Copyright � 2001 INFORMS0092-2102/01/3103/S008/$05.001526–551X electronic ISSN

MARKETING—SALES FORCEDECISION ANALYSIS—APPLICATIONS

This paper was refereed.

INTERFACES 31: 3, Part 2 of 2, May–June 2001 (pp. S8–S44)

Sales-Force Decision Models: Insights from 25Years of Implementation

Prabhakant Sinhaprabha.sinha@zsassociates.com

ZS Associates1800 Sherman AvenueEvanston, Illinois 60201

Andris A. Zoltnersandy.zoltners@zsassociates.com

J. L. Kellogg Graduate School of ManagementNorthwestern UniversityEvanston, Illinois 60208and ZS Associates

Over 25 years, we have developed many sales-force and mod-eling insights through over 2,000 projects with several hundredselling organizations in over 50 countries. Content insights areuseful in making sales-force decisions. Examples are that prof-itability is flat for a wide range of sales-force sizes; phasedsales-force growth is rarely optimal; focused strategies domi-nate scattered strategies; most sales territories (55 percent) aretoo large or too small; and no compensation plan satisfies ev-eryone. Implementation insights concern model building, use,and implementation, for example, a model’s economic valuecan come from such sources as reduced uncertainty, accuracy,increased speed, objectivity, and stakeholder involvement; the-ory and practice have different and complementary perspec-tives; experience and wisdom are sometimes better than mod-els; and models provide insights, while people make decisions.

Two global firms recently merged andasked us to help them design new

sales organizations in each of 40 countries.The integration would affect over 15,000salespeople. In just two months, over 100people from our consulting firm used well-

tested normative sales-force decision mod-els with strong implementation processesto design and integrate the sales forces.We could not have imagined such an un-dertaking when we began our careers asmarketing modelers in the early ’70s.

SALES-FORCE DECISION MODELS

May–June 2001 S9

At the beginning of his teaching careerin 1973, one of the authors (Zoltners) fear-lessly made a pronouncement to a busi-ness models class, “In 15 years, modelswill play a prominent role in most majorbusiness decisions.” In this paper, we ex-amine how this author would adapt hisproclamation after implementing modelsfor over 25 years.

Our modeling experience is narrow anddeep. We have used descriptive and nor-mative models repeatedly to address deci-sions concerning such sales-force issues assize, structure, resource allocation, incen-tive compensation, and geographic de-ployment. The descriptive models havecharacterized how markets react to vari-ous sales-force decisions while the norma-tive models were optimizers that searchedthe solution space to find the best sales-force decision (Table 1).

From 1983 to 2000, we, along with ourcolleagues at ZS Associates, a consultingfirm focusing on sales-force issues, haveimplemented these models in over 2,000projects for hundreds of organizations.Two to three percent of all of the fieldsalespeople in the US have been touchedby the results. The firms had pressing is-sues that required quick attention. Compa-nies sought help when merging separateselling organizations, when launching newproducts, when facing deregulation, orwhen faltering in performance.

We have learned, and our clients havelearned from this experience. Two types ofinsights have surfaced. Content insightsstem from observing the results of re-peated model applications across compa-nies, industries, countries, and contexts.They are insights about sales-force sizing,

sales-resource allocation, sales-force de-ployment, and sales-force incentive com-pensation. Implementation insights arelessons that we learned about modelbuilding, model usage, and modelimplementation.

Three project samples were chosen toquantify some of the content insights. Thesamples were convenience samples. Theyincluded projects that were well docu-mented and easy to access. The ZS-SRASample is designed to develop sales-forcesize and resource-allocation insights. Itcomprises 50 sizing studies in six coun-tries with sales forces ranging in size from35 to several thousand. All of the studieswere conducted for companies that manu-factured and sold health-care products.

The ZS-TA Sample is used to developsales-territory-alignment insights. It isbased on 36 different sales-force-alignmentimplementations in eight industries. Allimplementations were in the US andCanada.

The ZS-IC Sample is used to developsales-force incentive-compensation in-sights. It is based upon seven compensa-tion studies in six industries. All imple-mentations were in the US and Canada.

Our content insights are provided next.The Model Builder Learns Through theModel-Building Process

Descriptive marketing models are devel-oped to learn about market behavior andmarketing theory. Normative marketingdecision models are designed to help man-agers make good marketing decisions. Butnormative decision models can also con-tribute to marketing theory. They haveproduced many useful sales-force insightsfor their model developers.

SINHA, ZOLTNERS

INTERFACES 31:3 S10

Sales-force issue Relevant decisions Descriptive model Normative model

Sales-force sizeand resourceallocation

Determine sales-force size

Allocate sales-forceeffort to marketsegments andproducts

Sales-response modelsdescribe how selling effortaffects the sales for marketsegments and products

Usually nonlinearprogramming modelsthat maximize 3 to 5year profitability foralternative sales-forcesizes and product andmarket allocations

Sales-territorydesign

Assign accounts orgeographic unitsto salespeople

Models calculate coverage,disruption, and profitimpact of alternativeassignments

Usually integerprogramming modelsthat maximizecoverage, minimizedisruption, ormaximize profit

Incentivecompensation

Design incentiveplan

Nonstandard models relateincentive compensationplans to company sales andprofits, for example,disaggregate models linkplan elements with eachsalesperson’s utility fortime and money, and salesresponse models link eachsalesperson’s call effortwith territory sales todevelop the aggregaterelationship

Nonstandard models, forexample, individualutility-maximizingtime-allocation modelsfeed aggregate sales-response maximizers

Table 1: These sales-force issues lend themselves to implementable descriptive and normativemodeling initiatives.

Effort Drives SalesThat marketing investment drives sales

is a fundamental principle supported bydata (Figures 1 and 2). Most executives be-lieve this principle, yet they sometimesuse decision rules that run counter to itspremise:—While addressing his divisional vicepresidents of sales, a CEO of a Fortune 50firm stated that last year he “cut the totalcompany sales-force size and sales wentup.” He did not allow any of the divisionsto increase their sales forces and cut mostof them. Will sales go up even further if

he cuts the sales force again?—Having completed a rigorous model-based analysis, a national sales managermade the following recommendation tohis division president: “We need moresalespeople, a 10 percent increase in size,and I project that we can increase reve-nues by five percent.” The president’s re-sponse was, “So you think that you can in-crease revenues by five percent? Do it! Butkeep the head count the same—just get 10percent more efficient.” What an out-come—the president increased the salesmanager’s revenue goal without increas-

SALES-FORCE DECISION MODELS

May–June 2001 S11

Figure 1: This scatter plot using cross-sectional data shows a statistically significant relationshipbetween sales-force effort and sales for a market segment serviced by a medical sales force.Every dot represents a sales territory.

ing the sales force. Will this managermake such a proposal ever again?—At an internal sales-force productivityworkshop, a country general manager as-serted that he maximized profits. Whenasked how, he responded that he “keptsales-force costs at 11 percent of sales.”

These examples demonstrate that somemanagers don’t incorporate the premisethat sales-force effort drives sales intotheir investment philosophy. The CEOwould have had even higher sales had henot cut his sales force. The division presi-dent believes that the best way to increasesales is by increasing productivity. It wasinteresting to observe the president’s reac-tion when he was told that the sales-forcesize might actually need to be increased as

a result of a productivity improvement.Productivity improvement lowers the av-erage cost of a sales call. Consequently,customers and prospects that were notprofitable enough to call on become worththe call. Finally, the country general man-ager was evoking a backwards principleby suggesting that sales should drivesales-force effort.

Two rational reasons for these decisionrules are likely. First, top managers maybelieve that the extra investment neededto increase size would be wasted becauseweak management processes would notincrease effective sales activity. Second,they know that the costs are incurred nowand are certain, but most of the impact isin the future and is not guaranteed. How-

SINHA, ZOLTNERS

INTERFACES 31:3 S12

Figure 2: This scatter plot using longitudinal data shows a statistically significant relationshipbetween sales-force effort and sales for a product sold by a pharmaceutical sales force. Everydot represents a quarter of the year.

ever, a well-managed increase in sales-force size usually leads to incrementalshort- and long-term increases in sales butnot necessarily to short-term increases inprofits.

To determine the winners of its annualincentive trip, a large consumer-productscompany ranked its sales territories fromthe highest to the lowest performing. Theranking showed that the fourth best terri-tory (out of 250) in the country was a va-cant territory. Who should go on the trip?How does this happen?Carryover

In most industries, carryover from prioryears’ selling effort contributes to currentsales in a territory. Some selling environ-ments favor carryover more than others.Significant differences in carryover can

even be observed across products sold bythe same sales force. For example, in thepharmaceutical industry, acute-care prod-ucts, such as antibiotics or antihistamines,have low carryover because patients typi-cally take these medications for short pe-riods. Salespeople can often persuade phy-sicians to try new acute-care medicationsparticularly when they have little healthrisk associated with them. Chronic-careproducts, such as blood pressure medica-tions or Alzheimer’s treatments, typicallyhave high carryover. Once patients startusing such medications, they will likelystay on them for a long time. Physiciansare reluctant to switch patients from medi-cations that are working, and they tend toprescribe products for new patients thatthey have found effective in the past.

SALES-FORCE DECISION MODELS

May–June 2001 S13

Figure 3: The magnitude of sales carryover across companies in the ZS-SRA Sample. We definethe base-case effort plan for any company as its current three-year sales-force size, structure,and resource-allocation plan. The base-case scenario (solid line) represents the average perfor-mance across all of the base-case plans for the companies in the sample. We indexed sales at100 in each year because sales levels varied across companies and over time. The zero-effortscenario (dashed line) represents an average model-based projection of sales if the 50 compa-nies were to have no sales forces. It is an estimate of carryover.

Carryover can be quite pronounced. TheZS-SRA Sample provides insight into thesize of the carryover that can be expectedin the health-care industry. The 95-percentconfidence interval for the aggregate sales-force carryover for the 50 companies thatwere members of the sample is 75 to 85percent in the first year, 62 to 78 percent inthe second year, and 52 to 70 percent inthe third year (Figure 3).

Sales forces are undersized if the carry-over effect is not considered. The bestfirst-year sales-force size changes depend-ing on which criterion management de-cides to use. Based on the ZS-SRA Sample,the best sales-force size using a one-year

contribution criterion is 18 percent smalleron average than the best size using athree-year discounted contribution crite-rion (Figure 4).The Flatness Principle

Company profitability can be flat for awide range of sales-force sizes around theoptimal size. Other authors have observedthis result as well [Chintagunta 1993;Mantrala, Sinha, and Zoltners 1992; Tull etal. 1986]. The ZS-SRA Sample demon-strates the degree to which flatness is evi-dent in practical settings. The three-yeardiscounted contribution varied, on aver-age, two percent for sales-force sizes thatvaried plus or minus 20 percent from the

SINHA, ZOLTNERS

INTERFACES 31:3 S14

Figure 4: The impact of carryover and multiyear thinking can affect the best sales-force size.Based upon the ZS-SRA Sample, the best sales-force size (95 percent confidence interval) is be-tween nine and 27 percent smaller using a one-year contribution criterion than a three-year dis-counted contribution criterion. The solid curve represents the three-year discounted contribu-tion and the dashed curve represents the one-year contribution associated with differentsales-force sizes. The one-year contribution is defined as net sales minus consolidated variableproduct costs, advertising and promotion costs, field-support costs, and sales-force cost. Thethree-year discounted contribution is defined as the present value of the next three years ofcontribution.

optimal size (Figure 5).When they are asked to increase sales,

most sales executives will ask for moresalespeople. This makes sense since sales-force effort drives sales. The ZS-SRA Sam-ple calibrated a more subtle result: Re-source allocation has a bigger impact onprofitability than sizing. The cliche “worksmarter, not harder” is appropriate. Forthe 50 companies in the ZS-SRA Sample, asize and resource-allocation strategy wasavailable that would produce, on average,a 4.5 percent contribution improvementover the company’s current or base casethree-year sales-force strategy. Only 29

percent of the incremental improvementwas attributable to a size change; the restwas due to resource allocation (Figure 6).Variation in Elasticities

Optimization theory tells us that a re-source is allocated optimally if the incre-mental returns are equal across all mar-keting entities (for example, products,markets, activities, and marketing-mix ele-ments) that want the resource. How closedo companies actually come to allocatingtheir sales-force effort optimally? We cal-culated sales-response functions for eachof the over 400 products promoted by the50 companies in the ZS-SRA Sample. As-

SALES-FORCE DECISION MODELS

May–June 2001 S15

Figure 5: The ZS-SRA Sample demonstrates that long-term profitability is flat over a largerange of sales-force sizes. Three-year discounted contribution varied between one and threepercent (95-percent confidence interval) for sales-force sizes that ranged plus or minus 20 per-cent from the optimal size. We normalized all sales-force sizes to 100 and the three-year dis-counted contribution to 1,000 to facilitate comparison across the 50 companies in the sample.Three-year sales appear as a dashed line and three-year discounted contribution as a solid line.

suming that they implemented their cur-rent base-case effort-allocation strategies,the ratio of the largest incremental returnto the smallest incremental return aver-aged more than eight. These companieswere not very effective in allocating sales-force effort. The ratio of the largest incre-mental return to the smallest incrementalreturn was 5.57 for the five products pro-moted by one company (Figure 7).Upsizing and Downsizing Rules

Decision makers in the ZS-SRA Samplewere reluctant to assume the risk associ-ated with increasing sales forces. The man-agement teams sized their company’s salesforces, on average, at 97 percent of the op-

timal size when measured from a three-year perspective but only 83 percent of theoptimal size when measured from a five-year perspective. They favored short-terminterests over long-term interests.

Decision makers used different ruleswhen increasing and decreasing salesforces. They stopped adding people whenthe incremental return on their sales-forceinvestment dropped below 50 percent.They could have continued to add peopleand increased profits but chose this risk-averse cut-off criterion. For downsizingdecisions, they required only a positive,incremental long-term discounted contri-bution. That is, they maximized profits.

SINHA, ZOLTNERS

INTERFACES 31:3 S16

Figure 6: The ZS-SRA Sample compares the impact of improved sales-force sizing and sales-force resource-allocation decisions on company contribution. The joint impact is 4.5 percentwith a 95-percent confidence interval of (3.5–5.5 percent); 3.2 percent was due to resource alloca-tion and 1.3 percent was due to sizing.

A 50-percent incremental return-on-investment criterion would have requiredmore severe reductions of the sales forces.If they had interchanged these cut-off cri-teria they would have expanded theirsales forces by more under favorable cir-cumstances and downsized them more inunfavorable environments.Concerns Beyond Models

In many instances, business processesapart from the model can manage thepractical concerns about the output of amodel. Sales-territory-alignment modelssearch the space of all possible ways ofgrouping accounts and geographies intobalanced territories for salespeople tocover. Trillions of potential alignments canbe developed even for small sales forces.Most companies’ alignments are far fromoptimal. Zoltners and Lorimer [2000]showed that 55 percent of sales territories

in a typical company are either too large(impossible to adequately cover all ac-counts) or too small (salespeople are wast-ing calls on low-value customers). Mostcompanies can improve sales and profitsthrough better alignment. Yet sales execu-tives are reluctant to revise their align-ments because they don’t want to dis-rupt current salesperson-customerrelationships.

To shed light on the sales impact of dis-rupting salesperson-customer relation-ships, we analyzed empirical data for anindustrial distribution sales force that hadrealigned its territories using an optimiza-tion model. We tracked monthly salesprior to and following the realignment.We identified two groups of accounts: atest group and a control group. The testgroup consisted of about 4,500 targeted ac-counts whose salesperson had changed.

SALES-FORCE DECISION MODELS

May–June 2001 S17

Figure 7: These five sales-response functions were derived for the five products sold by amedical-equipment company. The dots represent the planned sales-effort allocation and ex-pected sales. The incremental contribution for each product was calculated by applying theproduct margin to the incremental sales at the planned effort level. The calculated incrementalcontributions are for Product A, $164,381; for Product B, $69,245; for Product C, $385,696; forProduct D, $270,206; and for Product E, $198,068.

Test-group accounts had maintained a re-lationship with the same salesperson for atleast eight months prior to the realignmentand then a different salesperson for sevenmonths following the realignment. Thecontrol group consisted of approximately44,800 targeted accounts not affected bythe realignment.

We segmented accounts within eachgroup based on their annual purchasingvolume. We created six volume segments:extra-small-volume purchasers ($2,000 to$4,000 per year), small-volume purchasers($4,000 to $8,000 per year), medium-volume purchasers ($8,000 to $20,000 peryear), medium-large-volume purchasers

($20,000 to $50,000 per year), large-volumepurchasers ($50,000 to $100,000 per year)and extra-large-volume purchasers (over$100,000 per year). We tracked averagemonthly sales for each account segmentover a 13-month prealignment period anda seven-month postalignment period andthen compared results for the controland test groups.

During the prealignment period, themonthly sales trends for the two groupswere similar. During the postalignmentperiod, however, some differences be-tween the test and control groupsemerged: The large-volume purchasers($50,000 to $100,000 per year) in the test

SINHA, ZOLTNERS

INTERFACES 31:3 S18

group purchased 20 percent less thanthose in the control group. These resultswere statistically significant at the 95-percent confidence level. The estimatedtotal loss in sales at these disrupted ac-counts was approximately $2 million.

For those purchasing under $50,000 peryear and those purchasing over $100,000per year, there was no significant differ-ence in sales to control and test accountsin the postalignment period (Table 2).

The strength of the relationship betweenthe salespeople and their accounts pro-vides a partial explanation of the results.Salespeople did not have strong relation-ships with accounts purchasing under$50,000 prior to the realignment. A changein relationship, therefore, had little impacton sales to these accounts. At accountspurchasing over $50,000, however, sales-person relationships before the realign-ment were stronger, and hence, a change

had a significant impact. At accounts pur-chasing over $100,000, the sales force tookthe transition seriously. Exiting salespeo-ple introduced the new salespeople toeach account in transition. Both salespeo-ple would share account responsibilityand commissions for a specified period oftime. Because of this special attention,these accounts showed no sales loss.

In this case, a business process apartfrom the model solution prevented a lossin sales when salesperson-customer rela-tionships were realigned. The model re-sults were useful and compelling, but thefirm relied on a disruption-managementprocess for successful implementation.Corporate Sponsors Develop InsightsAbout Market Behavior Through theModel-Building and ImplementationProcess

The repeated application of several nor-mative sales-force-decision models has

Extra-smallaccounts

Smallaccounts

Mediumaccounts

Medium-largeaccounts

Largeaccounts

Extra-largeaccounts

Annual purchasing volume $2,000 to$4,000

$4,000 to$8,000

$8,000 to$20,000

$20,000 to$50,000

$50,000 to$100,000 $100,000 �

Total sales volume in millions(percent of total)

$22.1(2%)

$65.2(6%)

$220.4(20%)

$291.7(27%)

$182.4(17%)

$306.6(28%)

Was purchasing affected bya change in salesperson(95% confidence level)? No No No No Yes No

Did strong salespersonrelationships exist beforerealignment? No No No Somewhat Yes Yes

Was relationship-transitionprogram implemented? No No No No Somewhat Yes

Table 2: A study establishing the impact on salesperson-customer relationships after a majormodel-based sales territory alignment shows that disruption can be managed using businessprocesses apart from the model. An account transition program minimized the disruption forextra-large accounts for an industrial distributor.

SALES-FORCE DECISION MODELS

May–June 2001 S19

Current plan Expansion Reduction

Number of salespeople 100 150 50

Sales $100,000 $120,000 $70,000Cost of goods sold (20%) $20,000 $24,000 $14,000

$80,000 $96,000 $56,000Sales-force cost $10,000 $15,000 $5,000Other marketing cost $5,000 $5,000 $5,000Administrative costs $5,000 $5,000 $5,000

Pretax profit $60,000 $71,000 $41,000

Sales-force costs as percent of sales 10.0% 12.5% 7.1%Sales per territory 1,000 800 1,400

Table 3: This example shows why cost-containment approaches are not profit maximizing. Pre-tax profit and sales-force costs as a percentage of sales are negatively correlated for sales-forcesizes smaller than the profit-maximizing sales-force size.

produced a series of insights that havelead to a number of valuable sales-forceinsights.The Percent-of-Sales Rule

Cost containment approaches for sizingsales forces do not maximize profits. Manycompanies employ a cost-of-sales percent-age calculation for their sales-force-sizingdecisions. They like to constrain the ratioof sales-force costs as a percentage of totalsales to be smaller than a preset value. TheUS average is 6.8 percent. This heuristicignores the principle that sales-force effortdrives sales. Sales-force cost ratios andprofits are negatively correlated for sales-force sizes less than the profit-maximizingsales-force size (Table 3). Companies thatfavor small sales-force cost ratios tend toundersize their sales forces.Timing-of-Sales-Force Impact

Changes in sales-force size, structure,and resource allocation do not alwayshave immediate impact. Carryover is quitestrong in many markets. Managers whoover-promise when they ask for sales re-

sources will be disappointed because theconsequence of their strategy will takelonger to appear than they anticipate.Managers who reduce their investmentwill frequently be very pleased with theirdecision in the short-term because saleswill be impacted minimally for six to 18months. The impact accelerates with time,however.Phased Growth

Several years ago, a small pharmaceuti-cal firm acquired the rights to market twoproducts in the United States. The com-pany had paid close attention to its finan-cial performance. The sales force had toearn its way, and only significant salesgrowth would warrant expanding thesales force. The sales force grew as theproducts succeeded in their markets—oneeven became the market leader. Executiveswalked around with smiles on their faces.

The company left hundreds of millionsof dollars on the table because it didn’tlaunch hard. It built sales too slowly,didn’t take advantage of carryover, and

SINHA, ZOLTNERS

INTERFACES 31:3 S20

didn’t compete aggressively with latermarket entrants. Many model implementa-tions have shown that phased sales-forcegrowth is suboptimal (Table 4).Launch Hard; Protect Strengths

When launching new products, compa-nies should launch hard but also protecttheir existing products. This usually re-quires adding salespeople.

A successful new product launch usu-ally demands a sizable selling investment.Launching a new product or entering anew market can take 50 to 60 percent of asales force’s time, drastically curtailing thetime available to existing products andmarkets. Assuming that existing productswill retain their sales in the absence ofselling effort is dangerous. Sales may bemaintained due to carryover for a shortperiod but will suffer in the long run evenfor strong products. Many existing prod-ucts fail to make their sales targets whencompanies launch new products.

The only way to launch hard and pro-tect existing products is by expanding thesales force. However, this strategy has arisk of its own. The firm may need the ex-panded capability during only a shortstrategic window. Some firms look to alli-

ances or partnerships in these cases; otherstake a risk, increase their sales forces, andrely on attrition to reduce the sales-forcesize if capacity is needed for only a shortperiod of time. US sales-force-personnelturnover rates average about 20 percent.This implies that a firm can manage itsnew product and service opportunity byincreasing the sales-force size and then useattrition to systematically bring the field-force size back to a desirable steady-statelevel.Focused Strategies

Focused strategies dominate scatteredstrategies. One company’s sales plancalled for its 100 salespeople to spend timeselling all of its 37 products—“sell every-thing in the bag”. How much time couldeach product receive? An optimizationalgorithm recommended the sales forcesupport only eight products. Profit-maximizing strategies focus the firm’s re-sources on many fewer customers andproducts than marketing managers oftenrecommend.

We performed three types of analyseson a convenience sample of size and re-source allocation studies for 14 companiesand found support for focused strategies.

Sales-force size 1998 3-Year

Strategy 98 99 00 01 Sales Profit Sales Profit

A 300 380 380 380 330 83 1090 351B 300 350 380 380 315 84 1035 321C 300 320 350 380 290 87 970 301

Table 4: The short-term and long-term consequences of three expansion strategies for a salesforce that was launching several new products shows that phased growth can be suboptimal.Strategy A is the quick-build strategy and Strategy C is the slow-build strategy. Three-year prof-its for Strategy A are forecast to exceed those of Strategy C by $120 million, while one-yearprofits for Strategy C exceed those for Strategy A by $4 million. Companies often face trade-offs between short-term and long-term profit maximization when expanding their sales forces.

SALES-FORCE DECISION MODELS

May–June 2001 S21

Segmentation basedon volume

Segment

Percent ofcurrentsales

Percent ofaccounts

Currentallocationof effort

Optimizedallocationof effort

Company A 1 60% 3% 29% 36%2 37% 46% 51% 64%3 3% 51% 20% 0

Company B 1 94% 61% 77% 100%2 6% 39% 23% 0

Company C 1 76% 35% 52% 60%2 17% 28% 30% 31%3 4% 15% 7% 9%4 3% 22% 11% 0

Table 5: The output of an optimization model demonstrates the degree to which companiesneed to change how they allocate their resources across volume-based market segments if theywant to maximize profits. The three companies are business-to-business manufacturers.

The different analyses were performed be-cause the size and resource-allocationmodels developed for the 14 companieswere all different and consequently, theresource-allocation decisions were notcomparable.

An optimization model for threebusiness-to-business manufacturers recom-mended that they increase their focus onhigh-volume accounts (Table 5).

The target audience for the 18 largestproducts sold by the sales forces from fivepharmaceutical companies were also ana-lyzed (Table 6). The optimization modelsshowed that the number of physicians thatwere targeted by the companies should bedecreased by 39.3 percent for 13 of theproducts if the companies wanted to maxi-mize their profits. The five remainingproducts were very large products requir-ing a large customer reach. The optimiza-tion models recommended no reduction inthe target market for these products. Theaverage decrease across all products was

27 percent.Examining the products requiring the

greatest sales-force effort and determininghow many of them consume 50, 67, and 75percent of total sales-force resources is an-other way to assess effort concentration.The fewer the number of products, themore concentrated the effort allocation.Optimization analysis across six pharma-ceutical sizing and resource-allocationstudies showed that effort concentra-tion increased approximately 10 percent(Table 7).Selling Partnerships

Selling partnerships come in severalforms. They can be copromotions in whichseveral firms sell one or more products, orthey can be arrangements in which onefirm contracts with another organizationor series of organizations to provide sell-ing effort for its products. In either case,the selling investment that would opti-mize overall system profits exceeds thesum of the selling investments that would

SINHA, ZOLTNERS

INTERFACES 31:3 S22

Product

Percent change in thenumber of physicians thatneed to be called on(optimized vs. current strategy)

1 �68.52 �63.23 �61.14 �58.35 �48.66 �40.07 �37.88 �31.99 � 28.910 �14.911 �14.312 �9.913 �9.314 015 016 017 018 0

Table 6: The output of an optimization modeldemonstrates the amount of reduction in thenumber of physicians that need to be calledon if five pharmaceutical firms want to maxi-mize their profits for the 18 largest productsthat they promote. The reduction ranges be-tween 0 and 68.5 percent and averages 27 per-cent.

maximize the profits of the individualpartners.

The best strategy for each partner is toallocate selling effort across customers andprospects until the marginal return of theselling effort is equal to the marginal costof the selling effort. The depth of penetra-tion depends upon product margins. Withhigher margins, a firm can call on morecustomers and prospects before its callsbecome unprofitable. Product margins areshared when organizations enter into part-nerships giving each partner only a frac-tion of the total margin. Each partner will

stop calling on accounts and prospectsthat would have been profitable with thehigher margin. Neither partner will in-crease its effort unilaterally beyond its de-sired level even though the consortium’sprofits would increase by calling on morecustomers.Realignment Overlooked?

Roughly 55 percent of sales territories inthe US are either too large to be coveredadequately or so small that calls arewasted. How does this happen?

There are trillions and trillions of goodalignments. Finding a good alignment ofterritories is a large combinatorial problemthat requires lots of effort. The task is of-ten delegated to local district sales manag-ers burdened with many other responsi-bilities. Even if the district managers doprovide good alignments, the global align-ment will be poor if the district bound-aries are not optimally designed or if thedistricts are sized incorrectly.

A second reason for poor alignments isthat companies seldom have good defini-tions of what constitutes a good align-ment. Some authors suggest that a goodalignment is profit maximizing [Lodish1975; Skiera and Albers 1996; Zoltners1976]. Most managers feel that it shouldbe disruption minimizing. In practice,good alignments balance territory work-load. The best workload measures are cali-brated to account for territory marketpotential.

Third, sales forces resist change. Manyfirms retain poor alignments because salesmanagers want to avoid the risks associ-ated with reassigning accounts amongsalespeople and because salespeople pressto service their established account

SALES-FORCE DECISION MODELS

May–June 2001 S23

Number of products receiving . . .

50% of thesales-force effort

67% of thesales-force effort

75% of thesales-force effort

Current allocation 21 32 40Optimized allocation 19 29 35Change �10% �9% �13%

Table 7: The number of products that receive 50, 67, and 75 percent of total sales-force effort forsix pharmaceutical firms were examined to see if effort concentration would improve their prof-itability. Products were ranked in terms of the effort that they received. The ones receiving themost effort were counted until they comprised 50, 67, and 75 percent of the total effort. Thefewer the number of products, the more concentrated the effort. The number of products ineach column is the total across all six studies. The number of products selected by the opti-mizer in each of the six studies was either the same or smaller than the number of productssold with the current strategy for each of the three effort categories.

relationships as opposed to establishingnew ones. Large realignments can requirethat salespeople and sales managers relo-cate. Sales executives fearful of losing goodpeople are reluctant to call for relocations.Finally, realignments can change reportingrelationships. Good manager-salespersonrelationships may be severed. All of thesechanges cause stress in the sales force.

Fourth, veteran salespeople obtain fa-vorable alignments. They know that terri-tory potential correlates highly (0.4 � r �

0.8) with territory sales—high-potentialterritories usually have the highest sales.Veteran salespeople do whatever it takesto get and protect desirable accounts, es-pecially when their compensation has alarge incentive component. New salespeo-ple either leave or learn how to play thegame. Sales managers find it hard to takeaccounts away from veteran salespeoplewho resist the change and whose signifi-cant sales help the manager achieve his orher sales goal.

Finally, sales executives are not alwaysaware of the value that results from goodalignment. Empirical studies [Zoltners and

Lorimer 2000] show that sales increase be-tween two and seven percent when salesterritories are aligned to optimize cus-tomer coverage (Figure 8).

Travel times are reduced by 10 to 15percent when models are used to alignsales territories.

Since territory potential correlates sohighly with territory sales, companies withhigh incentive components in their com-pensation plans tend to overpay peoplewith rich sales territories when territoriesare not aligned properly. Rewards andperformance evaluation systems are muchfairer when sales territories are balanced.No Plan Pleases All

Every salesperson has an opinion of hisor her compensation plan. Some wouldlike to see more salary; some would likehigher incentives; some would like to getincentive income before they hit the terri-tory goal; some feel that caps are de-motivating; some feel that team incentivesfeed free-riders; some would like more fre-quent payouts; some feel that they are re-warded in terms of factors that they can-not control; but almost everyone feels that

SINHA, ZOLTNERS

INTERFACES 31:3 S24

Figure 8: The impact of alignment improvement can be assessed by analyzing the relationshipbetween territory sales and territory market potential. The dots in the graph represent 25 salesterritories comprising a pharmaceutical sales region. The sales potential of each territory ismeasured by adding the sales of the company’s products to the sales of all competing products.The data show the positive relationship between territory potential and territory sales whilethe fitted curve reveals that this relationship has diminishing returns. This is because as poten-tial increases, territory workload begins to exceed a salesperson’s capacity, and sales opportuni-ties will be lost because it is impossible to cover all accounts effectively. Because the relation-ship between sales and market potential shows diminishing returns, the sales lost by reducinglarge territories will be more than offset by sales gained by increasing small territories. Wehave performed this analysis for many sales forces and find a typical predicted net incrementalgain in sales between two and seven percent.

his or her income should be larger.Many compensation models assume

sales forces are comprised of people withhomogeneous preferences. This is nevertrue. The advertising sales force at a largeUS newspaper was evaluated to assess thedegree of preference heterogeneity that ex-ists in practice. The 42-person sales forceevaluated seven plan options (Table 8).

None of these plans will satisfy all 42salespeople (Table 9). Every sales forcethat we have worked with has exhibited

similar heterogeneity in its preferences.Hidden Salary

A large producer of business forms de-cided that its sales force lacked motiva-tion. Its compensation plan was mostlysalary, and its managers thought that in-creasing the variable component would in-crease motivation. They decided that alow salary plus a commission structurestarting with the first dollar would reallyrev up the sales force. Compensation mod-eling showed that a commission structure

SALES-FORCE DECISION MODELS

May–June 2001 S25

Thresh-old 1(payouttrigger)

Rateabovepayouttrigger($/%)

Thresh-old 2

Rateabove thesecondthreshold($/%)

Thresh-old 3

Rateabove thethirdthreshold($/%)

Incentivepay atgoal

Indi-vidualweight

Depart-mentweight

Current 95% $2,500 100% $4,000 115% $2,500 $12,500 60% 40%Option 1 90% $2,000 95% $3,000 105% $2,000 $25,000 60% 40%Option 2 80% $1,000 90% $1,500 110% $1,000 $25,000 60% 40%Option 3 70% $667 85% $1,000 115% $667 $25,000 60% 40%Option 4 90% $2,000 95% $3,000 105% $2,000 $25,000 80% 20%Option 5 80% $1,000 90% $1,500 110% $1,000 $25,000 80% 20%Option 6 70% $667 85% $1,000 115% $667 $25,000 80% 20%

Table 8: To assess the level of preference heterogeneity, seven plan options were rank-orderedby members of a newspaper advertising sales force. The plans differed in three features: pay-out trigger, rate of payout, and the percentage of payout attributable to individual (versusteam) performance. The payout trigger was the percentage of goal attainment where the incen-tive payment was to begin. It could begin at 70, 80, 90, or 95 percent of the territory goal. Therate of payout was the amount of money to be earned for each percentage point above the trig-ger and subsequent threshold levels. The total incentive was partitioned into a departmentcomponent and an individual component. The department component was tied to the successof the entire selling team while the individual component was tied to the individual’s perfor-mance. The individual and department weights specified how the total incentive payout wouldbe split between these two components.

could be derived that would reward sales-people at the appropriate level so that theywould not lose income with the new plan.

We analyzed the firm’s carryover struc-ture and observed that customer-switchingcosts were very high. Someone using thecompany’s forms would face significantcosts in migrating to another producer’sforms. The sales force was in a mainte-nance mode. About 80 percent of thefirm’s sales were secure even if salespeo-ple did little selling. Consequently, thehigh-incentive commission plan wasmostly rewarding sales that salespeoplecould bank on. Modeling efforts in highcarryover environments reveal that highlyleveraged commission plans rarely put alot of pay at risk.The Share Growth Relationship

The assumption that it is harder to grow

market share in high-market-share territo-ries than in low-market-share territories isfrequently not true. Such assumptions canlead to unfair incentive plans. We sur-veyed a convenience sample of 91 firms in32 industries and found that 84 percent ofsales-force incentive plans are goal based.Sales managers setting goals for theirsalespeople believe it is easier to increasemarket share in low-market-share territo-ries than in high-market-share territories.The analysis of territory-level data sug-gests that this assumption is frequentlyfalse (Figure 9). An excessive increase inquotas for low-market-share territorieswill penalize those salespeople who havethese territories.Poor Goals Cost Money

About 84 percent of sales forces usegoals in developing their incentive com-

SINHA, ZOLTNERS

INTERFACES 31:3 S26

pensation. Many take goal setting forgranted. In fact, we rarely see statisticalmethods used for this purpose. The fol-lowing examples show that goal settingshould not be taken for granted; doing itpoorly can add unnecessary costs for asales force.

A large telecommunications companyintroduced a new product line. It devel-oped an aggressive compensation plan tomotivate the sales force. The line did ex-ceptionally well, providing the sales forcewith a windfall. Most salespeople attainedtheir quota by midyear and almost tripledtheir sales forecast by the end of the year.They made a lot of money. Managementwas pleased because its stock options be-came very valuable but did not realize ithad created a compensation nightmare.The company paid the sales force toomuch because it did not forecast sales ac-curately. The product line was terrific. Theproduct sold itself, but the sales force tookinappropriate credit for its sales. Not onlythat, management created an expectationamong the salespeople that they wouldcontinue to earn an income well above the

market rate.In another example, a health-care com-

pany was expecting a sales decline. Hop-ing to motivate its salespeople, it estab-lished stretch goals for them. By October,all the salespeople were certain theywould get no incentive pay because theirgoals were too high. Some of the best quit.Others stopped working and deferredsales into the next year in the hope thatgoals would be more realistic.

These examples point out the two detri-mental consequences of poor goal setting:Salespeople get free money if the goal istoo low and lose motivation if the goal istoo high.

The next simulation demonstrates howexpensive goal-setting errors can be. Inthis example, a company uses an incentivecompensation plan (Figure 10) in which itpays 2.5 percent commission on sales upto a territory goal and pays 7.5 percentcommission on sales that exceed this goal.This is a growth-oriented compensationplan in that it aggressively rewards extraeffort at a point where extra effort is re-quired. If the salespeople beat the quota,

Current Option 1 Option 2 Option 3 Option 4 Option 5 Option 6

1’s 2 14 8 4 9 6 02’s 5 7 10 4 6 8 13’s 5 10 12 3 5 5 34’s 7 1 6 7 5 7 85’s 4 7 3 4 6 7 116’s 0 3 3 15 6 8 77’s 19 0 0 5 5 1 12Average 5.0 2.7 2.9 4.6 3.7 3.7 5.3

Table 9: Significant heterogeneity in preferences emerged from the rank-order preference sur-vey administered to the newspaper advertising sales force. The rows in the matrix correspondto the respondent’s preference levels, where a 1 is most preferred and a 7 is least preferred. Thecells in the matrix represent the number of times that each preference level was chosen for eachof the compensation options. For example, 14 salespeople ranked plan option 1 as their firstpreference.

SALES-FORCE DECISION MODELS

May–June 2001 S27

Figure 9: This relationship between market share and market-share growth for a pharmaceuti-cal firm demonstrates that it is dangerous to assume that low-market-share territories havegreater opportunity to grow than high-market-share territories. Each dot represents a sales terri-tory. The territories are plotted in terms of their market share at time t and their change inmarket share from t to t � 1. R2 � 0.00002 for this group of territories.

they and the company benefit.In this example, the plan assumes that a

sales territory should sell $1,000,000 if thesalesperson puts forth a fair day’s work.The perfectly accurate territory goal wouldbe $1,000,000. Quotas above or below$1,000,000 constitute a goal-setting error.An inaccurate goal-setting process wouldhave a wide distribution around$1,000,000, while an accurate goal-settingprocess would have a small variancearound this sales level (Figure 11).

Assume that all the sales territories cansell $1,000,000. The incentive payout de-pends upon the accuracy of the goal-setting process. Incentive costs can be 24percent higher than necessary in the low-

accuracy case (90 percent of all territorygoals are plus or minus 50 percent of theperfectly accurate territory forecast). In-centive costs can be nine percent higherthan necessary in the moderate-accuracycase (90 percent of all territory goals areplus or minus 20 percent of the perfectlyaccurate territory forecast) (Figure 12).Companies that don’t set accurate goalsincur higher sales-force costs.

To put things in perspective, the incen-tive portion averages 40 percent of the to-tal income for a US salesperson and theaverage annual income for an experiencedsalesperson exceeds $100,000. US sales-force costs exceed a trillion dollars ayear.

SINHA, ZOLTNERS

INTERFACES 31:3 S28

Figure 10: A growth-oriented commission plan is used in the costing simulation. The commis-sion rate up to the territory sales goal is 2.5 percent and 7.5 percent above this goal. The per-fectly accurate goal for each sales territory is set at $1,000,000.

Figure 11: The error associated with goal set-ting can be represented as a distributionaround the perfectly accurate territory goal.Inaccurate goal-setting processes will have awider distribution (solid line) than more accu-rate goal-setting processes (dashed line).

Insights for Sales ManagementThe key insights from the ZS-SRA, ZS-

TA, and ZS-IC Samples combine to forman advisory for sales management. Theycan be organized around three topic areas.

The key insights for sales-force size,structure, and resource allocation are thefollowing:—Sales-force effort drives sales.—Carryover exists. Sales-force size is un-derstated if its effect is not considered.—The impact of changes in sales-forcesize, structure, and resource allocation isnot always immediate.—Company profitability can be flat for awide range of sales-force sizes.—Cost-containment approaches for sizingsales forces are not profit-maximizingapproaches.

SALES-FORCE DECISION MODELS

May–June 2001 S29

Figure 12: The impact of goal-setting error on the incentive plan costs for the example in Figure10 can be calculated. The cost is $25,000 per territory if the forecast is 100 percent accurate. Thegraph represents the incentive cost when 90 percent of all of the territory goals are plus or mi-nus the value on the horizontal axis. For example, incentive costs can be 24 percent higher thannecessary in the low-accuracy case (90 percent of all territory goals are plus or minus 50 percentof the accurate forecast).

—Resource allocation is more importantthan sizing.—The largest product elasticities can beeight times larger than the smallest prod-uct elasticities.—Phased sales-force growth is rarelyoptimal.—When launching new products, launchhard but also protect strengths. This usu-ally requires adding salespeople.—Focused strategies dominate scatteredstrategies.—Products and services receive insuffi-cient resources when multiple firms enterinto a selling partnership.—Sales-force executives tend to be ori-ented toward the short term and risk

averse when an increase in the sales forceis warranted and protective when down-sizing is necessary.

The key insights for sales-territory align-ment are the following:—Most sales territories (55 percent) are ei-ther too large or too small.—Good sales-territory alignment enhancescustomer coverage and increases sales.Sales will increase two to seven percentwhen sales territories optimize customercoverage.—Sales-territory alignment affects perfor-mance evaluation and rewards. Compa-nies overreward territories and under-reward salespeople when alignments arepoor.

SINHA, ZOLTNERS

INTERFACES 31:3 S30

—Good territory alignment reduces traveltime.—The cost of disrupting a sales forcethrough realignment can be managed witha relationship-transition program.—Software for optimizing and refiningterritory alignments saves sales managershundreds of hours of tedious manualwork.

The key insights for sales-force compen-sation are the following:—No compensation plan satisfies every-one in a selling organization.—Commission plans in high carryover en-vironments contain hidden salary.—The assumption that it is much harderto expand market share in high-market-share territories is frequently not true.Such assumptions can lead to unfair incen-tive plans.—Inaccurate goal setting is expensive.Modeling Provides a Benchmark forHow Much Improvement is Possible

When executives review their salesforces, these questions come up repeat-edly: Is our sales force the right size? Howgood is our compensation plan? Are wedeploying our resources the right way?How good are our current decisions? Ourmodel implementations reveal how muchimprovement is possible over currentpractice.

The model-based analyses for ZS-SRASample companies provide a benchmarkfor the improvement that is possible by re-sizing a sales force and deploying its salesresources differently. The sample showedthat on average firms can make a 4.5 per-cent improvement in three-year profitabil-ity. About 28 percent of the improvementcan be attributed to sizing.

The ZS-TA Sample suggests that about55 percent of all sales territories are toolarge or too small. We classify territoriesas too large if the salesperson’s workload

The estimated total loss wasapproximately $2 million.

is more than 15 percent above the idealworkload and too small if his or her work-load is more than 15 percent below theideal workload. The percentage of territo-ries that were too large or too small wasreduced to 26 percent after we used amodel-based alignment process to developan acceptable alignment for companies inthe ZS-TA Sample.

Most companies try to develop incentivecompensation plans that pay their sales-people for performance. The degree towhich they achieve this goal is measur-able. Salespeople for a large industrial dis-tributor were plotted in terms of their per-formance and incentive payout (Figure13). Salespeople in the Northwest quad-rant were overpaid—they received anabove-average incentive payout for sub-standard performance. Those in the South-east quadrant were underpaid—they re-ceived below-average incentive payout forabove-average performance. Of the 1,400salespeople, 146 were underpaid and 21were overpaid. These statistics and thecorrelation coefficient serve as metrics forthe degree to which an incentive compen-sation plan truly pays for performance.

The ZS-IC Sample provides an estimateof the degree to which companies overpayunderperformers and underpay over-achievers (Table 10). Companies appar-ently favor underpaying high performers.

SALES-FORCE DECISION MODELS

May–June 2001 S31

Figure 13: The relationship between the incentive payout and actual performance can estimatethe degree to which a company has a pay-for-performance incentive plan. The dots represent1,400 salespeople selling for an industrial distributor. We defined sales performance as the dif-ference between actual sales and the expected sales derived from a regression model that incor-porated the prior year’s sales. Both performance and incentive payout are plotted in standarddeviations above and below the mean. One hundred forty-six salespeople had above-averageperformance and below-average incentive pay while 21 salespeople had below-average perfor-mance and above-average incentive pay.

This may help explain the high turnoverrates that face some selling organizations.An improved compensation plan couldbetter balance performance and pay.

Our conclusions here summarize manystudies. We can usually measure howmuch a firm can improve by changing itssales-force size, resource allocation, terri-tory alignment, or incentive compensationplan after implementing a model custom-ized for the company.Implementation Insights

The economic value of a model cancome from many possible sources, such asreduced uncertainty, increased speed, ac-

curacy, objectivity, and stakeholder in-volvement. Model-centered thinking fre-quently focuses on the quantifiable.Models parameterize how customers andmarkets respond to different marketing-mix decisions. They maximize measures,such as sales, profits, and market share.However, models provide other, more-qualitative sources of value as well.

Mergers and acquisitions require quickintegration of multiple selling teams. Anunbiased view is desirable since most inte-gration decisions are contentious. Modelsprovide both objectivity and speed.

Models help reduce uncertainty. Many

SINHA, ZOLTNERS

INTERFACES 31:3 S32

Total number

Number ofsalespeople getting

Percentage ofsalespeople getting

Industry of salespeople overpaid underpaid overpaid underpaid

Animal health 66 13 18 19.7% 27.3%Consumer health 90 18 10 20.0% 11.1%Pharmaceutical 61 9 10 14.8% 16.4%Pharmaceutical 75 6 14 9.8% 23.0%Newspaper advertising 87 6 19 6.9% 21.8%Industrial distribution 1,400 21 146 1.5% 10.4%Eye care 68 3 7 4.4% 10.3%

Weighted average 262 10.9 32.0 4.1% 12.2%

Table 10: A study of seven companies shows the extent to which salespeople are over- orunderpaid with their current compensation plans. We define a salesperson to be overpaid if heor she is a below-average performer while receiving above-average incentive pay and under-paid if he or she is an above-average performer while receiving below-average incentive pay.

incentive compensation model builderstest their potential plans before making arecommendation, looking at total cost andhow much each salesperson is likely tomake under each plan scenario. The testsreduce the risk of spending too much oncompensation and of overpaying or un-derpaying individual salespeople.

One multinational organization devel-oped a model-assisted process for annualmarketing planning for various countries.A major benefit of the process was the dis-cipline and uniformity required from thecountries. Management was able to com-pare solutions, scenarios, and plans acrosscountries.

Large-stake decisions, such as resizingand restructuring a sales force, require topmanagement involvement. Models that ex-plore environmental assumptions and de-velop alternative solutions are very engag-ing to executives, promoting theirparticipation in the decision making andtheir confidence in the results.

What makes the model valuable canhelp the modeler establish the scale, com-

plexity, and nature of the modeling effort.For example, if speed is critical, or if train-ability and flexibility are important andthe process is to be used across manycountries, the models need to be simple.The Invisible Model

The best model is often one the decisionmaker never sees. A variety of people par-ticipate in implementing the typicalmodel: a sponsor, decision makers, themodel creator, and the project team.

The sponsor, usually the president of adivision or a country manager, insuresthat the project has sufficient corporatepriority and human and financial re-sources. Acting as a change agent, thesponsor has the leadership and authorityto get the project results implemented.

Many authors of Management Science ar-ticles talk about “the decision maker” asthe model user. We have found that deci-sions are usually made by a group of peo-ple. The decision makers could include theproject sponsor, the vice president of sales,and several other top-level sales and mar-keting managers.

SALES-FORCE DECISION MODELS

May–June 2001 S33

The model creator creates a generaliz-able model that can be used for many situ-ations or companies, does not participatein all model implementations, but is usu-ally available to customize the model ifnecessary.

The project team is comprised of severalmembers. The project manager assemblesthe team, plans the project steps, delegatesresponsibilities, guides the team, and is ac-countable for completing the project suc-cessfully and implementing the results.

The math was right, but theanswer was wrong.

The project team is responsible for doingthe work. One member, the model builderand user, is responsible for using themodel. He or she specifies the model in-puts, parameterizes the model, and orga-nizes the model outputs. The model userinteracts continuously with other membersof the team.

The several roles are often distinct. Thedecision makers, in particular, rarely per-form any downstream functions. They donot have the narrow, deep perspective andskills to be effective model users. Decisionmakers are valued most for their intuitionand judgment. To use complex modelsthat combine factual and judgmental in-puts, one needs to combine skills that aninfrequent modeler can’t develop andcontext-specific judgment that a modelercan’t have.

Decision makers have diverse perspec-tives and capabilities. In the 2,000 studiesconcerning sales-force size and deploy-ment that we have done, the decisionmaker has never also been the model user.

In at least 80 percent of the cases, the proj-ect manager and the model user were dif-ferent, and in over 70 percent, the modeluser and model creator were different. Wefind that the combination of context or is-sue experts and modeling experts creates apowerful synergy and that this kind ofspecialization is a dominant approach. Analternative is to design the model to be di-rectly accessible to the decision maker.Our narrow but extensive experience hasbeen skewed toward the call-the-doctormodel and not the here-is-a-car drive-itmodel.Perspectives: Theory and PracticeThe objectives of marketing modeling in-clude the following:—Develop a precise and complete under-standing of market behavior.—Create rigorous and sophisticated esti-mation and optimization techniques.

The objectives of marketing modelingpractice include the following:—Solve the right problem.—Build realistic models.—Build adaptive models.—Generate implementable solutions.—Get it done quickly.

In the ideal world, theoretical model ad-vances will continually feed the practicalmodeling needs of real-time decisionmakers.Right Math, Wrong Answer

In an assessment of sales-force size andresource allocation in France, we built amodel that suggested that a firm could im-prove sales and profits by significantly in-creasing investment in a specific product.When we communicated this result to theproject team, the members threw up theirhands in bewilderment. Since none of us

SINHA, ZOLTNERS

INTERFACES 31:3 S34

understood French, we spent half an hourdeciphering what they were concernedabout. The product was a memory en-hancer, popular with high school studentsstudying for their exams! The medicalbenefits were dubious, and it would beprofessionally inappropriate to throwheavy investment at such a product. Allthat the model focused on was the highsales response to sales-force activity forthis product. The math was right, but theanswer was wrong.

In other situations, the diagnosis ofwhat needs fixing can be wrong. Thisleads to addressing the wrong problem.We have often been involved in “incentivecompensation projects” only to discoverthat the problem lies elsewhere. Aninappropriate sales-force structure orperformance-management system may bethe culprit. For example, if a sales force isspending too little time with some prod-ucts or customers, a manager’s first incli-nation can be to boost the incentive pay-out for these products or customers. But

Our team members foundtheir PCs locked up.

the real solution may be to create a spe-cialized sales force focusing on the prod-ucts that are difficult to sell. Changing in-centives may not be enough. The incentivemodel can give a great answer but itdoesn’t solve the real problem.Is Wisdom Better than Models?

Model developers find fertile groundwith some issues and environments, forexample, sales-force analysis in the phar-maceutical industry. The industry has a lotof valuable and reliable data that can be

used to calibrate models for sales-force siz-ing, resource allocation, and territoryalignment. The data is tracked at the phy-sician level and includes monthly pre-scriptions for every product that everyphysician uses, as well as monthly call ac-tivity and other promotional programs di-rected at physicians. Sizing, resource-allocation, and territory-alignment modelshave had a big impact in this industry.One of the earliest documented exampleswas a project done for Syntex [Lodish etal. 1988]. Syntex’s sales force was grosslyundersized and its main product was un-derfunded. By departing from its currentstrategy, Syntex greatly improved profits.The modeling effort convinced Syntexmanagement to make important changes.

Models provided financial projectionsfor different strategic and tactical alterna-tives. Financial people love them. Deci-sions go beyond intuition, and salesmanagers can be held accountable forachieving the targets established with thehelp of the models.

However, it is more difficult to developeffective models for other sales-force deci-sions. The complexity inherent in compen-sation decisions, for example, makes itvery difficult to model well (Figure 14).

Managers make many interdependentdecisions. Individual components can bemodeled, but it is very difficult to developa practical comprehensive model. An ex-perienced compensation consultant whoknows what works and what does notwork will find all of the foibles associatedwith a model-based solution.

Lack of measurability is another factorthat hinders effective model developmentand implementation. For example, some

SALES-FORCE DECISION MODELS

May–June 2001 S35

useful criteria for evaluating sales-forcestructure choices are the following:—For customers, the choice should pro-vide responsiveness and continuity.—For employees, the choice should pro-vide clarity, meaningful work, and lowuncertainty.—For the firm, the choice should enhanceefficiency and effectiveness while main-taining adaptability.

Many of these criteria are not measur-able or are difficult to measure. Marketingdecision models require measurability forsuccess. Consequently, managers mustrely on wisdom and good judgment tomake sound decisions about sales-forcestructure.

The two dimensions of complexity andmeasurability help define where market-ing decision models can be implemented(Figure 15). Simple models are appropriatewhen measurability is high but complexityis low. Such problems are amenable tomodels that are easy to build, control, andunderstand. In situations in which com-

plexity is high and measurability is low,experienced managers will dominate mod-els. Processes to organize the managers’judgment are very useful in this decisioncategory.

Various sales-force issues are amenableto modeling to different degrees (Figure16). Determining sales-force structure islargely qualitative and judgment driven.However, models can be used to deter-mine size and deployment in data-richindustries but not in data-challengedindustries.

In designing incentive plans, simplemodels can be used to estimate the cost ofpotential incentive compensation plans.But developing a sales-force incentive-compensation structure depends more onthe use of judgment and experience (Fig-ure 17). Sophisticated modeling methodsare required to assess sales-force prefer-ences and the impact on sales of alterna-tive plans.The Model is the Tip of the Iceberg

Early in our modeling careers in the

Figure 14: Many key decisions must be made to determine an effective compensation plan. Theincentive-payment-design wheel enumerates the decisions that need to be made to establish aperformance-payout relationship. Such complex decisions can be assisted by models but can’tbe solved using models.

SINHA, ZOLTNERS

INTERFACES 31:3 S36

1970s, our thinking was centered on mod-els and we believed that the model was alarge and prominent part of solving sales-resource-optimization problems. Over theyears, we have realized that we spendmuch more energy on other activities,such as articulating the issues, buildingdatabases, and dealing with change man-agement and implementation. For exam-ple, in the geographic deployment workwe have done, we spend over 95 percentof the time in activities unrelated to modelbuilding.

In realigning sales territories, we fre-quently use a six-step process (Figure 18)with the alignment objectives tailored to

the individual sales force. This process issimilar to those we use in decisions aboutsales-force size, structure, resource alloca-tion, and compensation.The Roles of Models and People

Models provide insights; people makedecisions. Arriving at a decision and im-plementing it tends to be a multistage pro-cess in which people use models to gaininsights but are really in charge ofdecisions.

Years ago, we did a sales-force-sizingstudy for an international subsidiary of alarge company. We developed manywaves of sizing scenarios as we exploredpossible states of the future environment.

Figure 15: The role of wisdom and models differs for different levels of measurability andcomplexity. Models are useful in moderate-to-high measurability situations, and the issue com-plexity drives model complexity. Experience and wisdom are useful in high-complexity, lowmeasurability situations.

SALES-FORCE DECISION MODELS

May–June 2001 S37

The results showed that the firm shouldreduce its sales force even under the mostoptimistic assumptions. Returning fromlunch one day, our project team membersfound their personal computers locked up.The project sponsor had locked the roomto prevent the team from evaluating anymore scenarios. He did not like the direc-tion of the results, even though the localsales and marketing managers expected asales-force reduction. A month later thesponsor presented a recommendation tointernational headquarters that the firmmaintain the current sales-force size. Mod-els provide insights, but people make thedecisions.

Having worked with some managers re-peatedly over a decade or more, we haveobserved patterns in the ways managersuse consulting assistance and models.

Most begin with an intuitive feeling forthe answer: “For the new product launch,I think I may need another sales team,” or“There are too many generalists in this or-ganization. We need to specialize.” Theythen use model assistance to quantify thesituation, to give precision to the answer,and also to set an expectation of the salesand profit results associated with a changein direction. The decision makers influencethe role that the model will play in the de-cision making. Some are completely open-minded, whereas others begin with preciseexpectations of the ultimate decision.Project Examples

Four examples follow of the types ofprojects from which we gained the in-sights we have discussed. In them, weorganize our descriptions around threetopics:

Figure 16: A complexity-measurability characterization of sales-force structure, size, and de-ployment problems: Sales-force-structure decisions are complex and difficult to quantify. Sizeand deployment decisions are quantifiable in some industries but not in others.

SINHA, ZOLTNERS

INTERFACES 31:3 S38

—Background,—Project description, and—Results, benefits, and insights.Example 1: Ensuring Speed andConsistency in a Merger

How do you make a giant fleet of foot?Two major companies faced this challengeas they moved toward a merger. Theyneeded to integrate the sales and market-ing strategy of two huge organizationswith over 15,000 sales representatives anda presence in every developed country inthe world, and they needed to do itquickly to return the company focus to thecustomer. The new company set itself theambitious target of developing a sharedstrategy across its 40 largest affiliates in

two months.With this tight deadline, the project

team would have little time to communi-cate in detail across 40 affiliates once themerger was underway. It would have toensure consistency of purpose and ap-proach from the outset, reflect specificlocal issues, and ensure that local manage-ment teams supported and implementedthe answer.

We designed a project structure basedaround two key elements—an integratedstrategy-development model to ensureconsistency, and a team-communicationstructure to allow for course correctionsand local flexibility. The strategy-development model was structured

Figure 17: A complexity-measurability characterization of different components of incentive-plan-design problems: Incentive-structure problems are complex and have low measurability,whereas estimating the cost of incentive plans is a quantifiable problem, amenable to model-ing. Other incentive issues are in between.

SALES-FORCE DECISION MODELS

May–June 2001 S39

around four stages.In the first stage, we used a common

framework to capture the existing strate-gies of the two organizations, formulatedbefore the merger. This enabled the projectteams to define strategies in a languageshared by both halves of the new organi-zation and to challenge existing assump-tions. This baseline was a stable point ofreference during a very volatile period,and would later form the basis for com-parison with postmerger strategies.

We then focused on what would hap-pen if the firm did something different.Clearly the new, larger company had farmore promotional options than each com-pany had had individually. By framingthis analysis around an econometric modelstructure, we were able to share knowl-edge and insights across joint product-category teams very efficiently. The detailsof the analysis supporting this modelingvaried from country to country to reflectthe local environment, but the core model

structure remained common.We also identified duplicated activities,

shared customer interests, and commonchannels that could be sources of cost syn-ergies. We also targeted activities thatwere driven by specific products but thatincreased investment needs.

In the fourth stage, the firm based its fi-nal strategic decisions on an econometricevaluation of the incremental return onpromotional investment. We consistentlyapplied this metric throughout the organi-zation to find the most valuable use ofeach dollar that the firm invested in salesand marketing. With a shared format foranalyzing the promotional strategies, com-parison and assimilation at corporateheadquarters was rapid.

A team that was structured to ensurehigh quality and consistency implementedthis process at the global level while en-abling local project teams to adapt to localcircumstances. A few senior project man-agers spanned projects in several countries

Figure 18: A typical alignment generally follows a six-step process, but the exact objectives,activities, and participants are tailored to the specific situation.

SINHA, ZOLTNERS

INTERFACES 31:3 S40

and communicated with the corporatesteering group and with each other.

Using this integrated process, the com-pany developed a worldwide integratedstrategy within two months and deliveredcost and sales synergies. In over 300 meet-ings, managers around the globe came to-gether to focus on a common goal, and inthe process, they built their knowledgeand their teams. Furthermore, becausethey did this so quickly, it became the firstmajor merger in the industry in which afirm increased its market share during amerger.

The company firmly believed that speedwas key to the success of a large merger:“To be fast is more important than to be

Models provide insights, butpeople make the decisions.

accurate.” By driving through this inte-grated strategy model, the companysucceeded in being large and fast andaccurate.Example 2: Creating a VirtuousModeling-Learning Cycle

We worked very closely for over 10years with a pharmaceutical companythrough several restructurings and merg-ers. During this period, the firm launchedmany products with varying degrees ofsuccess. It also made many changes in topand middle management. Like others inits industry, this company invests heavilyin purchasing and organizing data aboutits customers, competitors, and markets.The value of the investment is routinelyquestioned, but the value is difficult to es-tablish. In spite of the data investment,managers still made most major decisions

subjectively.Beginning with projects on specific is-

sues, we developed a relationship with thecompany whereby we created a systematicprocess for aiding decisions on sales-forcesize and structure, customer targeting,segmentation, marketing-resource alloca-tion, and new-product-launch-strategydevelopment.

We created a sales and marketing datawarehouse, a centralized repository of cus-tomer, competitor, market, and activity in-formation, and established processes forcollecting, cleaning, verifying, and orga-nizing the information to make access easyfor sales and marketing personnel.

We created a sales decision-support sys-tem to help sales representatives and salesmanagers gain access to information andto make information-based decisions.

The company also uses a systematicresource-planning process annually to as-sess the market responsiveness of all thecompany’s brands and customer segments.It combines the results with judgmentaldata to determine investment levels and toallocate sales and marketing resources.

In addition, the company uses models tosupport periodic decisions. The headquar-ters team uses information to evaluatenew-product-launch strategies and part-nership opportunities and to help it makeseveral tactical and strategic decisions. Forexample, the company uses a market-monitoring mechanism to track new-product launches. As a result, it has madeseveral course corrections that ensuredproduct success.

The core benefit for the company hasbeen its creation of a culture ofinformation-based decision making. As

SALES-FORCE DECISION MODELS

May–June 2001 S41

the headquarters group learned the impor-tance of gathering objective informationand analyzing data, it exported the prac-tice to the sales organization.

Today the company plans better forlaunching new products than it used toand is also more targeted, aggressive, and

The managers saw them aspoisoning the answer.

successful in its launches. The entire orga-nization uses a scorecard that includesqualitative and quantitative measures toevaluate its current success and its aspira-tions for the future.

Over the last 10 years, the company hasbeen in a virtuous modeling-learningcycle.Example 3: Using Models for Control andConsistency

A global healthcare organization wantedto rationalize its allocation of sales andmarketing resources across a diverse set ofgeographic markets. The president of thefirm recognized that he and his immediateadvisors lacked the local knowledge theyneeded to effectively evaluate the compet-ing proposals for resources from over 50countries.

For this modeling implementation tosucceed, the management team in eachcountry had to understand, accept, andimplement a modeling approach that wasrobust, actionable, and tailored to localmarket conditions and that also enabledcross-country aggregations and trade-offs.

Our firm simplified and tailored a fam-ily of proven optimization models basedon sales response for use across the cus-tomer’s countries. We trained sales and

marketing personnel on the concepts, con-sulting process, and software models. Be-cause the problem of allocating resourcesrecurs annually, the local country teamsneeded to internalize this knowledge.

We achieved much of the process con-trol and consistency through commontraining, common tools, and a consistentsupply of external resources to help the lo-cal teams in implementing the models intimes of need. The president’s staff alsoachieved control and consistency by ask-ing similar questions from the managersof the various countries:—What resources and implementationprocesses do you think are most suitablefor your market? What sales and profitline can you deliver if the company pro-vides these resources?—If your resource request is cut back by10 percent, what part of your preferredimplementation would you give up? Whatsales and profit line will you achieve un-der these circumstances?—If your resources were limited to thecurrent level, how would your implemen-tation be affected?

This modeling approach, coupled withthe implementation process elements, fa-cilitated a transformation in this globalfirm’s decision-making culture. A cultureof horse trading was transformed into aculture of commitment. Affiliates commit-ted to outcomes that were transparentlyconnected to resources; the president com-mitted to reduced expectations when re-ducing resources.

While this modeling approach improvedthe organization’s resource-allocation deci-sions, the implementation process greatlyenhanced the practices of sales and mar-

SINHA, ZOLTNERS

INTERFACES 31:3 S42

keting personnel, who adopted a commonlanguage and a sense of commitment topromised outcomes.

We gained a very interesting insightfrom this work. The participants in theprocess typically accepted the rational andinformed nature of the modeling and theimportance of accuracy and integrity ofmodel inputs. General managers wantedtheir sales and marketing managers to

After a 14-year gap, thecompany is working with usagain.

provide their best estimates of the saleslevels they could achieve at any given re-source levels. Likewise, the global head-quarters personnel wanted the generalmanagers to put forward their best esti-mates of the aggregate expectations intheir markets. However, from the other di-rection in the decision chain, each chainmember wanted to introduce cushion inthe models. They wanted to promise lesssales but beat their forecasts. These cush-ions had the unfortunate effect of reducingthe resources allocated to those adding thecushions.

In this situation, models were used toensure consistency and control in the allo-cation of scarce resources.Example 4: Good Models are Not Enough

A company had three sales forces ofabout 500 salespeople each. The com-pany’s products had been divided amongthe three sales forces, primarily for histori-cal reasons. Each sales force called onsimilar customers.

The company had decided to redistrib-ute the products among the three sales

teams in a more rational manner andwanted to size the three new sales teamsbased on the revised product portfolios. Areduction in total sales-force size was ex-pected. Then the company would need todesign sales territories for each of thethree sales teams and assign the varioussalespeople from the three old sales teamsto a territory in one of the three new salesteams. All unassigned salespeople were tobe laid off.

We assembled a large team from ourconsulting firm to work on this projectwith a core team of key sales directorsfrom the client.

The joint team first developed sales-response relationships for the productsunder study, using a combination ofhistorical-data analysis and judgment fromthe client team and from other client man-agers. By moving products around amongpossible team configurations, we identifiedthe configuration that made the mostsense in terms of product synergies andcustomer overlaps and then used thesales-response relationships to size thesethree new sales teams optimally. The anal-ysis called for a sales-force reduction of150 people.

For each sales team, we defined the op-timal sales-territory locations. We defineda set of business rules to assign the exist-ing salespeople to these new territory cen-ters, using such criteria as past perfor-mance, seniority, geographic fit, and priorexperience with that new sales team’sproducts. Using these rules, we assignedall the existing salespeople to a new ter-ritory or slated them for termination.Second- and third-level sales managersreviewed and refined these assignments.

SALES-FORCE DECISION MODELS

May–June 2001 S43

The company’s legal and human resourcesdepartments then approved the locations.

Once we finalized the personnel assign-ments, we designed actual sales territoriesaround these locations for the new sales-people. We essentially created three com-pletely new territory alignments, each cov-ering the entire country for its uniqueproduct line. These territories themselveswere reviewed and refined at several acri-monious meetings by over 100 first- andsecond-line sales managers for the threesales teams.

We distributed the information to a va-riety of functions within the client’s com-pany for implementation. The companyput the new structure into place, withnearly 1,400 people working in their newroles, with different products, perhapsnew bosses, and new territories, in lessthan five months from when the projectwas first discussed.

After what seemed to us to be a success-ful conclusion to a high pressure, stress-filled, fast-paced sales-force-restructuringproject, we anticipated that we would behearing requests for other help we couldprovide to the company. Instead weheard . . . silence. There were no projectsfor the next 14 years, and our attempts atcontacting the senior managers were re-buffed as well.

Only some years later did we find out,as people in the company eventually cameto tell us, that top management had notbeen as pleased with the results of theproject as we had thought. It turned outthat senior-most management felt that wehad placed too much emphasis in the ter-ritory alignment on designing geographi-cally good territories, rather than on

minimizing disruption to existingsalesperson-customer relationships. Theythought that this had hurt sales perfor-mance in the first months after they imple-mented the new structure. They also

Judgment helps insure thatmodels are implementedsuccessfully.

thought that the core team members fromthe client side had injected too much judg-ment into the inputs that drove sales-forcesizes, personnel assignments, and territorydesigns, pulling the recommendationsaway from what the strictly objective datashowed. While at the time we thoughtthese refinements were improvements, af-terwards the client’s top managers sawthem as poisoning the answer.

In short, the client expected us to havedriven the process more than we did, rely-ing more on our expertise than on its ownpeople’s judgment or on criteria they haddevised for us to use. Learning this lessonwas perhaps even more valuable than theproject itself had been for us.

After a 14-year gap, the company isworking with us again. We hope we havebecome better modelers and problem solv-ers. This time, we plan to not just modeland analyze but to also listen and advise.A Final Thought

Many years have passed since one of uspostulated that models would be used formost business decisions. Models certainlywere very useful for the sales-force proj-ects we described. Yet in the entire spec-trum of business decision making, formal-ized models are not very common. Mostdecisions are based on experience, knowl-

SINHA, ZOLTNERS

INTERFACES 31:3 S44

edge, and wisdom. Our own thinking ondecision making has moved in this direc-tion as well. As we organized our model-ing experiences for this paper, we noticedthat our appreciation of experience andjudgment, of implementation processes,and of stakeholder involvement has esca-lated appreciably. Our belief in the powerof models is still strong, but our apprecia-tion of the softer elements of problemsolving and change management is muchstronger than it was before. Models helpshape judgment, and judgment helps in-sure that models are implemented success-fully. The combination of modeling andjudgment is very powerful.Acknowledgments

We thank the current and former em-ployees of ZS Associates, who werelargely responsible for the projects whoseinsights populate this paper. Specialthanks go to Jaideep Bajaj, Ellen VenturaCoury, Sally Lorimer, Songjun Luo, ChrisMorgan, Dan Peterson, and Chris Wrightfor contributions to the examples in thispaper, and to Linda Kluver and SamanthaAlfassa for their care in producing it. Wealso recognize Arvind Rangaswamy andGary Lilien for their persistence and insis-tence that we document our experiences,and we thank the reviewers for their con-structive comments.ReferencesChintagunta, Pradeep K. 1993, “Investigating

the sensitivity of equilibrium profits to ad-vertising dynamics and competitive effects,”Management Science, Vol. 39, No. 9 (Septem-ber), pp. 1146–1162.

Lodish, Leonard M. 1975, “Sales territory align-ment to maximize profit,” Journal of Market-ing Research, Vol. 12, No. 1 (February), pp.30–36.

Lodish, Leonard M.; Curtis, Ellen; Ness,

Michael; and Simpson, M. Kerry 1988, “Salesforce sizing and deployment using a decisioncalculus model at Syntex Laboratories,” Inter-faces, Vol. 18, No. 1 (January/February), pp.5–20.

Mantrala, Murali K.; Sinha, Prabhakant; andZoltners, Andris A. 1992, “Impact of resourceallocation rules on marketing investment-level decisions and profitability,” Journal ofMarketing Research, Vol. 29, No. 2 (May), pp.162–175.

Skiera, Bernd and Albers, Sonke 1996,“COSTA: Contribution optimizing sales teri-tory alignment,” Marketing Science, Vol. 17,No. 3, pp. 196–213.

Tull, Donald S.; Wood, Van R.; Duhan, Dale;Gillpatrick, Tom; Robertson, Kim R.; andHelgeson, James G. 1986, “Leverageddecision-making in advertising: The flat max-imum principle and its implications,” Journalof Marketing Research, Vol. 23, No. 1 (Febru-ary), pp. 25–32.

Zoltners, Andris A. 1976, “Integer program-ming models for sales territory alignment tomaximize profit,” Journal of Marketing Re-search, Vol. 13, No. 4 (November), pp. 426–430.

Zoltners, Andris A. and Lorimer, Sally E. 2000,“Sales territory alignment: An overlookedproductivity tool,” The Journal of Personal Sell-ing and Sales Management, Vol. 20, No. 3(Summer), pp. 139–150.

Copying of this document is by permission of the Institute for Operations Researchand Management Sciences.

Copy by permission of the Institute for Operations Research and ManagementSciences. Requests for permission to republish or redistribute should be addressedto: INFORMS, 901 Elkridge Landing Road, Suite 400, Linthicum, MD, 21090,USA.