Evaluation of Salesforce Size and Productivity Through Efficient Frontier Benchmarking Dan Horsky + Paul Nelson University of Rockester Abstract The efficient operation of a salesforce is a critical ele- ment in the profitability of many firms. Three factors play key roles: the salesforce’s size, its allocation and its productivity. This gives rise to the following questions: can salesforce performance be improved by (1) hiring ‘more salespeople, (2) allocating them more effectively to the various sales districts and/or (3) improving salesperson productivity through better calling patterns in terms of consumers and product line items? The practice of most firms and the methodology used in most of the academic literature to address salesforce design and productivity questions is a “Bottom Up" approach. This approach starts with assessments by each salesperson of the sales and ef- fort corresponding to each customer and prospect in their territory. These assessments are then aggre- gated to the territory, district and national levels, This paper takes an alternative "Top Down” ap- proach. It is based on an estimated relationship be- tween district level sales and salesforce size, effort and other variables. This more macro level decision tool can be used by management in parallel to, and as an objective check of, the more conventional and more subjective “Bottom Up” approach. We develop an efficient frontier methodology which allows us to estimate how total district sales respond to salesforce size, district potential and competitive activ- ity in the firm’s best performing districts. The method- ology utilized is based on Data Envelopment Analysis (DEA) and yields a benchmark measure of each dis- trict’s efficient frontier sales (sales assuming the dis trict’s salesforce allocates its effort as done in the best performing districts). Based on the estimated response function we discuss the three potential sources of increased profitability: .0732.2389/96/1504/0301801 25 ‘Copy © 1 tie Opts Reach closing the inefficiency gap of each of the lower per- forming districts, optimally reallocating the current salesforce to the various districts, and changing the cur- rent size of the salesforce to its optimal level. The inef- ficiency gap issue is addressed through comparison of the parameter estimates for the best districts obtained through our methodology with those of an average dis- trict sales response function obtained using regression analysis. This comparison points to an important meth- odotogical finding, The use of multiple estimation re- sults may lead to an improved understanding of the phenomenon being studied (in our case, the identifica: tion of the likely causes of district productivity ineffi- ciencies). The latter two sources of increased profitabil- ity, salesforce reallocation and changes in the current salesforce size, are addressed analytically given the dis- trict level efficient frontier sales response function. The proposed “Top Down” procedure using the ef- ficient frontier methodology and the insights it provides are examined by evaluating the operations of two dif- ferent salesforces, one selling manufacturing equipment and the other business equipment. In both cases, regression-based analysis would have resulted in a dec- laration that the status-quo was close to optimal, while the frontier-based analysis pointed out that strong gains ‘were possible in certain districts. In particular, for both firms, the greatest increases in profit are obtained through improved salesforce efficiency in the lower per- forming districts, not through salesforce size or district allocation adjustments. At the more micro-level, a com- parison of the frontier and regression parameters made it possible to identify which specific changes in the daily ‘operations of the salesforces would allow the realization of these potential productivity gains. In our two cases this could be obtained through more emphasis on pur- suing prospective accounts (Salesforce; Benchmarking; Frontier Estimation) MARKETING SCIENCE/Vol. 15, No, & 1996 ‘pp. 301-320,EVALUATION OF 5) THROUGH EFFI 1. Introduction Many firms, such as producers of manufacturing and business equipment, are critically dependent on their salesforces to inform consumers about their products and to perform sales transactions. The results achieved by the salesforce are dependent upon its size, geo- ‘graphic breakup, and the allocation of individual sales People’s time to specific types of consumers and to spe- Cific product line items. A methodology that can iden- tify the facets of a firm’s salesforce operation that are ripe for improvement can be a key to increased profit ability. For example, will hiring more salespeople im- prove profitability or is the current salesforce size op- timal? Can the salesforce be better allocated to districts? Can the members of the salesforce improve their per- formance by changing their calling patterns? This paper develops and illustrates a methodology that is capable of providing insights into these questions, Within an organization, technical salesforces are usu- ally broken down into regions and then further into dis- tricts. Several salespeople operate within each district and often each salesperson has an exclusive territory in which to sell the product line of the firm through peri- dic calls on current and potential clients. While overall sales goals are determined by upper management, the detailed operations of a salesforce are usually planned using a “Bottom Up” approach. With a district man- ager, each salesperson evaluates the likely sales to the customers and prospects in their territory. Sales goals and quotas are set, and when these are combined with those of the other salespeople in the district, a district sales forecast results. This forecast in turn can be aggre- gated with those of other districts to generate a national sales forecast. If the national forecast does not match the national sales goal, the individualized quotas can be ad- justed. When the current salespeople are judged to be operating near their utmost capacity, salespeople may be added and territories realigned. The existing literature on salesforce problems mirrors this “Bottom Up’ approach. The basic unit of measure- ‘ment is either the individual purchasing account or the individual salesperson’s territory, and the responses of these units to the number of sales calls are either esti- mated or subjectively determined. Knowledge of these response functions allows companies to plan the allo- cation of each salesperson’s time and to realign sales 302 ESFORCE SIZE AND PRODUCTIVITY ENT FRONTIER BENCHMARKING. territories. Models representing the response of sales to the level of sales effort have been formulated and esti mated by Lucas, Weinberg and Clowes (1975); Beswick and Cravens (1977); Lilien (1979); and Ryans and Wein- berg (1979, 1987). The allocation of a salesperson’s time to customers and prospects and across products has been examined by Montgomery, Silk and Zaragoza (1971); Lodish (1971, 1974, 1976, 1980); Fudge and Lod- ish (1977); and Lodish, Curtis, Ness and Simpson (1988). The design and operation of sales territories has been studied by Hess and Samuels (1971); Lodish (1975); Zoltners (1976); Zoltners and Sinha (1983); and Rangaswamy, Sinha and Zoltners (1990). Within any individual salesperson’s territory, the al- location of a salesperson’s time to customers and pros. pects—and across the firm’s products—need not be controlled by these direct allocation methods. Indirect control of the salesperson’s efforts can be achieved through an appropriate compensation scheme. Assum- ing that each salesperson knows his or her territory bet- ter than anyone else in the firm, a compensation scheme which allows simultaneous maximization of the sales- person's and firm’s objectives (ie, income and profit) will lead the salesperson to allocate time such that it ‘maximizes firm profit. The indirect control of salesforce efforts through compensation schemes has been studied by Farley (1964); Davis and Farley (1971); Srinivasan (1981); Basu, Lal, Srinivasan and Staelin (1985); Cough- lin and Sen (1989); Rao (1990); and Lal and Srinivasan (1993) ‘The “Bottom Up" approach requires very detailed in- formation for its successful implementation, For exam- ple, for each customer and prospect an accurate func- tion reflecting sales response to salesperson calling time is needed. Since these functions are usually estimated subjectively, their accuracy is far from perfect (eg., Charkravarti, Mitchell and Staelin 1981) and aggrega tion compounds this problem. Therefore, it would seem. worthwhile for top management to augment the “Bot- tom Up” planning of the salesforce operation with a “Top Down” evaluation when making decisions on such important matters as the size of the national sales- force. The “Top Down” approach is a more aggregated approach which allows top management an indepen- dent assessment (.e., with little or no input from the salesforce) of the salesforce size, its geographic breakup Magkenine Sctence/Vol. 15, No. 4, 1996HORSKY AND NELSON and time allocation. In this approach, the basic unit of measurement is the sales district. District level sales re- sponse to the number of salespeople is estimated. Based ‘on this function, the optimal number of salespeople in each district and thus the nation as a whole can be de- termined. The more detailed district level operational issues, such as the allocation of salespeople within the district to territories, customers, product lines, etc., can then be achieved using the “Bottom Up” methodology. ‘The benchmarking methodology developed in this paper estimates a function that reflects the “efficient frontier” sales response to salesforce size, effort and. other variables. This efficient frontier sales function re flects the productivity of only the firm’s top performing districts. Using this response function, the profit maxi- mizing number of salespeople may be found, each dis- trict’s efficient frontier sales can be calculated, ineff cient districts can be identified and the extent of these inefficiencies can be revealed. Through comparison with regression-based estimates of the average sales re- sponse function, possible reasons for these inefficiencies can be uncovered. ‘The following study outlines and demonstrates this “Top Down’ approach. The first section models district sales response to the number of salespeople, reviews the efficiency frontier methodology used to estimate this function and contrasts this procedure with regression analysis. The second section derives the optimal sales- force size in each district and nationally from the sales response analysis. In addition, the optimal reallocation of the national salesforce to districts when its size is constrained (at least in the short run) is derived, The third section examines the operations of two separate salesforces in very different industries and illustrates the efficient sales frontier estimation methodology and. the optimization procedures. The fourth section pro- vides a summary and suggests some further applica- tions of the proposed benchmarking procedure. 2. Sales Response to the Number of Salespeople Sales of a firm's products, especially when industrial and commercial goods are involved, depend on the size and productivity ofthe firm's salesforce. As advertising does for consumer goods, a salesforce for industrial or MARKETING SciENCE/Vol. 15, No. 4, 1996 fie Frontier Benchmarking commercial goods informs customers and prospects of the existence and attributes of a firm’s products. In ad- dition, for these goods the salesforce performs direct sales transactions. Aggregate (national) sales response models are fre- quently used to measure the effectiveness of advertis- ing. Estimation commonly is conducted through time series analysis. (For a review, see Hanssens, Parsons and Shultz 1990.) Unfortunately, the use of time series analysis to measure salesforce effectiveness is generally not feasible because not enough variation in the sales- force size occurs over time. Unlike advertising budgets which are handled by an outside advertising agency and are often drastically adjusted in response to busi- ness cycles and other events, salespeople are part of the firm's organization and their employment is smoother over time. As a result, a cross-sectional rather than a time series analysis is more appropriate in salesforce applications. The natural unit of measurement for this, ‘cross-sectional analysis is the sales district. Typically, salesforces operating in the U.S. are divided into 25 or ‘more separate sales districts, each usually having a dif- ferent number of salespeople. 24. The District Sales Response Function ‘The firm’s sales in any district depend on the potential for its product line in that district and the selling effort exercised. A general formulation is SALES, = (POTEN,)(SPEOP,)”, a where for district i, SALES, equals the firm's sales, PO- TEN, is its sales potential and SPEOP, is the number of salespeople assigned to the district. The constant y mea- sures how sales are affected by the size of the salesforce. Since additional salespeople will have to call on less promising accounts, diminishing returns are associated with additional salespeople. This means that 0 < y Since lagged sales in part define a citric’ potential, strictly speak: ing discounted profits over time should be maximized and dynamic optimization techniques, such as optimal contol, shouldbe used. For the sake of simplicity, this ese 8 ignored and a state optimization is sed as an approximizatio, NSIZE* Itis evident from Equation (9) that districts with larger potentials, POTEN,, will be assigned more salespeople. It also follows from Equations (9) and (10) that—if the ‘gross margin, g, is increased or the cost per salesperson, ¢, falls—the optimal district and national salesforce sizes will increase. Moreover, a similar outcome occurs if the salesforce becomes more efficient as mapped through an increase in either y or the parameters gov- erning the potential, a, and fr In the short run, if the national salesforce is not al- lowed to change its size, NSIZE, but a redeployment of its members is acceptable to management, a con- strained profit maximization problem results. In this case, the national profit maximization problem be- Max 5 = ¥ [g(POTEN, MSPEOP,)” ~ e(SPEOP)1 (11) st 5 SPEOP, = NSIZE ‘The corresponding optimal district salesforce sizes are sobor = Nae PORN ier ay ‘Thus, the existing salesforce is redeployed such that each district receives an allocation which is depen- dent on its relative potential. Ifthe national salesforce size is allowed to grow but is constrained to be below the optimal national size, NSIZE*, the allocation to districts will also follow Equation (12).* As with the nonconstrained profit maximization, the actual dis- trict salesforce sizes obtained through Equation (12) * Based on Equation (12) the marginal profit obtained from the last salesperson in each ofthe districts is identical. [n Equation (9) this is ao true. In addition, in Equation (9), all marginal profits equal zero making any increase or decrease in salesforce size result ima profit less than the masdeal prof obtained at SPEOP? MARKETING SCIENCE/VVol. 15, No. 4, 1996HORSKY AND NELSON must be rounded either up or down to the more prof- itable integer value while maintaining the total na- tional size. Alternatively, one can use an incremental, integer-value-based algorithm to find the constrained. profit maximizing integer value for district salesforce sizes 4. A Study of Two Salesforces The “Top Down’ salesforce evaluation methodology forwarded in this paper was used to examine the op- erations of two different salesforces, District level sales response functions were estimated. Next, the empirial results were used to investigate how addi- tional profit could be made foreach frm through (1) improving the efficiency ofthe existing salsforce in less efficient districts, (2) improving the operation of the existing salesforce by better allocation to districts, and (3) increasing or decreasing the sizeof the sales force. Firm A is a business equipment manufacturer while Firm B is a chemical production equipment manufacturer. Each firm is one ofthe largest manu- facturers in its respective product class. Firm A em ployed 230 salespeople in 26 districts while Firm B employed 129 salespeople in 27 districts. In both firms, the pricing and advertising policies were iden- tical across their sales districts. For each istrict, the following yearly data were collected: sales, the num- ber of a company’s own and competitive salespeople (and some related salesforce “size” variables), and measures of the customer and prospect bases. Cost per salesperson and gross margin also were obtained for each frm. ‘The number of a company’s own salespeople was collected from monthly logs and and. working months aggregated to form non-integer values for the salesforce size. For Firm A, a 1 to 10 rating of each salesperson’s “quality” as assessed by his or her distriet manager and eye-to-eye contact hours also were available, Adjusting for the “quality” of each salesperson allows a relaxation of the assump- tion, implicit in our development of the sales re- sponse function, that every salesperson has equal trinsic selling ability. The eye-to-eye contact hours do not include time spent on nonselling activities MARKETING SciENCE/Vol. 15, No. 4, 1996 ficient Frontier Benchmarking such as travel and as such may prove to be a better measure of salesforce selling effort than the number of salespeople. The number of competitive salespeople was col- lected through a survey of each firm's salespeople and district managers. Some subjective assessments were required as some of the smaller competitors employed dealers instead of sales representatives in certain districts and their “people” equivalent had to be approximated. Moreover, management at Firm B felt that some of the smaller competitors were pay. ing more attention to the Los Angeles and New York City areas (for example, by employing sales repre- sentatives rather than dealers), and as a result, the effectiveness of the competitive sales operations in the two districts was larger than in the other dis tricts. Consequently, for Firm B two. variables were used to represent competitive activity. One variable being the number of competitive salespeo- ple and the other variable being a dummy variable equal to one for these two particular districts and zero otherwise. It follows that the parameter asso- ciated with this dummy variable is expected to be negative. ‘The customer base for each district was measured either by the number of customers in the previous year or lagged sales. The latter is preferable since it takes into account the purchasing power of the cus- tomers. Unfortunately, this was available only for Firm B. As stated previously, the size and geographic dispersion of the various customers (and prospects) if not distributed equally across districts may impact the productivity of district salesforces. However, both firms used a centralized national accounts sales ‘group to deal with especially large customers, such as the U.S. government. With respect to the geo- graphic dispersion of the customer and prospect bases, for Firm B a rough measure in the form of each district's area in square miles was included. For Firm A, the eye-to-eye contact hours measure deals with this issue. Data concerning the prospect base were not di rectly available so various measures of overall mar ket potential, such as the population size or the num- ber of refineries in a district, were used as proxies. For each firm, about twenty different proxy variablesHORSKY AND NELSON les Response Function Paramer Estimates eG uF Constant sare 2072 B, Number of Customers oor oa Br Population 0153 0389" 8, Numer of Comptive 0.069, 0136" Salespeonle Number of Salespeople 042" 0.302" (Aas FP) 093 (092) 7 ” % 26 * Satisicaly significant at 10% level, Stats signtiean at $% lee ‘were collected. Some were chosen based on the types of businesses and industries that did business with Firms A and B. Other proxy variables were general economic and demographic indices. These proxies were obtained from U.S. County Census reports as well as commercial data sources. The data generally were available at the county level and had to be ag- _gregated to correspond to the county composition of each firm's sales districts.” 4,1 Sales Response Estimation A priori it was not clear which market potential vari- ables best represent the prospective customer base. Since a high degree of collinearity was found be- tween many of these proxies, their number was first reduced by examining their correlation matrix and choosing a relatively independent subset. This re- sulted in about five proxy variables for each firm. Then for each firm, the sales response function was estimated using REG and EFF with the remaining proxy variables as well as with the other explanatory variables, Based on goodness of fit, statistical signif- icance and the intuitive appeal of the parameters, certain proxy variables were chosen to remain in the analysis. In principle, significance in either or both of the estimation techniques led to the inclusion of a “The identification of current data sources forthe various proxi and their arrangement according tothe firm's districts was the most time ‘consuming par of each projet. 310 fen Frontier ercmaring proxy variable, The sales response functions chosen for both firms are FIRM A: SALES = e"(LCUSTY*(POPUL)** x (COMPTY"“SPEOP)” 3) FIRM B: SALES = ¢*(LAGSL)""(FOODM)"*"(REFINY"* x (COMPTY*"(e%4")(SPEOP)” (14) LCUST = Number of customers in the previous period LAGSL = Lagged sales POPUL = Population size FOODM = Number of food manufacturing plants REFIN = Number of refineries COMPT = Number of competitive salespeople Dias = Dummy variable for Los Angeles and. New York City SPEOP = Number of own salespeople. ‘The estimated parameter values of the sales response functions (13) and (14) using REG and EFF are reported in Tables | and 2. The REG results represent an average sales response. The EFF results reflect only how sales in the top (efficient) districts respond to salesforce size and Table 2 Firm B: Sales Response Function Parameter Estimates ar ® Constant ase" 7516" 5, Lagged Sales 0398" 0400"* Bey Number of Food 0085 0.106" anu. Panis bax Number of Retneries 036" 0.050"* Bay Number of Compete 054 00% Salespeope Ba, NYE Orla 0212 -0284* y Numb of Salespeople ese" osra* (Adjusted) 087 (83) ~ a 2 + Statistical signicant at 10% lve, ** Staseay sonia at 5% level MARKETING SciENCE/Vol. 15, No. 4, 1996HORSKY AND NELSON the other variables.° The statistical significances of the EFF parameters are measured using the bootstrap method referred to earlier. Sales response for Firm A also was estimated using, the effective number of salespeople based on the district managers ratings of their salespeople and eye-to-eye contact hours rather than the number of salespeople. These alternative measures of salesforce size yielded statistically inferior results and, therefore, were not pur sued further. For similar reasons, for Firm B, the district square mileage variable used to measure the density of the customer (and prospect) base was dropped. 42 Parameter Values and Benchmarking Actual sales as well as predicted district sales based on the REG and EFF parameter estimates for Firms A and B are provided in Figures 3 and 4. As expected the sales predic: tions are higher when EFF parameter estimates are used than when REG estimates are used, Comparison of actual district sales and EFF-based predicted sales provides a mea- sure of district salesforce performance. The ratio of these two values provides an intuitive efficiency rating of how close each distric’s performance is to the frontier. These ratios are provided in Figures 5 and 6. The rather random distribution ofthese efficiency ratios over district sales lev- cls shows that (a) there is no apparent bias in the frontier ‘estimation procedure regarding the sales level and (b) that “igor “small” districts are not inherently more efficient. ‘The sales response function parameter estimates pro- vided in Tables 1 and 2 generally conform with expecta- tions and provide further insights relating, to salesforce ‘management and performance. Comparisons across the two estimation techniques show that a company’s own salesforce size is a consistently significant factor. However, ithas a slightly lower impact on district sales inthe premier districts. Since the more efficient districts do not necessarily have larger salesforoes, this cannot be due to scale effects It seems that, indeed, the addition of salespeople in the more efficient districts is marginally less productive. This * Due to confidentiality sues the data reported have been rescaled. Dstt sales figurs and the coe per salesperson are multiply the sae cone tant, This rescaling fects oly the size of the interept erm a The pti ‘leserce si sls are not affected nor are the rave changes in sales fed profits due fo the various manager scons The rescaling does affect the solute magnitudes of he sales and profi agus ep MARKETING ScIENCE/Vol. 15, No. 4, 1996 ficient Frontier Benchmarking ‘may be due to the finding discussed below that these sales- forces already generate more sales from the clientelein their districts. Consequently, additional sales are mote difficult, ‘As expected, the competitive salesforce size has a detti- mental, but small, effect on the sales of Firm A. While this parameter is positive for Firm B it is small and is not sta- tistical significant. This “incorrect” sign may result be- ‘cause competitors have assigned more salespeople to the more lucrative districts. In addition, the Dy -wy parameter {Bs for Firm Bis negative. As expected the effectiveness of competitive sales efforts in the Los Angeles and New York ity districts is found to be greater than in other districts The parameters for past sales activity (LCUST or LAGSL) and for potential clients (proxied by total mar- ket potential variables POPUL or FOODM and REFIN) reveal a key insight. For Firm A, relative to the average response function estimated using REG, the best pro ducing districts reflected by the EFF parameters show a much larger sales impact of POPUL while the impact of LCUST is smaller. For Firm B, the efficient frontier sales response function also shows an increased impact of REFIN and FOODM, while the lagged sales parameter is essentially unchanged. This observation is further re- inforced when the significance of the parameters is taken into account. Not all the prospect base proxy vari ables, such as POPUL for Firm A, are significant in REG but all are significant in EFF. What is thus evident for both firms is that in the best producing districts sales are more strongly tied to the potential customer base. ‘The increased sizes of the potential customer param- eters under the EFF formulation imply that in the effi- cient districts either more attention or better focused at- tention is paid to prospects. Conversely, the reduced. size of the existing customer base parameter for Firm A. implies that less attention is directed at these accounts in total, In all likelihood the less promising customers (Gn terms of sales versus effort required) are given less attention or dropped in the more efficient districts. For Firm B, the same level of response is witnessed for cur~ rent accounts across efficient and inefficient districts al- though as pointed out above more attention is paid to promising prospects in the better performing districts.” "This inference could have been cros-vaidated had the eyetoeye contact hours for Fir A been separately specified for existing custom fers and prospects, Unfortunately they were not aitHORSKY AND NELSON ficient Frome Benchmarking Tables Distt etal oF 1 99 as 2 846 246 a 99 a9 4 500 500 5 0383038 6 1100 11.00 7 800 200 8 523 523 9 758 154 0 700 700 " 938 938 2 954 954 3 ra eas 6 16381638 15 500 600 6 500 500 7 808 808 8 8.00 200 9 885 885 20 400 40 2 an an 2 700 700 23 631 63t 24 an an 25 589 669 28 538 538 2 400 400 Total Salesforce Sv 2208372983 Total Sales ($000) 314262 966.082 Total Profit (5000) 212 3902 (ange in Pratt (%) 292 become efficient, all districts would realize efficient frontier sales and corresponding profits. Tables 3 and 4 provide the actual sales, profits and. salesforce allocations for each firm as well as those that would be achieved if all districts became efficient, if the existing salesforce were reallocated across dis tricts to maximize profit, and if an optimal salesforce size and allocation were implemented. The latter val- ues are calculated using both the REG and EFF param- eter estimates. It is interesting to see that for Firm A the optimal REG-based salesforce size is marginally smaller than the current one while the EFF-based size au Aes eG ° 10 9 0 9 8 9 0 0 10 10 0 5 5 5 5 2m B By) 25 9 10 9 0 8 7 8 7 7 6 6 7 8 8 8 8 7 6 7 7 0 10 0 0 0 9 8 9 "7 "7 ” 8 4 14 4 “ 7 7 7 7 5 5 5 5 5 9 9 0 9 10 8 0 0 10 0 " 5 5 5 5 5 5 5 5 5 5 5 5 1 7 7 7 6 6 6 6 7 a 7 8 a ? 8 8 2 2 8 4 20 20 er 209 saga 367,105 BHT S84 372540 3038381780338, 199, 04 00 is slightly larger. Note that the optimal salesforce size as specified in Equation (9) is dependent on the values of y and POTEN,. Despite y being slightly larger un- der REG, the optimal salesforce size is larger under EFF due to its larger potentials. For Firm B, both es- timation procedures end up recommending huge, roughly four-fold increases in salesforce size. Con- trary to Firm A, the REG-based salesforce size is larger than the one implied by EFF. In this case, the difference in the size of the y parameters is large enough to overcome the larger potentials estimated using EFF, MARKETING ScieNce/Vol. 15, No. 4, 1996HORSKY AND NELSON ficient Frontier Benchmarking Table 4 current location Distt Curent EFF 1 225 225 2 500 600 3 575 575 4 3a 3a 5 508 503 6 800 600 7 392 392 8 608 508 9 40 400 10 500 500 1" 49 492 2 392 392 1a 500 500 “4 600 500 15 300 300 6 200 200 7 550 650 8 278 295 9 390 350 20 908 908 a 43 483 2 400 400 23 583 563 24 483 a3 2 438 458 2% 675 675 2 400 400 Tota Suesorce Sue 12879 128,79 Total Sales ($000) 170388 211.176 Total Pott ($000) roe 31004 Change in Prot (4) 08 Tables 3 and 4 also detail the change in profit which ‘would result from the three managerial alternatives out- lined earlier based on the EFF parameters. For Firm A, profits can be increased by 29.2% if ll current salespeo- ple become efficient. Reallocating the salesforce across districts would only improve profits by another 0.4%. AA further move to the optimal salesforce size would re- sult in a negligible profit increase. Firm B, in addition to increasing productivity in its inefficient districts, may also need to increase its salesforce size. Its profits can be increased by 40.8% if its salespeople act more effi- ciently. An additional 27% profit increase could be ob- [MARKETING ScIENCE/Vol. 15, No. 4, 1996 Firm B: Actual and Predicted Impacto Salesforce Size and Alcaton on Sales and Profit Optimal Realocation Opteral location res oF eG ee 5 2 9 5 4 6 3 4 15 1 2 5 5 8 5 a 2 7 a an 2 6 6 2B 2 7 10 2B ar 5 6 2 a a 5 ry 20 5 6 9 a 5 3 20 2 4 7 7 8 6 7 28 26 1 1 5 4 1 1 5 3 4 2 5 8 2 2 8 7 2 2 7 7 3 5 54 we 7 6 2 2 4 4 1a “4 0 7 2 2 3 3 2 3 4 5 7 18 6 8 2 0 2 2 8 9 129 19 529 ao rade 218035 439,t00 455,520 2612 31828 2820 2.856 21 348 tained by reallocating these salespeople and a further 34.6% added by increasing their number. However, the increase in size is immense. Is this justified? Figures 7 and 8 trace the potential national profit that results when the national salesforce is optimally allocated bbased on the REG and the EFF parameters for various national salesforce sizes. For Firm B, starting at about 300 salespeople, the profit function becomes rather flat raising a question about the value of adding hundreds of salespeople for a minimal expected increase in prof its, Even assuming that recruitment and training costs are negligible (neither is taken into account in our 315HORSKY AND NELSON fice Frontier Benchmarking ne fF 22 a £ 2 F as so ge 8 8 8 8 2 [NUMBER OF SALESPEOPLE analysis), the risky nature of the small expected addi- tional cash flow relative to the certain additional fixed costs may be enough to tame the implied hiring flurcy.* This risk issue is further enhanced by the extrapo- lation of Firm B's district salesforce sizes to levels far exceeding those currently witnessed. In Tables 3 and 4, we see that the current district salesforces vary in size from 4 to 30.38 for Firm A and from 2 to 9.08 for Firm B. When changing the salesforces to their optimal size, the district salesforce sizes for Firm A range from 5 to 26, actually less than the current range. For Firm B, the range is from 3 to 37 with 19 of the districts having a salesforce size greater than 9, the maximum size currently. This extrapolation issue raises a concen over whether the estimated coefficient of salesforce size, 7, might be improper for use in this higher size “This fatness of the profit function over a wide range of salestorce size is an empirical eulasity (Zolners and Sinks 1996), 316 range. In Tables | and 2, we see that the EFF value of +y is 0.382 for Firm A and 0.573 for Firm B. Given that Firm A generally has larger district salesforces this might imply that, had there been districts with larger salesforces for Firm B, its value of y would be smaller. Correspondingly, a smaller optimal salesforce size would result. The practical implication is that while the national salesforce of Firm B should be increased in size, its growth should be gradual. This will allow the value of y to be updated to reflect sales response to a higher range of district salesforce sizes before a more drastic growth in the number of salespeople is implemented. Quicker learning about the value of y at larger salesforce sizes can be achieved through exper- mentation, if doing so is acceptable to management. 44 Conclusions The bottom line observation on the operation of each of the two salesforces is that their national size and district allocation were determined reasonably well by man- MARKETING ScIENCE/Vol. 15, No, 4, 1996HORSKY AND NELSON fin Frontier Benchmarking Fue 8 i bad PROFIT (5000 1000 agement. For both of these firms, the most promising ‘way to improve profitability is not the shifting of sales- people among districts or the hiring of more salespeo- ple. Rather, it is through improved salesforce produc- tivity. In particular, more emphasis on prospects seems to be a key tool for achieving this improved productiv- ity. In poorly performing districts, individualized “Bot- tom Up” plans with a stronger focus on prospects should be implemented. Moreover, a nationwide com- pensation scheme which better rewards salespeople for sales to new accounts might merit consideration. It thus appears that for these two firms how they spend their marketing budget is at least as important as how much they spend. A similar conclusion is reported in an ad- vertising study by Eastlack and Rao (1986) and a nu- merical analysis by Mantrala, Sinha and Zoltners 1992) 45. Implementation Some elaboration is warranted about the use of the es- timation results for actual managerial action with im- Markerinc Science/Vol. 15, No. 4, 1996 plications for the future. The parameter values esti mated and the optimal policies implied are valuable in ‘evaluating the current operation of the salesforce. Rec- ‘ommendations concerning future salesforce size and al- location ate only correct if no changes occur in the fu- ture concerning either the values of these parameters or the values of the variables affecting district potentials, 1f any of these values change, a new optimal national salesforce size and district allocation will emerge. With respect to the parameters, this implies that ideally the estimation should be redone each year and even then some subjective changes to the parameters may be needed, For example, if a new product is about to be introduced, shifts in the parameters could be antici- pated. In terms of the levels of the variables affecting district potentials, forecasted values for the PROSP and COMPT measures should be used as should current year values for LCUST measures. ‘The operations of salesforces A and B were actually ‘examined as parts of two separate consulting, projects 317HORSKY AND NELSON ficient Frontier enchanking some time ago. At that time the “Top Down" approach discussed in this paper was conducted on the basis of the REG estimation only. For Firm A, it was concluded, that its current salesforce size and allocation was nearly “optimal. The national sales manager did, however, view the analysis as informative and requested that it be re- done the following year using the current year’s data to assess whether a newly implemented compensation scheme was effective. This estimation revealed that in- deed the salespeople parameter, y, increased in size, The implication being a small increase in salesforce size was warranted. Consequently, the size of the force was increased with the new cohort being allocated to dis- tricts based on the constrained allocation formulation, (12). For Firm B, it was concluded that the size of the salesforce should be gradually increased. Formulation (12) again was utilized for this purpose. In both firms, the national sales managers expressed great interest in the identity of the under-performing and over- performing districts. After identifying the “good” and "bad" districts (based on the REG analysis) a rather speculative discussion followed as to the likely causes of the productivity differences. Several years later the benchmarking analysis described in this paper was pur sued by the authors and the results presented to the ‘managers of Firm A. These managers found our conclu sions as to the likely causes of salesforce inefficiency to be particularly insightful and plausible. In addition, they pointed out that several firms have recently broken their salesforce into two groups—one for maintaining, existing accounts (“farming”) and the other for pursu- ing new accounts (“hunting”)—in recognition of the fact that some salespeople are better at turning pros- pects into customers than others 5. Summary and Other Applications In this study, a procedure for evaluating the size allo- cation and productivity of a salesforce was suggested and illustrated, This “Top Down” procedure ean be tse by management in parallel fo, and as an objective check of, the more conventional and more subjective “Bottom Up” approach to salesforce effort allocation, Even though this procedure can be viewed as a macro- level decision tool, it can also generate interesting ricro-level managerial insights. 318 In the first stage of the “Top Down” procedure, linear programming is used to estimate a district-level, effi- Cient frontier sales response function reflecting the re- lationship of sales to salesforce size and to other vari- ables that exists only in the top performing districts. ‘This allows the firm to benchmark each district’s sales against a sales estimate based on performance that is in line with that of the best performing districts. Moreover, comparison of the average regression and efficient fron- tier estimation procedures points to an important meth- codological finding, The use of multiple estimation re- sults may lead to an improved understanding of the phenomenon being studied (in our case, the identifica tion of the likely causes of district productivity ineff iencies). The estimation stage of the “Top Down” pro cedure is followed by a second stage in which the op- timal reallocation of the current salesforce and the optimal salesforce size are determined and the profit- ability gains associated with these actions and with im: proving efficiency are evaluated, ‘The proposed “Top Down procedure was illustrated for two salesforces. In both cases, the regression-based analysis would have resulted in a declaration that the status quo was close to optimal, while the frontier-based analysis pointed out that strong gains were possible in certain districts. Given both methods of analysis, a com: parison of their parameters made it possible to identify which specific changes in the daily operation of the salesforce would allow the realization of these potential productivity gains. ‘The benchmarking methodology proposed in this study can be useful in other contexts as well. In partic ular, in smaller salesforces, individual salesperson per~ formance and territory level response functions may be evaluated. Moreover, the methodology can be applied to nonsalesforce related situations in which other mar- keting efforts, such as advertising, are expended with different degrees of success. [Appendix A: A Bootstrap Based Significance Test ‘The idea isto approximate the standard deviations of the efficent ‘eontier parameters using a computational approach. Given an est mate ofa parameters standard deviation, a simple tor normal distr bation based statistical sigifcance test follows, A common technique used in such situations i bootstrapping (Efron and Gong, 1983, Elron and Tibshirani 1993). This procedure generates information conceen- tng.the parameters standard deviation through repeated estimations MARKETING ScIENCE/Vol. 1, No. 4, 1996HORSKY AND NELSON ficient Frontier Benchmarking sing bootstrap samples ofthe N available observations. First the co- efficients ofthe efficient frontier sles response function. = (8, $1. and corresponding eror terms, are estimated following problem (6), Then B diferent bootstrap samples are generated by “boottrap- ping the residuals.” Each bootstrap sample generate 3s follows. To the estimated fontice sales level for each distr an error term is ap pended, Ths err term «(b) i selected randomly with replacement from the p, generated by (6). Ths results in a bootstrap sample of district sales levels, Sale.b) = PSPEOPe%. These bootstrap sample sales levels are then related to the independent variables via problem (6) the result being 2 bootstrap replicate of the sales response coe cents, (6. Note that each bootstrap sample involves diferent Ss es(b) and, hence, results in different parameter estimates. The stan dard deviation of those bootstrap replicates fora particular parameter 4, an estimate of standard deviation of? We [Eee -$(Eemy] ' 1000. Consequently, test statistic forthe hypothesis that equals an (an “Ths statistic converges weakly to the standatd normal distribution. Horsky and Nelson (19951 demonstrate the ability ofthis est to iden: Ly sigutican parameters in the context of linear programming based cetimation of an lfcency frontier. Note thatthe bootstrap test statistic desenbed above is the most basic. More advanced and computationally intensive methods that kenerate more accurate bootstrap confidence intervals are aimed at accounting for things like bas and skewnes in the distribution ofthe bootstrap replicates ee Ffron and Tibsiran 1993). Wetried the mest ‘commonly used procedure ofthis type, the percentile interval method, And, inoue context, found its performance tobe similar to that of (A.2). Inother applications, mote sophisticated methods than (A.2) may be needed? "The procedure proposed in Horsky and Nelson (1995) was reviewed and ceric by the Meleing Science reviewers ofthis manuscript References Basu, A, R Lal, V Srinivasan and R Staelin (1985), “Salesforce Com- pensation Plans: An Agency Theoretic Perspective,” Marketing Science, 4, 267-291 Bauer, P. W. (1990), "Recent Developments in the Econometric Estimation of Frontiers,” Journal of Econometrics, 46, 39~ 56 Beswick, CA. and D. W. Cravens (1977), "A Mulistage Decision “Model fr Salesforce Management” Joural of Marketing Re search, 14, 135-148 MARKETING ScIENCE/Vol. 15, No. 4, 1996 CCharkravati,D, A. Mitchel and R Staelin (1981), "“udgement- based “Marketing Decsion Models: Problems and Possible Solutions’ Journal of Marketing 45, 13-23. ‘Chames, A, W-W. Cooper and T. Sueyoshi (1988), “A Goal Program: rming/Constrained Regression Review of the Bell System Breakup,” Management Science, 34, 1-26 Coughlin, A and S Sen (198), "Salesforce Compensation: Theory and Managenal leplications,” Marketing Science, 8, 324-342 Davis 0. A and JU. Farley (1970, “Allocating Sas Force Effort with Commissions and Quotas.” Managem Sim, 18, 55-6, Eastlack and A. Roo (1986), "Modeling Response to Advertising and Pricing Changes fr ‘V.8" Cocktail Vegetable Juice,” Mir Ieting Science, 5, 245-259 hom B. ane G. Gong (198), “A Lisurely Look tthe Bootstrap the Jack rife and Cros-Valdatin” The Amma Satan, 57, 3-48 — and R |. Tibshirani (1983), Am Intraduction to the Bootstrap, New York Chapman and Hall Evans, D.S.and J.J. Heckman (1988), “Natural Monopoly and the ell System Response to Charnes, Cooper and Sueyos,” Man agement Science, 27-38 Farley | U.(1968), "An Optimal Pl fr Salesmen’s Compencation,” Journal of Marketing Research, 1. 39-18. Pudge, W-K.and LM. Lodish (1977), “Evaluation ofthe fectiveness ‘of Mode! Based Selesman's Planning System by Fld Exper: limentation,” Inerfces, 8, 97-106, Hanssens, DM, L: Parsons, and RL Schulte (1990), Market Re: sponse Modes Econometric and Time Series Analysis, Boston, MA, Kluwer. Hess, 8. W. and S. A. Samuels (1971, “Experiences with a Sales Dis: ‘rcting Model: Criteria and Implementation.” Management Sc cence, 1, 41-54 Horsky, D. and P. Nelson (1991), “Determination of Sales Potential for Salesforce Districts," TIMS Marketing Science Conference, University of Delaware Dupont and — (1995), "Testing the Significance of Linear Program ming Estimators,” Working Paper, Wm. E, Simon Gradu- ate School of Business Administration, University of Rock Lal, Rand V. Srinivasan (1993), “Compensation Plans for Single and Malti-Product Slesforces: An Application ofthe Holmstrom Milgrom Model" Management Science, 39, 777-193 Lilien, G. (1979), “Advisor 2: Modeling the Marketing Mix for Indus trial Products" Management Science, 25, 191-204. ods, L: M. (1971), "CALLPLAN: Aa Interactive Salesman’s Call Planning System,” Management Science, 18, 25-40 0978), Vapuely Right” Approach to Sales Force Allocations” Harsand Business Remew, 52, 19-126 — (1975), “Sales Teritory Alignment to Maximize Profit” Joursl of Marketing Research, 12, 30-36 — (976), “Assigning Salesmen to Accounts to Maximize Profit, Fournal of Marcting Rescerch, 13, 40-144 (1980), "A User Oriented Model for Sales Force Size, Product and Market Allocation Decisions," Journal of Marketing, 44, 70-78 319HORSKY AND NELSON Lodish, LM, & Curtis, N. Ness, and M.K. Simpson (1988), “Sales Force Sizing and Deployment Using a Decision Calculus Model at Syntex Labocatories,”Inerfacs, 18, 5-20 Lucas, HC, CB Weinberg, and K. Clowes (1975), “Sales Re- sponse asa Function of Territorial Potential and Sales Rep esentative Workload,” Journal of Marketing Research, 12 298-305, “Mahajan, J.(1991), “A Data Envelopment Analytic Model for Assess- ing the Relative Efficiency of the Seling Function,” European Journal of Operational Research, $3, 189-205. Mantrala, MK, Sinha, and A. Zotners (1982), “Impact of Resource Allocation Rules on Marketing investment Level Decisions and Profitability,” Journal of Marksting Research, 29, 162-175. Montgomery, DB, A.J. Sik, and C.F. Zaragora (1971), “A Mulapl- Prt Sales Force Allocation Medel” Mansgenant Sc, 18,3 24 Parsons, L. (1990), “Assessing Salesforce Performance with Data Envelopment Analysis” TIMS Marketing Science Conference, University of ins, Urbana, — (1991), "Esimation of Frontier Production Function fora Sales: force," TIMS Marketing Science Conference, University of Del aware/ Dupont. — (1994), "Productivity vs. Relative Etfictency in Marketing,” in Laurent, Lilln and Pras, (Eds), Research Traditions in Marke ing, Boston, MA. Kluwer ficient Frontier enchmarking and P. Vanden Abeee (1981), “Analysis of Sales Call fective: es" Journal of Marketing Research, 18, 107-113. Rangasowamy, A. P. Sinha, and A. Zoliners (1990), “An Integrated Model Based Approach for Sales Fore Restacturng,” Markt Ing Science, 9, 279-298 Rao, R (1990), "Compensating Heterogeneous Salesforce: Some Ex plicit Solutions.” Markting Scene 9, 319-341 Ryans, A.B. and C. B. Weinberg (1979), “Sales Territory Response Fowrnal of Marvin Research, 16, 453-46, —and — (1987), "Terrtory Sales Response Models: Stability Ov Journal of Marketing. Research, 24, 229- 233, Seiford, LM. and RM. Thrall (1990), “Recent Developments in DEA: ‘The Mathematical Programming Approach o Frontier Analy 138" Jouralof Econometrics, 6, 7-36 ‘Stinivasa,V. (1981, “An vestigation ofthe Equal Commision Rae Po cy fora Mulprxtct Salesoree” Mangement Se, 27, 731-756 Zoltners, A, (1978), “Teger Programming Models for Sales Territory Alignment of Manimize Profi" Journal of Marketing Research, 1326-450 — and Sinha (1983), "Sales Temtory Alignment: A Review and Model" Management Science, 29, 1257-1256, — and — (1936, “Sizing and Struceuring Sling Organizations: Insights rom Practice,” Talk given at INFORMS Marketing Sei- ‘ence Conference, Gainesville, Florida Time, ‘Ths per was ected June 13, 1994, ad has Been withthe authors 9 mnths for 2 reson proves by Gary L. Lilien. MaRketine SciENCE/Vol. 15, No. 4, 1996PUBLISHED PAPERS “Multiattribute Utility Models”, Elgar Companion to Consumer Research and Economic Psychology, (1999), 392-400. “Positioning”, Elgar Companion to Consumer Research and Economic Psychology, (1999), 445-453. “Software Acquisition: The Custom/Package and Insource/Outsource Dimensions”, with Bill Richmond and Avi Seidman, Advances in Computers, 47 (1998), 342-367. “Evaluation of Salesforce Size and Productivity Through Efficient Frontier Benchmarking”, with Dan Horsky, Marketing Science, Vol. 15, No. 4 (1996), 301-320. “Belief Change as Change in Epistemic Entrenchment”, with Abhaya Nayak and Hanan Polansky, Synthese, 109 (November 1996), 143-174. “Two Dimensions of Software Acquisition”, with Bill Richmond and Avi Seidman, Communications of the ACM, 39,7 (July 1996), 29-35. “New Brand Positioning and Pricing in an Oligopolistic Market”, with Dan Horsky, Marketing Science, Vol. 11, No.2 (Spring 1992), 133-153. Winner of John D. C. Little Best Paper Award, 1992; Winner of the Frank M. Bass Dissertation Paper Award, 1992. “A Comparison of Ranking, Rating and Reservation Price Measurement in Conjoint Analysis”, with Shlomo Kalish, Marketing Letters, Vol. 2, No. 4 (1991), 327-335. WORKING PAPERS “Stating Preference for the Ethereal but Choosing the Concrete: How the Tangibility of Attributes Affects Attribute Weighting in Value Elicitation and Choice”, with Steve Posavac and Dan Horsky, under review at Journal of Consumer Research. “Testing the Significance of Linear Programming Estimators”, with Dan Horsky, under revision for resubmission to Management Science. “Attribute Importance Weights and Retail Placement”, with Debbie Desrochers. “Product Line Considerations in New Product Development”, with Dan Horsky. WORKING PAPERS continued “Bot Commerce: Implications of Markets Utilizing Search Engines”, with Rajiv Dewan. “Competitive Multiattribute Demand, Cost and Repositioning”, with Jeongwen Chiang and Dan Horsky.“Determinants of Information System Replacement”, “Price and the Multiattribute Model”. CURRENT RESEARCH “Information Processing Affects on Reservation Price Measurement”. “Shelf Space Arrangement and Attribute Level Perceptions”, PRESENTATIONS “st Preference for the Ethereal but Choosing the Concrete: How the Tangibility of Attributes Affects Attribute Weighting in Value Elicitation and Choice”, with Steve Posavac and Dan Horsky, INFORMS Marketing Science Conference in Wiesbaden, June 2001 “Impact of Search Engine Ownership on Underlying Market for Goods and Services,” with Rajiv Dewan and Marshall Freimer, Hawaii International Conference on System Sciences in Maui, January, 2001 “Product Line Considerations in New Product Development”, with Dan Horsky, INFORMS Marketing Science Conference in Los Angeles, June 2000. “Bot Commerce: Implications of Markets Utilizing Search Engines”, with Rajiv Dewan, Pennsylvania State University, April 2000. “Creating a Brand to Fit Your Niche,” The Association for Women in Communications Professional Development Conference, Rochester, May, 1999. “Market Implications of Electronic Shopping”, with Rajiv Dewan, University of California - Irvine, October 1998. “Product Line Positioning and Pricing under Competition”, with Dan Horsky, BCRST. Conference in Syracuse, March 1998. “Product Line Positioning and Pricing under Competition”, with Dan Horsky, INFORMS Marketing Science Conference in San Francisco, March 1997. PRESENTATIONS continued “Determinants of Software System Replacement”, with Bill Richmond, Conference on Advanced Information Technologies, Syracuse, New York, October 1996. “Price and the Multiattribute Model”, with Dan Horsky, INFORMS Marketing Science Conference in Gainesville, March 1996. “Evaluation of Salesforce Size and Productivity Through Efficient Frontier Benchmarking”, with Dan Horsky, INFORMS Marketing Science Conference in Gainesville, March 1996,“Product Line Positioning and Pricing under Competition”, with Dan Horsky, ORSAVTIMS Marketing Science Conference in Sydney, July 1995 “Information Technology and Outsourcing - Some Empirical Results on the Changing Boundary of the Firm”, with Bill Richmond and Avi Seidman, Conference on Information Technology and the Changing Boundaries of the Firm, Philadelphia, January 1995. “Outsourcing Software Development Projects: Does Practice Follow Theory”, with Bill Richmond and Avi Seidman, WISE "94, Vancouver, December 1994. “Evaluation of Salesforce Size and Productivity Through Efficient Frontier Benchmarking”, with Dan Horsky, ORSA/TIMS Marketing Science Conference, Tucson, March 1994, “Belief Change as Change in Epistemic Entrenchment”, with Abhaya Nayak and Hanan Polansky, AI'93 Conference: Belief Revision: Bridging the Gap Between Theory and Practice in Melbourne, November 1993. “Evaluation of Salesforce Size and Productivity Through Efficient Frontier Benchmarking”, with Dan Horsky, ORSA/TIMS Joint National Meeting, Phoenix, November 1993. “Cross-Substitution Effects and Product Cannibalization”, with Jeongwen Chiang, TIMS Special Interest Conference: New Directions and Current Issues in the Analysis and Use of Scanner Data in Toronto, September 1993, “Cross-Substitution Effects and Product Cannibalization”, with Jeongwen Chiang, ORSA/TIMS Conference in San Francisco, October 1992. “Product Line Positioning and Pricing under Competition”, with David Braden and Dan Horsky, ORSA/TIMS Marketing Science Conference in London, July 1992. PRESENTATIONS continued “An Empirical Investigation of District Potential and Salesforce Effectiveness,” with Dan Horsky, ORSA/TIMS Marketing Science Conference in Wilmington, March 1991 “The Effect of Product Line Considerations and Consumer Dynamics on Optimal Brand Positioning and Pricing,” ORSA/TIMS Marketing Science Conference in Urbana, March 1990. ‘The Updating of Attribute Weights as Preferences Tum into Choice,” with Dan Horsky, National ORSA/TIMS Meeting in New York, October 1989 “New Brand Positioning and Pricing in an Oligopolistic Market”, with Dan Horsky, ORSA/TIMS Marketing Science Conference in Durham, March 1989 “The Impact of Imperfect Data on the Performance of MDS Models: A Comprehensive ‘Monte Carlo Study”, with Peter Crosbie and Patrick Dupareq, ORSA/TIMS Marketing Science Conference in Durham, March 1989“Testing the Significance of Attribute Weights Estimated from Ordinal Preference Data”, with Dan Horsky, National ORSA/TIMS Meeting in Denver, October 1988 “New Brand Positioning in an Oligopolistic Market”, with Dan Horsky, ORSA/TIMS. Marketing Science Conference in Seattle, March 1988, “An Empirical Evaluation of Multiattribute Utility and Reservation Price Measurement”, with Shlomo Kalish, ORSA/TIMS Marketing Science Conference in Seattle, March 1988 “Estimation of Multiattribute Utility and Cost Functions”, with Dan Horsky, National ORSA/TIMS Meeting in St. Louis, October 1987 PROFESSIONAL ACTIVITIES Referee for Marketing Science, Management Science, Journal of Marketing Research, Journal of Consumer Research, Journal of Operations Management, Journal of Business and Psychology and the Communications of the ACM Session chair at various Marketing Science INFORMS Conferences Member of INFORMS, INFORMS College on Marketing, American Marketing Association and Product Development and Management AssociationDOCTORAL DISSERTATION COMMITTEES Chair, Debra Desrochers, “Product Evaluations and Retail Placement,” 1999 Member, Debabrata Talukdar, “A Theoretical and Empirical Investigation into Consumers Brand Choice Process Under Information Uncertainty,”1995 Member, Chi Kin (Bennett) Yim, “Price Expectations and Optimal Sales Promotion Policies,” 1989 ADMINISTRATIVE AND SERVICE ACTIVITIES Member, Administrative Committee, William E. Simon Graduate School of Business Administration, University of Rochester, 2000 - present Member, Ph.D. Program Review Committee, William E. Simon Graduate School of Business Administration, University of Rochester, 1996 - 1999 Member, Deans Advisory Committee, William E. Simon Graduate School of Business Administration, University of Rochester, 1997 - 1998 Member, Committee on Teaching Excellence, William E. Simon Graduate School of Business Administration, University of Rochester, 1996 - 1997 Member, Committee on MBA Programs, William E. Simon Graduate School of Business Administration, University of Rochester, 1993 - 1996 Member, Computing Committee, William E. Simon Graduate School of Business Administration, University of Rochester, 1994 - 1995 Member, Ph.D. Committee, William E. Simon Graduate School of Business Administration, University of Rochester, 1991 - 1994 Member, Quality Management Committee, William E. Simon Graduate School of Business Administration, University of Rochester, 1992 - 1993 Member, Administrative Committee, William E. Simon Graduate School of Business Administration, University of Rochester, 1990 - 1993 Member, Ph.D. Program Admissions Committee, Krannert Graduate School of ‘Management, Purdue University, 1987-1990 Member, Masters Program Admissions Committee, Krannert Graduate School of ‘Management, Purdue University, 1987-1988 SELECTED AWARDS AND HONORS John D. C. Little Best Paper Award, 1992 Frank M. Bass Dissertation Paper Award, 1992 Beta Gamma Sigma (Business Honor Society), 1989 American Marketing Association Doctoral Consortium Fellow, 1985Phi Beta Kappa, 1981 Omicron Delta Epsilon (Economics Honor Society), 1981 Pi Mu Epsilon (Mathematics Honor Society), 1981 ‘National Merit Scholar, 1978