■e « / '
s / : '
c / - '
.2
(0
/ ! '- '
'
CO / ' ! '
g 40 - / : '
I-
- ' 'ļ
'ļ ' '
' ' '
'
20- / ; ;
0 .005 .010 .015 .020 .025
Commission Rate
heterogeneous plan (the baseline or best-case scenario).able

In to choose the type space of the agents concomitant with
incentives.
Figure 5, the x-axis represents profitability (revenues
minus payout) of each agent under the fully heterogeneous What is the mechanism that enables the firm to come
plan. The y-axis represents agents' profitabilities under
close to the profitability of the fully heterogeneous plan by
the partially heterogeneous plans with and withoutoptimizing
the the composition of its agents? The intuition is
ability to optimize composition. Solid dots represent straightforward. When constrained to set a homogeneous
profits when composition is optimized; open dots repre-plan, a firm can do much better if the agents it must in-
centivize are more homogeneous. Consider an extreme
sent profits when composition is held fixed at the status
quo. Each dot represents an agent. Axis scales are caseinwherein the firm could find any number of agents of
millions of dollars. Looking at Figure 5, we see thatany thetype to fill its positions (i.e., no search costs for labor).
The firm would first pick the agent from whom it could
ability to choose composition is important. In particular,
the profitability at the agent level when firms are obtain
con- the highest profit (output minus payout) under the
strained to partially heterogeneous contracts without fully tailored heterogeneous contract. It would then fill the
the ability to optimize composition is much below N available positions with N replications of that agent.
the
profitability in a situation wherein contracts can be fully
tailored to each agent's type. Indeed, the ability to choose
agents seems to be able to mitigate the loss in incentives Table 4
implied by the constraint to homogeneity. The profit-
PROFITS UNDER FULLY AND PARTIALLY HETEROGENEOUS
ability under composition-optimized, partially hetero-
PLANS
geneous contracts comes very close to that under fully
tailored contracts.
We think this result is an important takeaway. In the real Composition
world, firms can choose both agents and incentives, not Compensation Plan Status Quo Optimal
incentives alone. Firms do face constraints when setting
Fully heterogeneous $60.56 million $60.56 million
incentives, but our results suggest that the profit losses Partially heterogeneous $55.78 million $59.18 million
associated with these constraints are lower when firms are

200.3.152.96 on Fri, 19 Aug 2022 03:56:44 UTC
Figure 4
EMPIRICAL CDF OF IMPLIED EFFORT AND SALES UNDER DIFFERENT COUNTERFACTUAL COMPENSATION AND COMPOSITION
PROFILES
.0 .2 .4 .6 .8 1.0 .5 1.0 1.5 2.0

Effort Sales
- o - Status quo partial-het plan - 0- Status quo partial-het plan

- B - Optimal partial-het plan - b - Optimal partial-het plan
- © - Status quo full-het plan - 9 - Status quo full-het plan
E[tc({A,A})]
Then, the uniform commission it charged for the sales =E[TC({B,B})] >E[TC({A,B»].
force as a whole would be optimal for every agent in the
It is in this sense that the firm has a preference f
firm. Thus, heterogeneity involves costs. In this sense, an
geneity reduction. Optimal contracting requires th
increase in heterogeneity has two effects on a firm con-
strained to uniform contracts. On the one to satisfy
hand, both the IR and the IC constraints for
it increases
agents. Allowing for agent-specific salaries enable
the chance that the firm employs high-quality agents (a
to satisfy IR constraints for the agents it wants
positive). However, on the other hand, it also increases the
However, the constraint to a common commissio
level of contractual externalities (a negative). The optimal
that IC constraints become harder to satisfy when
composition has to balance these competing forces.
pool becomes more heterogeneous. Thus, a firm th
To state this more formally, consider a firm that has
choose the agent pool should prefer one that is relat
demand for two agents, which it can fill with agents of type
homogeneous, ceteris paribus.
A or B. Let 0 index type, and suppose type A agents
To empirically assess this intuition, we com
generate more profit than type B agents when employed at
measures of the spread in the type distribution o
their individually optimal contracts:
force under the optimal partially heterogeneous
©a * ©b, and E[tt({A})] > E[rt({B})].with and without the ability to choose agents. As
dispersion in types is complicated by the fact tha
Then, all things held equal, the firm should prefer compo-
space is multidimensional. We can separately com
sition {A, A} over {B,B}:
variance-covariance matrix of types in the sales f
E[rc({A,A})] >E[n({B,B})]. the two scenarios. To compute a single metric th
marizes the distribution of types, we define a m
Now, suppose that the types are such that although ©a * ©b>
spread, d m> as the trace of the variance-covarian
agents of types A and B generate the same expected profit
of agent characteristics:7
when employed at their individually optimal contracts:
(21) Úm = tr (1m) •
0A * 0B, but E[tc({A})] = E[tc({B})].
We find that d^StatusQuo = 78, 891.6, and d^^
Then, even though individual profits are the same, the firm
where d^statusQuo is the trace under the opti
should prefer to have the composition {A, A} or {B, B} over
{A,B} because composition {A,B} generates contractual
externalities: 7The trace of a matrix is the sum of its diagonals.

200.3.152.96 on Fri, 19 Aug 2022 03:56:44 UTC
Figure 5 Figure 6
PROFITABILITY AT THE INDIVIDUAL SALES-AGENT LEVEL OPTIMAL COMPOSITION DISPLAYS A DEGREE OF OUTLIER
UNDER FULLY HETEROGENEOUS PLAN AND PARTIALLY AVERSION
HETEROGENEOUS PLAN WITH AND WITHOUT OPTIMIZED
COMPOSITION
still retained, presumably on account of their higher abil-

ities or better fit with the rest of the agents.
In an important study, Raju and Srinivasan (1996) make
an analogous point, that allowing for heterogeneous quotas
partially heterogeneous plan when all agents in the firm in are
a common-commissions setting can closely approximate
retained, and d^0ptimal is the trace under an optimally chosenoptimal salary -plus-commission incentive scheme for a
the
heterogeneous sales force when those quotas can them-
partially heterogeneous plan when the set of agents retained
in the firm is jointly optimized. We see that the optimal selves reflect agent-specific differences. Our point is
configuration involves an approximately 35% reduction in
analogous, that a firm constrained to a homogeneous slope
heterogeneity. As another metric, we use d^ = det(X^).
on its incentive contract can come very close to the op-
The determinant can be interpreted as measuring the timumvol- by picking the region of agent types that it wants
ume of the parallelepiped spanned by the vectors of agent
to retain. However, the mechanism we suggest is differ-
ent. Raju and Srinivasan (1996) suggest addressing the
types. To the extent that the volume is lower, the spread
in types may roughly be interpreted as lesser. We findproblem
that of providing incentives to a heterogeneous sales
the determinant-based measure of spread shows a decline of by allowing for additional heterogeneity in contract
force
80% when the firm can pick its agents and incentives, terms. We suggest addressing the problem of setting in-
relative to picking only incentives. Both measures illustrate
centives to a heterogeneous pool of agents by making the
that under the optimal strategy, the firm chooses agents sales
such force more homogeneous. In another contribution,
that the residual pool is more homogeneous. AlthoughLai thisand Staelin (1986) and Rao (1990) show that a firm
finding is intuitive, what is surprising is that firm profits facing a heterogeneous sales force can tailor incentives to
under this restricted situation come so close to profits underthe distribution of types it faces by offering a menu of sales
fully heterogeneous plans. This result can only be assessed force plans. Their approach uses agents' self-selection into
empirically. plans as the mechanism for managing heterogeneity; it
To assess the intuition visually, we plot in Figure 5, the requires the firm to offer a menu of contracts, taking the
salaries and commissions of the entire set of agents when sales force composition as given. In our model, the firm
each can be offered a personally tailored contract (i.e., the offers only one contract to an agent but chooses which
salary-plus-commissions scheme from the fully heteroge- agents to offer attractive contracts to (thus, the margin of
neous case). The black dots in Figure 6 denote the agents choice for agents in our model is not over contracts but
who are retained in the optimal composition, and the red over whether to stay in the firm or leave). Our model
dots denote the agents who are dropped. Also plotted is the endogenizes the sales force's composition and may be seen
convex hull of the salary/commission points. We see that as applying to contexts in which offering employees a
the optimal configuration exhibits a degree of outlier menu of plans to choose from is not feasible or desired. We
aversion: the agents dropped from the optimal composition think the three perspectives outlined above for the practical
are all on the extremes of the distribution. Note at the same management of heterogeneity in real-world settings are
time that being an outlier does not automatically imply an complementary to each other. In the next section, we
agent is dropped: we see that some agents on the edges are discuss the latter mechanism further.

200.3.152.96 on Fri, 19 Aug 2022 03:56:44 UTC
Sorting Agents into Divisions Figure 7

We now discuss whether we canPERFORMANCE
further OF DIVISIONS IN MANAGING HETEROGENEITY
improve the
management of heterogeneity in the firm by sorting
agents into divisions. We consider a sales force archi-
tecture in which the firm creates 'J' divisions and assigns
each agent it retains to one of the 'J' divisions. The
divisions correspond to different compensation profiles.
We allow each division to have its own commission but
require that all agents within a division are given the
same commission. Salaries are allowed to be heteroge-
neous, as before. Such architectures are commonly ob-
served in the real world. For instance, sales agents
targeting large "key accounts" could be assigned to a
division that offers more incentive pay, whereas those
targeting smaller clients could be in a division that offers
more salary than commission. Alternatively, sales agents
targeting urban versus rural clients could be in two dif-
ferent divisions, each with its own commission scheme.
Regardless, the observed empirical fact is that commis-
sions are invariably the same within a division. This ar-
chitecture reflects that.
For each value of 'J |, we solve simultaneously for the
match between agents and divisions and the optimal
commissions across divisions, along with the optimal
salaries across agents given their assignment to a division.
Formally, we solve the following modified bi-level opti-
mization problem:
the profit profile in which we allow the firm to sort agents
into divisions and to optimize the composition as de-
max n-I 1rs, - Wxj(Si)]d^r(Si|ei),such that scribed previously. The difference in the profits under the
Mi jejJicM} blue versus the green line indicates the extent to which
composition choice adds to profitability above the ability
to sort agents into divisions.
WMi J = arg max [ Y [Si - W(Si)]d^(Si|ei), (IR,IC),and
J WewN J ifļļiļ We first discuss the situation wherein we allow the firm
to sort agents into divisions but not to choose sales
U M' = MN,
j ej force composition. The dashed horizontal line in Figure 7
represents a profit of $60.56 million, the maximum profit
where the last "adding-up" constraint ensurespossible
that aunder the fully heterogeneous contract (see
given agent is assigned to one of 'J' contracts. For
Table 4). Looking at the green line in Figure 7, we see
brevity, we do not write out the IR and IC constraints
that even without the ability to choose composition, the
explicitly. In the final solution to the previous
firmequa-
is able to come very close to this value with as few as
tion, the set assigns to each agent i a number six divisions. Even two divisions does a remarkably good
{0, 1, ...,j, ..., I J'}, where j = 0 implies the agentjob
is of managing heterogeneous incentives: profits under
dropped from the firm and j > 0 implies the agentthe is green line for the two-division case are more than
$59,000. Thus, one empirical takeaway is that a small
assigned to division j with wage contract Wmj - Cur goal
is to assess empirically how many divisions 'J' are
amount of variation in contract terms seems to be sufficient
required to fully span the heterogeneity and to cometo manage a large amount of heterogeneity in the firm, at least
close to the profits under the fully heterogeneous case.
in the context of these estimates.
Additionally, we want to assess the extent to which theWe now discuss the situation wherein we allow the firm
ability to choose agents interacts with this mechanism
to sort agents into divisions and to optimize its sales force
for managing heterogeneity. composition. We see that the results are similar to the
In Figure 7, we report the results of simulating the
previous case, but the firm is able to achieve a higher profit
firm's profits from creating up to 'J' = 6 divisions. Thegain
x- with fewer divisions (the blue line in Figure 7 is always
axis of Figure 7 plots the number of divisions considered above the green line). Thus, the ability to choose com-
('J'). The y-axis of Figure 7 plots the total profits to the
position is powerful even when one allows for sorting into
firm for each 'J'. Each point corresponds to solving the divisions. With ' J' = 6 divisions, we find the firm ends up
modified bi-level problem shown in the previous para- dropping five agents from the optimal composition (vs. six
graph for the corresponding value of ' J'. The green line agents with only one division). Thus, allowing for sorting
shows the profit profile in which we allow the firm to sortof agents into divisions does not automatically imply that
agents into divisions but do not allow it to optimize the composition choice is not needed; the right perspective is
composition (i.e., in the bi-level problem, we do not that sorting and composition choice are two strategies for
allow j = 0 as an option). The blue line in the figure shows
managing heterogeneity, and when used in combination,

200.3.152.96 on Fri, 19 Aug 2022 03:56:44 UTC
Figure 8 together permit very effective m

HETEROGENEITY REDUCTION IN THE SALES FORCE WITH neity, even with restricted contra
MANY DIVISIONS
DISCUSSION
We developed the exercise presented in this article under

the assumption that the firm knows each agent's typ
perfectly and can sort agents into divisions on the basis o
that knowledge. As mentioned previously, the theory has
emphasized an alternative mechanism by which the firm
offers a menu of contracts to all agents and each agent self
selects into one of the offered contracts according to his o
her type, which is unknown to the firm (this situation is
analogous to nonlinear pricing). This strategy helps a firm
manage heterogeneity when the firm does not know type
perfectly, and it emphasizes adverse selection as the main
difficulty in contract design, as opposed to the mor
hazard we emphasize. In practice, it is likely that both
strategies are at play in many real- world contexts. Optim
contract design with both adverse selection and mora
hazard is beyond the scope of this study. In sales forc
contexts, we believe the approach we have outlined above
is more realistic than self-selection contracts. First, unlik
nonlinear pricing, self-selection contracts are rarely ob-
served in sales force compensation (perhaps because of
concerns with dynamic signaling - if an agent chooses
contract with low commissions, he/she signals his/her typ
to the principal, which is information the principal could
they unlock powerful complementarities in the provision of
firmwide incentives. use to update the agent's contracts in subsequent periods)
The more common observation is of agent assignment to
Finally, we show that heterogeneity reduction plays a
sales force divisions. Second, adverse selection in sales
role in improving profitability with sorting. In Figure 8,
force settings is usually addressed by monitoring, pro-
we plot the log-distortion implied by the divisions against
the number of contracts offered. The distortion metric is bation, and training. New hires are often placed on a
salary-only probation period during which their perfor-
simply a way to summarize the average heterogeneity
mance is observed. The employment offer is made full
within a division. It captures the mean squared deviation
time conditional on satisfactory performance during the
from the average salesperson across divisions. Formally,
probation period. New hires are also provided significant
let 0ij denote the 6 x 1 vector of characteristics for agent
sales training during the probation period and are often
i allocated to division j > 0, and let 0^,r = 1,...,6 de-
asked to shadow an established sales rep; real-time
note the rth element of 0y . Denoting Nj as the number of agents
training is imparted and performance on the field is ob-
allocated to division j, let 0-^ = ^ average
served. This monitoring helps firms assess agent types
value of characteristic r inside division j. We define the
before full-time offers are made. Thus, in our view, for
distortion as long-term sales force composition and compensation with
full-time sales agents, adverse selection may be a second-
|J| 2 order consideration. A limitation of our model is that it
(22) d^min - X (ef - apply
does not ēf) 2
to the •
noteworthy dynamics previously
outlined that are associated with new employee hiring and
learning.
In Figure 8, the lower line (red) corresponds to the model that
allows for composition to be optimized
Finally, if a firm doesjointly with
not know agent types division-
perfectly, the
specific commissions, whereas profit
theit uppercan make whenlineoffering(blue)
a menu of divisions
ignores is the
strictly lower
composition aspect. As one would than the profitas
expect, it can make when contracts
more it knows
types perfectly and the
are added to the compensation structure, can assign each type to its preferred
distortion falls, but
allowing the firm to manage division (as in the simulations above).
composition results We reported in in this
a more
significant reduction. In essence, with
study that when types area small
known perfectly, number
firms still gain of
from the ability
contracts, the firm finds it optimal to to choose composition. Wethe
eliminate interpret this
outlying
result as implying that even
agents and use the increased flexibility to ifbetter
a menu of contracts is of-
compensate
the agents who are retained. As fered,atheresult,
ability to choosetheagents will still have power in
heterogeneity in
terms of profits in
the pool of retained agents is managed the contextbetter.
much of our empirical example.
Ultimately,
if the number of contracts increased to assumption
A primitive match the
in our total
analysis number
is that the firm
of agents, the two curves would knows thecoincide.
agents' types with Thatcomplete certainty.
is, if Al- every
agent got a customized contract,though this is a standard
there would assumption
be in no principal-agent
distortion.
models with pure
The broad point is that sorting and moralcomposition
hazard, it poses some relevantchoice

200.3.152.96 on Fri, 19 Aug 2022 03:56:44 UTC
questions for our analysis: Do the results continue

Figure to
9 hold
if the firm only has access to "estimates" of DROPPED
IDENTITY OF AGENTS the agents'
FROM THE OPTIMAL
types? If the results are different, what are
COMPOSITION AS the nature
A FUNCTION OF THE and
PERCENTAGE BY
magnitude of the differences? To answer thesePARAMETERS
WHICH ESTIMATED questions, ARE PERTURBED
we implemented extensive Monte Carlo simulations in
which the firm recognizes it has access to parameters of
agent types that are estimated with error and maximizes
an expected profit function that integrates out this esti-
mation error.8
Under the assumption that firms acknowledge the
presence of estimation error, we need a way of in-
corporating the parameter uncertainty into the firm's
decision-making process. We assume the firm has access
to the sampling distribution of the estimates. The firm
then uses this information to integrate out the uncertainty
and maximize expected profits' We simulate various
levels of uncertainty by perturbing the parameters we
estimated. We add to each point estimate a noise term
that is normally distributed with mean zero and with
standard deviations that vary such that the effective
range within which the firm believes each parameter can
lie is from ± 1% to ± 100% of the point estimate. This
corresponds to situations in which the firm can estimate
the parameter with a high level of precision, as well as
situations in which the parameter estimates are no longer
significant at the a = .05 level (i.e., range contains 0). We
used these perturbed parameters to compute the expected
profits faced by the firm. We then used continue
some agents the same opti-
to be (optimally) excluded in the
mization technology described previously to compute
solution set, suggesting that the broader point that
the optimal compensation and composition composition choice under
can help these
mitigate externalities under
simulated scenarios. All comparisons rigid contracts
with the continues
status to hold
quo even under extreme
and heterogeneous plan outcomes are also
parameter based on
uncertainty. the set of Monte Carlo
In another
appropriate expected profit functions. simulations (not reported here), when we resampled the
Our simulations show that the inclusion set of agents withof replacement,
parameter the results varied more
uncertainty does not alter our results qualitatively.
significantly, We the heterogeneity
suggesting that varying
find the composition results areinrobust across
the composition is moresimula-
relevant to the profitability of
tions, in that the same agents are eliminated
the firm from
than is the estimation the
error. Overall, these
analysis as before in most cases. To show this,
simulations show thatwe ourplot
findingsinabove are driven by
Figure 9 the retention or exclusion of each
meaningful agent
differences in types
in agent the(i.e., by hetero-
optimal composition as a function of
geneity) thethan
rather degree
by parameterof uncertainty per se.
perturbation to the parameter estimate. The
Looking at profits, agents'
we find that the IDs
profit levels do not
are plotted on the y-axis, and the degree vary muchof across these simulations, is
perturbation even when the pa-
on the x-axis (range: ± 1% to ± 100%). Each column rameters are perturbed by as much as ± 100%, the dif-
represents a perturbation level, and each square in a ference between the profits presented here and the
vertical column represents an agent in the configuration. counterf actuals differ by only around 7%. Finally, the
Agents dropped in the optimal configuration are repre- relative profits show the same patterns as before, with
sented in red. The first column plots the original results the homogeneous plan faring the worst and the plan with
with no perturbation to the parameters. Each column the optimal composition coming fairly close to the fully
thereafter plots the optimal composition found by heterogeneous plan.
maximizing expected profits (subject to appropriate IR
and IC constraints) given a certain level of perturbation.
CONCLUSIONS
Expected profits are found by Monte Carlo simulation
with R = 1, 000 draws over the parameter range. Looking We consider a situation in which a firm that is con-
at Figure 9, we see that the number of retained agents is strained to set partially heterogeneous contracts across
fairly stable (usually around the same level as the original its agent pool can optimize both its sales force compo-
result that included seven eliminations), and the identity sition and its compensation policy. We find that the
of the agents dropped from the pool is roughly preserved. ability to optimize composition partially offsets the loss
When the perturbation error is substantial (>±80%), there in incentives from the restriction to uniform contractual
appear to be frays in the optimal composition. Even so, terms. Homogeneity also implies a particular type of
contractual externality within the company. The pres-
8For the sake of brevity, we do not include complete details of ourence of an agent in the firm indirectly affects the welfare
simulations here. They are available from the authors on request. and outcomes of other agents through the effect the first

200.3.152.96 on Fri, 19 Aug 2022 03:56:44 UTC
but would not

agent induces on the common element ofchange our main point
contracts. Thisabout contractual
externalities
externality exists even in the absence ofand the codependence of compensation
complementarity
in output across agents, team and composition whencommon
production, contracts cannot ter-be tailored. In
ritories, or relative incentive terms. our data, weSimulations
do not have a way of and an the dis-
estimating
application to a real-world sales force
tribution suggest
of worker that
types in the the or the dis-
population
ability to choose sales force composition tribution of search costs has for empirical
labor among firms in this
power in terms of payoffs, sales effort,
market, both ofand which sales.
are critical inputs to a credible
This study explores the consequences empirical model of uniformity
of labor market sorting. With access to
but not the reasons for uniformity better data,in contracts
an extension within
of this sort could be pursued.
firms. Motivations for uniformity The reader could
should notebe that
aversions to in con-
such competition
sales agent inequity, concerns for fairness
tracts across firmsin has evaluation,
been relatively understudied in
preferences for simplicity, orempirical different kinds
work. Finally, ofmargin
another menu along which
costs. Lo, Ghosh, and Lafontaine (201 1)
the principal may conducted
manage heterogeneity field is to optimize
interviews with managers at the industrial
match between firms agents and in four
territories (e.g., Skiera
sectors (i.e., electrical and nonelectrical machinery
and Albers 1998). Analyzing andproblem
this matching
transportation equipment and instruments). while endogenizing the They report
compensation contract is out-
that the two main reasons managers side the scope of this article but is using
cite for not the subject of our
agent-specific sales force compensation ongoing work. plans are (1)
computational costs of developing complex
Although our contextplans and
is sales force (2)
compensation,
costs associated with managing ex ideas
similar post conflict
to the one exploredamong
here arise in other
sales agents induced by differential evaluation.
contexts of interest to marketing.Re- One area is joint
latedly, in a survey of 130 business-format franchisors,
choice of consumers and promotions. For instance,
Lafontaine (1992) reports that 73%
Belloni of
et al. surveyed
(2012) fran-
discuss an algorithm that enables a
chisors who chose uniform royalty university torates dida desirable
jointly choose so for mix of students
reasons of consistency and fairness and the leveltoward franchisees,
of scholarships required to attract them.
and 27% chose uniformity because it reduces the The complication associated with sales force compen-
transaction costs of administering and enforcing con- sation relative to these situations is the presence of
tracts. It seems, therefore, that fairness and menu costs
moral hazard. To the extent that we discuss the impli-
play a large role in driving such contract forms.9 cations of endogenizing the mix of agents at a firm, we
Notwithstanding the reasons, the fact remains that the
believe our analysis motivates development of richer
ability to choose agents and the restriction to partially
empirical models of the joint choice of how and to whom
heterogeneous contracts is pervasive in real-world
to offer product options to consumers in marketing and
business settings. However, principal-agent theory is economics.
surprisingly silent on both endogenizing the composi-
tion of agents and exploring the consequences of uni-
formity. We hope our first cut on the topic will inspire APPENDIX: BEHAVIOR OF THE CONDITIONAL
richer theory and empirical work on the mechanisms that VALUE FUNCTION
cause firms to choose similar contracts across agents and
on the consequences of these choices. We have simulated some data to illustrate some of the
We abstracted away from hiring and from the prin- properties of rc(ß). Agents are generated by spreading
cipal's policies for learning new hires' types, which lognormal noise around parameters (d,r,U°) = (1,2,0).
would complicate the model by introducing dynamics We then plot 7t(ß) in Figure Al for various N.
In the upper left quadrant, the conditional value
function is seen to be continuous and piecewise differ-
9We conducted interviews with salespeople and sales managers to entiable. The kinks are at the points where some agent's
understand why salaries are typically heterogeneous but commissions are
IR constraint just binds and the agent either enters or
invariably homogeneous in sales organizations. The common story we
have heard is as follows: Salaries are typically indexed against those at a exits the composition. At all other points, the criterion
hired employee's previous job (typically set as a percentage raise). Thus, is a sum of continuous and differentiable functions.
they reflect agents' outside options. Agents perceive the differences as Although the conditional value function has a clearly
fair because the variation is justified by managers as the costs to "beat" accentuated maximum in ßG [0, 1], this is not a general
competitive salaries to hire their colleagues. There is some evidence in
feature of the problem. The reader may note that the
the psychology literature that agents perceive variation as fair when it is
linked to "justifiable" costs. For example, Kahneman, Knetsch, and criterion has two local maxima, the second being slightly
Thaler (1986) document that agents do not perceive price discrimination above .2. Multiple maxima is a general feature of the
across consumers as unfair if they think it derives from differences in problem.
costs as opposed to the desire to extract more surplus from consumers The algorithm runs in linear time because only the
with higher valuations. On the other hand, commissions (and other forms
of incentive-based pay) reflect a percentage payout to an agent of rev-
profit contributions of each agent must be calculated at
enues brought into the firm. A dollar in revenue brought in by an agent A each iteration. For all N in this example, the optimization
is equally valuable to the principal as a dollar brought in by another agent is executed in less than one one-hundredth of a second
B; because of this, it becomes difficult for the principal to justify why A using standard numerical derivatives-based methods of
and B are rewarded different proportions of the dollar as commissions.
Such a policy is typically seen as "unfair." Clearly, the perception of optimization. In the lower right quadrant, the algorithm
fairness is linked to the deeper psychology of how human beings evaluate allows a directed search of a space of 25,000 possible
these kinds of trade-offs. compositions in fractions of a second.

200.3.152.96 on Fri, 19 Aug 2022 03:56:44 UTC
Figure A1
CRITERION FUNCTION FOR VARIOUS COMPOSITION SIZES
N = 3 N = 5
4
' 2,5 "

3 "
2 - '
Î2 '
*5 ' *5 '
*5 2 2 - ' ' *5 o 1.5 • ' '
o. ' a. '
.5 - '
o
0 .2 .4 .6 .8 1 0 .2 .4 .6 .8
ß ß
N = 10 N = 5,000
12
10 ' ' 4,000 ■ N.

8 - ' - '
ä
ä '¡h
' Ä' 3'000
Ä »SS" '
'
¡h
a.
2 6 -
'
' -2
o.
»SS '
'
' 2,000 - '
4 - ' '
2 1'000 " X.
0 .2 .4 .6 .8 1 0 .2 .4 .6 .8
ß ß
REFERENCES Desai, Prey as S. and Kannan Srinivasan (1995), "Demand Signaling

Under Unobservable Effort in Franchising: Linear and Nonlinear
Ackerberg, Daniel and Maristella Botticini (2002), "Endogenous
Price Contracts," Management Science , 41 (10), 1608-23.
Matching and the Empirical Determinants of Contract Form,"
Journal of Political Economy , 110 (3), 564-91. Godes, David and Dina Mayzlin (2012). "Using the Compensation
Armstrong, Mark (1996), Multiproduct Nonlinear Pricing,Scheme to Signal the Ease of a Task," working paper, University
Econometrica , 64 (1), 51-75. of Maryland.
Green,"In-
Bandiera, Oriana, Iwan Barankay, and Imran Rasul (2007), Jerry and Nancy Stokey (1983), "A Comparison of Tour-
naments and Contracts," Journal of Political Economy , 91 (3),
centives for Managers and Inequality Among Workers: Evidence
349-64.
from a Firm-Level Experiment," Quarterly Journal of Eco-
nomics, 122 (2), 729-73. Hamilton, Barton H., Jack A. Nickerson, and Hideo Owan (2003),
Basu, Amiya, Rajiv Lal, V. Srinivasan, and Richard Staelin 'Team Incentives and Worker Heterogeneity: An Empirical Anal-
(1985),
"Salesforce Compensation Plans: An Agency Theoretic ysisPer-
of the Impact of Teams on Productivity and Participation,"
spective," Marketing Science , 8 (3), 324-42. Journal of Political Economy , 111 (3), 465-97.
Belloni, Alexandra, Mitchell J. Lovett, William Boulding, and Bengt (1979), "Moral Hazard and Observability," Bell
Holmstrom,
Richard Staelin (2012), "Optimal Admission and ScholarshipJournal of Economics, 10 (1), 74-91.
Decisions: Choosing Customized Marketing Offers to Attract a
nomics, 13 (2), 324-40.
Desirable Mix of Customers," Marketing Science , 3 1 (4), 621-36.

200.3.152.96 on Fri, 19 Aug 2022 03:56:44 UTC
Lo, Desmond (Ho-Fu), Mrinal Ghosh, and Francine Lafontaine

(2011), "The Incentive
Provision of Intertemporal Incentives," Econometrica , 55and Selection
(2), Roles
of Sales Force
303-28.
Compensation Contracts," Journal of Marketing Research ,
American Economic48 Review
(August), 781-98. , 84 (4), 972-99
Mantrala, Murali K., Prabhakant
Ichniowski, Casey, Kathryn Shaw, Sinha, and andAndris A. Giovan
Zoltners
'The Effects of Human Resource
(1994), "Structuring Manage
a Multiproduct Sales Quota-Bonus Plan for a
Heterogeneous Sales
Productivity," American Force: A Practical Model-Based
Economic Review Ap-
Kahneman, Daniel, Jack L. Knetsch,
proach," Marketing Science , 13 (2), 121-44. and Ric
Milgrom, Paul and
"Fairness as a Constraint on JohnProfit
Roberts (1990), "The Economics of
Seeking:
Market," American Modern
Economic Review
Manufacturing: Technology, , 76 (
Strategy, and Organiza-
Kalra, Ajay and Mengze tion," American Shi (2001),
Economic Review , 80 (3), 511-28. "Desig
Contests: A Theoretical
Misra, SanjogPerspective," Market
and Harikesh S. Nair (201 1), "A Structural Model of
170-93. Sales-Force Compensation Dynamics: Estimation and Field
Kandel, Eugene and Edward P. Lazear (1992), "Peer Pressure and Implementation," Quantitative Marketing and Economics , 9 (3),
Partnerships," Journal of Political Economy , 100 (4), 801-17. 211-57.
Lafontaine, Francine (1992). "How and Why Do Franchisors Do
What They Do: A Survey Report," in Franchising, Passport for "Salesforce Design with Experien
Growth and World of Opportunity: Sixth Annual Proceedings of Transactions , 36 (10), 941-52.
the Society of Franchising, Patrick J. Kaufmann, ed. Lincoln, NE: Mookherjee, Dilip (1984), "Optimal In
International Center for Franchise Studies. Agents," Review of Economic Stud
Prendergast, Canice (1999), "The Pro
and Franchising Contracts: Evidence Firms," from theJournal
United States,"
of Economic Litera
Entrepreneurial Business Law Journal Raju, , 3 Jagmohan
(2), 381-434. S. and V. "Seenu" Sr
Lai, Rajiv and Richard Staelin (1986),Compensation "Salesforce Compensation
Plans for Multi-Ten
Plans in Environments with Asymmetric force," Management
Information," Market- Science , 42 (1
ing Science , 5 (3), 179-98. Rao, Ram C. (1990). "Compensating
Some Explicit Solutions," Marketin
Rochet,
and Multi-Product Salesforces: An Application Jean-Charles
of the Holmstrom- and Philipp
Milgrom Model," Management ScienceSweeping,
, 39 (7), 777-93.
and Multidimensional Scr
(4), 783-826.
Larkin, Ian (2010). "The Cost of High-Powered Incentive Systems:
Gaming Behavior in Enterprise Software Sales," working
Rotemberg, paper,
Julio J. and G. Saloner (2
Harvard Business School. and Strategic Direction," RAND
Lazear, Edward P. (2000a), "Performance Pay and Productivity," 31 (Winter), 693-716.
American Economic Review , 90 (5), 1346-61. Skiera, Bernd and Sönke Albers (1
Maximizing Sales Territory Alignme
Review , 90 (2), 410-14. (3), 196-214.
Steenburgh, Thomas J. (2008), "Effo
Optimum Labor Contracts," Journal of Political Bonuses,"
Lump-Sum Economy ,Quantitative
89 (5), M
841-64. (3), 235-56.
Lim, Noah, Michael Ahearne, and Sung Ham (2009), "Designing Zoltners, Andris A., Prabhakant Sin
Sales Contests: Does the Prize Structure Matter?" Journal of (2001), The Complete Guide to Ac
Marketing Research , 46 (June), 356-7 1 . formance. New York: American Ma

200.3.152.96 on Fri, 19 Aug 2022 03:56:44 UTC

Contratos Homogéneos para Agentes Heterogéneos - Alineación de La Composición y Compensación de La Fuerza de Ventas

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Contratos Homogéneos para Agentes Heterogéneos - Alineación de La Composición y Compensación de La Fuerza de Ventas

Uploaded by

Copyright:

Available Formats

Homogeneous Contracts for Heterogeneous Agents: Aligning Sales Force Composition

Stable URL: http://www.jstor.com/stable/43832424

This content downloaded from

Keywords: sales force compensation, incentives, hiring, agency theory,

Homogeneous Contracts for Heterogeneous

© 2016, American Marketing Association Journal of Marketing Research

This content downloaded from

Our second point

This content downloaded from

This content downloaded from

choice of agents and incentives, result

This content downloaded from

This content downloaded from

function Ui(Wi) = -exp{-riWi}, Equation

IR: C£' > U? Vi = 1, N. (11) E[n(ß)]=X(hi-U

where e[(ß) = = jļf. Solving Equation 9 gives the optimal

This content downloaded from

This content downloaded from

Composition a Profits Commission Rate Sales Compensation

Uniform commissions M E[n(ß)] ß E[XS¡] E[Wm]

improvements in computational time over changing

cedure. In the inner loop, we find the optimal composition

This content downloaded from

This content downloaded from

total sales, is augmented by theGiven this specification,

(16) n m Q,+ o " _ í Qt + Qt ifFirst,

This content downloaded from

This content downloaded from

optimal composition) exert an externality that brings the

This content downloaded from

A: Contribution Under Status Quo

-.5 .0 .5 1.0 1.5

.0 .5 1.0 1.5 2.0 2.5

This content downloaded from

---■ Optimal partial-het plan ļ f sv

0 .005 .010 .015 .020 .025

heterogeneous plan (the baseline or best-case scenario).able

This content downloaded from

.0 .2 .4 .6 .8 1.0 .5 1.0 1.5 2.0

- o - Status quo partial-het plan - 0- Status quo partial-het plan

This content downloaded from

still retained, presumably on account of their higher abil-

This content downloaded from

Sorting Agents into Divisions Figure 7

This content downloaded from

Figure 8 together permit very effective m

We developed the exercise presented in this article under

This content downloaded from

questions for our analysis: Do the results continue

This content downloaded from

but would not

This content downloaded from

' 2,5 "

10 ' ' 4,000 ■ N.

REFERENCES Desai, Prey as S. and Kannan Srinivasan (1995), "Demand Signaling

This content downloaded from

Lo, Desmond (Ho-Fu), Mrinal Ghosh, and Francine Lafontaine

This content downloaded from

You might also like