TIMOTHY M.

SMITH, SRINATH GOPALAKRISHNA, and RABIKAR CHATTERJEE*

The marketing and sales functions in many firms are often at odds despite their common goal of increasing revenue and profit. The finger pointing goes both ways: Marketing complains of poor lead follow-up by sales, and in turn, sales grumbles about the quality of leads generated by marketing. This disconnect can be damaging; high lead volumes generated through effective marketing campaigns could actually hurt downstream sales because of wasted effort on poorly qualified leads and/or delays in sales follow-up resulting from limited sales force capacity. To examine the revenue and profit implications of coordinated communications efforts at the marketingsales interface, the authors develop a three-stage model that captures the effects of sequential marketing/sales communications on lead generation, appointment conversion, and sales closure. The results, which are based on a collaborative effort with a large home improvement retailer, suggest a complex interplay among marketing efforts (multiple media that generate leads), delays in follow-up (time lag between inquiry and sales force contact), and sales efficiencies (appointment and sales conversion). The findings underscore the impact of multimedia spending on the timing and effectiveness of subsequent communications, implying that improved internal collaboration between marketing and sales can offer significant upside potential for the firm. Finally, the authors develop a managerial decision support tool to simulate the impact of varying communications budgets, timing, and allocation on the marketing and sales planning system.

A Three-Stage Model of Integrated Marketing Communications at the MarketingSales Interface

The effective integration of various elements of the marketing communications mix is an important challenge for practitioners and academics alike. Corporations spend heavily trying to communicate with their current and prospective
*Timothy M. Smith is Associate Professor of Corporate Environmental Management and Adjunct Professor of Marketing/Logistics Management, Carlson School of Management, University of Minnesota (e-mail: timsmith@umn.edu). Srinath Gopalakrishna is Associate Professor of Marketing, University of Missouri, Columbia (e-mail: srinath@missouri. edu). Rabikar Chatterjee is Professor of Marketing, Katz Graduate School of Business, University of Pittsburgh (e-mail: rabikar@katz.pitt.edu). The authors contributed equally to this article and thank the cooperating partner firm and the Forest Products Management Development Institute for their support of this research, as well as the contributions of Sergio MolinaMurillo in the development of the decision support tool. They also acknowledge the helpful comments from Leigh McAlister, special issue editor of the Marketing Science Institute Special Issue, and the three anonymous JMR reviewers.

customers, often through multiple media. Although previous research has examined the effectiveness of individual elements, such as advertising (Hanssens and Weitz 1980) and trade shows (Gopalakrishna and Lilien 1995), the synergy across elements is also important; that is, spending on one source may enhance the effectiveness of another (Gatignon and Hanssens 1987; Gopalakrishna and Chatterjee 1992). This perspective, called integrated marketing communications (IMC), has received some attention in the literature, though research in this area is scarce (Naik and Raman 2003). In practice, the impact of communications activity on the purchasing process is often sequential, especially when the process involves multiple stages. However, the timing of exposure to sequential communications is a key unexplored issue in the context of effective resource deployment. Our research stems from discussions with a major home improvement retailer, which we refer to as HIR for confiJournal of Marketing Research Vol. XLIII (November 2006), 564579

2006, American Marketing Association ISSN: 0022-2437 (print), 1547-7193 (electronic)

564

Integrated Marketing Communications at the MarketingSales Interface dentiality reasons. The firm communicates with its target market for installed home improvement products through several channelsprincipally, direct mail, radio advertising, newspaper, and trade shows. Leads generated from these sources are followed up (by prior appointment) with a sales call. The management at HIR implicitly believes that all leads are important and require prompt sales force follow-up; thus, customer appointments for sales visits are set up (through the call center) at the earliest mutually acceptable time slot. However, the limited sales force capacity creates significant delays in the salespeoples inhome visits, especially in the high season. The longer wait can result in a lost sales opportunity due to the prospects declining interest. Such declines in purchase likelihood bear some relationship to the lead generation process at the front end of the system. The aforementioned scenario, which motivates our conceptual development and analysis, lies at the heart of a hotly debated issue: the marketingsales disconnect. On the surface, the marketingsales relationship seems symbiotic and complementary, though in practice, coordinating the two functions is rarely an easy task. In many organizations, this coordination is limited to handing off information, creating tensions over resources and questions about both functions roles in enhancing customer experiences and the impact on the bottom line (Marketing Science Institute research priorities 20042006). The business press extols the virtues of marketingsales harmony, but industry reports suggest that the sales force ignores as many as 70% of leads that marketing generates (Watkins 2003). In turn, salespeople argue that marketing managers have become isolated from the sales process and cannot discern a good lead that is worthy of timely follow-up. Such observations highlight the practical problems in effectively integrating all communication activities. Management at HIR asked us to examine the complex problem of developing an effective and integrated communications mix. Leads generated by marketing communications were creating stress for the salespeople because of the uneven volume. As Figure 1 shows, the average time to service leads closely follows weekly lead volume. In addition, the data imply that a longer customer wait causes a Figure 1
ACTUAL LEADS CREATED AND FOLLOW-UP DELAY

565

lower purchase likelihood. Leads followed up within five days of being identified are converted into sales at 20% higher rates than those that wait longer than ten days for a sales visit. Thus, increasing marketing spending and, consequently, lead volume may not be worthwhile. Similarly, the timing of marketing expenditures and allocation across sources may involve desirable spending levels that maintain the overall balance. In addition, delays in servicing leads can persist for several weeks as the limited sales force works through the inventory of leads. A likely side effect of such imbalances in the system is the adverse impact of lower lead conversion rates on sales force morale and increased tension between marketing and sales personnel. Our objective is to use the available data to develop and estimate a model that captures the dynamics of the customer buying process and to evaluate the impact of alternative marketing and sales budgets on HIRs performance. To help managers run such simulations as an aid to marketing and sales planning, we develop a user-friendly decision support tool. In the process, we address the following questions:
Which elements of the marketing communications mix are more effective than others in generating leads? What carryover and complementary (interactive) effects, if any, exist? What is the impact of lead volume on follow-up delay (the time lag between lead generation and sales appointment) for the prospect? What is the impact of delay on the likelihood of conversion of a lead to a sales appointment and to subsequent closure of the sale? Does this vary by lead source? What is the impact of seasonality at each stage of the buying process? How do variables such as salesperson quality and prospect characteristics affect the process?

We first review the literature in the IMC area and discuss the managerial context and data available. We then describe our model in terms of a conceptual framework that examines the sales process as a sequence of three stages: lead generation, appointment conversion, and sales closure. Next, we discuss the estimation results of the multistage model and report on model validation. We then describe the tool to help managers assess the impact of different media expenditures on outcomes (based on the model parameter estimates) and provide some illustrations. We conclude by summarizing the contributions, managerial implications, and avenues for further research. IMC Although marketers intuitively embrace the integration (IMC) perspective, empirical research in this area is scarce. For planning and budgeting, practitioners have acknowledged the interaction among marketing communication elements (Acheson 1993; Morrill 1970), though leveraging this synergy remains an unexplored area. The IMC framework is built on the foundation that if multiple communications are deployed appropriately, they can enhance one anothers contributions (Belch and Belch 2003). The IMC concept has been characterized as both a relational process and a business competency (Reid 2003). Its goals and outcomes are often linked to building relationships with customers and other stakeholders through ongoing dialogue and the subsequent effects on sales and profits (Duncan and

566

JOURNAL OF MARKETING RESEARCH, NOVEMBER 2006 tions and sales force deployments) and seasonality on marketing resource deployment. THE MANAGERIAL CONTEXT HIR is a large manufacturer in the United States that produces quality replacement windows promoted through multiple media and sold through a direct sales process. The firm provides a unique product/service offering in the industry and communicates with its target market for installed home improvement products through direct mail, radio advertising, newspaper, trade shows, and so forth. Sales leads generated from these sources are followed up by a sales visit if a prior appointment is set up. The nature of the product is such that a sales visit at the prospects home must happen before a sale can be concluded. Typically, the salesperson ascertains the specific needs, offers a quote for window replacement, and, if possible, tries to close the sale in that one visit. The data pertain to the lead generation and sales processes in a major metropolitan market. Specifically, inquiries in 2002 and 2003 frame the sample; there were a total of 19,496 inquiries, 16,309 in-home sales visits, and 6068 purchasing events. At the prospect/customer level, the disaggregated data set provides information about the date of initial inquiry, the marketing medium/campaign prompting the inquiry, sales units anticipated, and the date of the in-home sales visit. Note that the marketing medium prompting the prospects initial inquiry is a critical piece of information; each medium specifically directs the prospect/ customer to call a toll-free number (unique to that medium and campaign) to arrange for a free in-home estimate. We examine nine separate sources of leads: print advertisements (newsprint and newspaper supplements), direct mail (individual mailings and marriage mail inserts), exhibition and event sponsorships (consumer-oriented home and garden shows and booths at various local events), radio advertisements, telephone directories, Internet communications, retail showrooms, referral programs, and repeat business. These sources account for more than 98% of the leads in our analysis and cover all the marketing communications expenditures incurred in this market. Communications expenditures were available by media type, along with archival records of purchase transactions and zip codelevel demographic data (provided by a third party). HIR manages media expenditures on a weekly basis, with an emphasis on key metrics, such as cost per lead and cost per appointment. Transactional data for all customer purchases are at the individual (household) level. Data from this source include matching customer information, purchase date, units purchased, and purchase amount. In addition, demographic data (estimates of home value, age of home, age of head of household, household income, and length of residence) were available at the zip code (not household) level. These data enable us to explore a process that begins with impersonal marketing communications, which in turn generate customer inquiry (contact with the call center); this is followed by a sales call (after a time lag) and culminates in a potential sale. We describe the process in more detail in the next section, but it is important to recognize the link between multimedia expenditures and the delay between inquiry and sales appointment (see Figure 1). The delay adversely affects the purchase likelihood because consumer

Caywood 1996; Smith, Gopalakrishna, and Smith 2004). Furthermore, as a business competency, IMC suggests integrated management of multiple media to achieve superior outcomes (Duncan 2002; Naik and Raman 2003; Naik, Raman, and Winer 2005). Synergy among multiple media in dynamic settings is of strategic importance. Naik and Raman (2003) propose a model that incorporates synergistic effects in dynamic budgeting decisions. They provide important, managerially relevant insights; however, the analysis is based on monthly data at the market level, ignoring the sequential realities of communication exposure at the individual level. Naik, Raman, and Winer (2005) extend this work to include competition in an extended Lanchester model that allows for interaction between advertising and promotion. Again, although this research makes important theoretical and empirical contributions, it does not consider lead and lag effects or the implications of sequential communications. In the business press, Schultz (2004) discusses the role of simultaneous media in communications planning to achieve possible synergies. However, even in the most technology-driven settings, multiple communications (e.g., advertising and personal selling) may be most effective in a sequential rather than simultaneous setting. The time lag between two sequential communications affects their synergistic impact and the effectiveness of the overall deployment in terms of bottom-line outcomes. Interactions among marketing variables form an important aspect of our understanding of marketing effectiveness. Although recent work in IMC has focused on the advertising domain, researchers have explored the role of interactions across the marketing and sales divide. Several studies in the business press and academic literature have documented the role of advertising in terms of creating a favorable climate for the sales call (Couretas 1984; Levitt 1967; Morrill 1970). Indeed, the McGraw-Hill man-in-chair advertisement, considered a business marketing classic, suggests that advertising can have a positive impact on sales force efficiency. With regard to resource allocation, several studies explore the mix of personal and impersonal communications expenditures, but the analyses are often at the market level and limited to two sources (e.g., sales force and advertising, direct mail and advertising). Gatignon and Hanssens (1987) explore the optimal ratio of personal and mass communications expenditures numerically within a static (single-period) case. Gopalakrishna and Chatterjee (1992) assess the joint impact of advertising and sales force expenditures through a dynamic sales response model. Smith, Gopalakrishna, and Smith (2004) assess the complementary effect of trade shows on sales force performance and suggest normative implications for optimizing sales force allocations based on previous communications exposure. In summary, the literature offers limited empirical and theoretical insights into the process of integrating communications. Specifically, there is little help for marketing managers in planning communications strategies across multiple media and understanding their combined impact on sales force effectiveness. We build on these gaps by exploring the dynamics of IMC, including not only the carryover and interaction effects of lead-generating media but also the effects of decay in sales force effectiveness (stemming from the time-lag between media communica-

Integrated Marketing Communications at the MarketingSales Interface decisions to invest in home improvement tend to be transient, competing with alternative uses of the available funds. Significant delays in the sales visit, associated with high lead volume, warrant careful analysis of the allocation/ timing of lead-generating communications that best complement follow-up sales effort. THE MODEL We model the sales process as a series of stages marked by concrete outcomes. They are (1) the generation of sales leads, (2) the conversion of leads into sales appointments, and (3) the conversion of appointments into sales. Figure 2 displays this process as a sequence of stages, marked by the actions of both the firm and its prospective customers. We briefly describe the dynamics of the process leading to the outcome (i.e., sales leads, appointments, or actual sales orders) for each stage. In adopting model specifications and underlying functional forms, we were guided by the objective of capturing the essential relationships and dynamics consistent with the underlying phenomena and with the patterns exhibited by the data in a reasonably parsimonious and robust manner. We considered several viable candidate forms, and the final choices best met the criteria of theoretical and descriptive soundness (Lilien, Kotler, and Moorthy 1992, p. 674). Stage 1: Lead Generation Sales leads arrive in the form of telephone calls to the companys call center from potential customers who express an interest in HIRs products; typically, these callers want to schedule an appointment for an in-home visit. These leads are triggered primarily by the companys marketing communications through radio and newspaper advertising, direct mail, and exhibitions (consumer-oriented home shows). Leads are also generated by a limited retail/ showroom presence, telephone directories, the Internet,

567

referrals, and repeat business. The communication source that led the prospect to initiate the call is identified through a unique extension within the telecommunications system, or it is obtained verbally from the customer by the call center representative. An examination of the data and insights from HIR managers suggest the following:
Exhibitions produce some carryover effect on lead volume, but the impact of other key media (for which spending is controlled and may be varied over timespecifically, radio, newspapers, and direct mail) is instantaneous, with no discernable carryover effect. There are interactive effects in which the volume of leads generated by one source is affected by communications expenditure on other sources. The impact of communications spending on lead volume shows seasonal variation, given the highly seasonal nature of the business. Rather than a discrete step change from low to high season, there appears to be gradual shoulders at either end of the high season; that is, the impact ramps up (from a low level) at the beginning, reaches a plateau, and ramps down at the end of the season. The two shoulders (ramps) each appear to be approximately 12 weeks long.

On the basis of the preceding discussion, we propose the following model specification for lead generation:
(1a ) L it i L i ( t 1) = 0 i ( 1i

)S ( X it )
t

ii

(1 + X )
jt j = 1, j i

ij

exp e1it , and 13 t if 1 t 12 St = 0 if 13 t 40 , t 40 if 41 t 52

( )

(1b)

Figure 2
SALES PROCESS AS A SEQUENCE OF STAGES MARKED BY CONCRETE OUTCOMES

568

JOURNAL OF MARKETING RESEARCH, NOVEMBER 2006 not all leads are converted into sales appointments. Attrition (nonconversion) may occur because the customer is not ready to schedule an in-house sales visit, the time lag between the call date and the earliest sales-appointment date (delay) is unacceptable to the customer, or the customer cancels a sales visit after it has been scheduled. The latter two factors suggest that attrition increases with delay. Data suggest that the attrition rate is specific to the communications source. Furthermore, the attrition rate is likely to be different in the off-season relative to the high season (though the effect of seasonality on the conversion of leads into appointments should not be source related) and may possibly vary with the total volume of leads. Given these considerations, the conversion rate of leads into appointments can be modeled as follows:
A it S S = 0 i ( 1 ) 1 ( 2 ) 2 ( L it ) i Lag t L it

where i indexes the communications source uniquely identified by the system or verbally by the customer (i = 1, , N); t indexes the time period (in weeks, t = 1, , 52, with t = 1 denoting Week 1 of the calendar year); Lit is the lead volume (number of customer calls) attributed to source i in week t; Xit is communications expenditure (in dollars) for source i in week t; 0i, 1i, ii, ij, and i are the lead generation model parameters (to be estimated); and e1it is the error term, which is assumed to be distributed N(0, 1i) and possibly correlated serially and across sources. The parameter 0i captures the scale effect in the sense that, all else being equal, a higher value of this parameter implies a proportionately higher number of leads generated by the particular source in a given week. The slope of the off-season ramps (on either side of the high season) is captured by 1i; to have upward/downward sloping ramps, 0 < 1i < 1, where smaller values of 1i indicate steeper declines in the ability of communications to generate leads. The parameters allow for nonlinearity in the main and interactive effects. Specifically, 0 < ii < 1 would suggest a concave response function for the main effect that is positive but with declining marginal impact due to saturation. The term (1 + X ij )ij captures the interactive or synergistic effect of source j on leads generated by source i; ij > (<) 0 would imply positive (negative) synergies, and ij = 0 would signal no interactive effect. Note that adding the 1 in the interaction terms, (1 + X ij )ij , ensures that the main effect ( X it )ii remains intact even if Xjt is zero.1 Finally, the carryover effect, i, is estimated in the cases in which such an effect might exist (exhibitions), but it is assumed to be zero for media sources for which there is no carryover. Furthermore, we observe that HIRs presence in certain major exhibitions (which occur in specific months) is significantly larger in terms of expenditures and scope than its involvement in other (minor) shows. The impact of these major exhibitions in lead generation appears to be qualitatively different from that of the minor ones. Accordingly, we modify Equation 1a in the case of exhibitions to reflect this possible difference, as follows:
(1a ) L it i L i ( t 1) = 0 i ( 1i )S ( E )E t ( X it )ii

2i

which can be rearranged to yield the following model specification for sales appointments:
(2) A it = 0 i ( 1 )
S1

( 2 )S2 ( L it )1i ( Lagt ) 2i exp ( e2it ) ,

(1 + X )
jt j = 1, j i

ij

exp ( e1it ) , for i = exhibitions, i

where Et is a dummy variable indicating the type of exhibition; Et = 1 if the exhibition in week t is a major; and Et = 0 if otherwise. On the basis of an inspection of the data, we classified an exhibition as major when its expenditure exceeded $10,500. The parameter E captures the differential impact of a major (versus minor) exhibition on lead generation; we would expect that E > 1 if major exhibitions indeed have a greater impact. Stage 2: Conversion of Leads into Appointments The lead in the form of a telephone call from the customer (outcome of Stage 1) provides the call center with an opportunity to schedule an in-house sales visit. However,
1Our formulation captures nonlinear effects in a constant elasticity specification, in contrast to a bilinear specification, which is linear in Xi for fixed Xj (and vice versa), as in the work of Naik and Raman (2003).

where Lit is as defined previously; Ait is the number of sales appointments converted from leads in week t attributed to source i; Lagt is the median lag time between the leads generated in week t and the resulting sales appointments; S1 and S2 are seasonality indicators, where S1 = 1 if week 0 t 12, and 0 if otherwise, and S2 = 1 if week 41 t 52, and 0 if otherwise; 0i, 1, 2, 1i, and 2i are the conversion model parameters (to be estimated); and e2it is the error term, which is assumed to be distributed N(0, 2) with the possibility of serial correlation. The high-season scale effect, which is captured by 0i, is modified to (0i 1) for the 12-week off-season period before the high season and to (0i 2) for the 12-week offseason period after the high season; note that the off-season adjustment parameters 1 and 2 are assumed to be independent of the lead source. The (possible) nonlinearity in the effects of Lit and Lagt on conversion is captured by 1i and 2i. Because these effects should be positive for Lit and negative for Lagt, we expect that 1i > 0 and 2i < 0. Stage 3: Sales Closure During the sales appointment (outcome of Stage 2), the salesperson collects the information that is necessary to prepare a quote for the customer, after which the customer places an order for supply and installation (the order may or may not result). We model this stage in two parts: (1) the probability of closure (sales order), given that the appointment has taken place, and (2) the size of the sales order, given that an order has been placed. The probability of closure decreases as the time lag between the initial customer contact (sales lead) and the scheduled sales visit increases, similar to the dynamics in Stage 2. Again, the initial probability and decay rate may vary by communication source; our initial scrutiny of the data suggests that this is so. Furthermore, seasonality and the effectiveness of the salesperson who calls on the household can influence the probability of closure. Unlike the sales lead generation and appointment conversion models in Stages 1 and 2, the availability of appropri-

Integrated Marketing Communications at the MarketingSales Interface ate data (household level) enables us to model the probability of sales closure at the individual household level. Finally, rather than the median lag (between sales lead and appointment), we can consider household-specific lags. The data (and discussions with HIR managers) suggest that individual cases of lags that are greater than the corresponding median lag (for that week) tend to be customer driven in nature (i.e., the customer requests a specific appointment date beyond that which HIR offers). Such customer-driven lags will not have the same adverse effect on the probability of closure as HIR-driven lags that arise because of sales force capacity constraint. On the basis of these considerations, we employ the following logit specification to model the probability of closure at the individual household level:
Ph | it (3) ln = 0 i + 1S1 + 2S2 + Y 1 Ph | it + 1i min Laghit , Lag t

569

communications source generating the initial lead. We specify the order-size model as follows:
(4) Sh | i = i ( Vh ) ( )
Y

(Z
l =1

lh

)l exp ( e4 h ) ,

) )

+ 2 i max Laghit , Lag t Lag t +

Z
l l =1

lh

+ e3h ,

where h indexes the individual household; l indexes the household descriptor variable; i and t index the communications source and period, respectively, as before; S1, S2, and Lagt are defined as before;2 Phit is the probability that household h places an order, given that the initial lead was generated by source i in week t; Y is a dummy variable indicating salesperson effectiveness based on prior performance data (Y = 1 if the salesperson is above the median effectiveness level, and 0 if otherwise); Laghit is the lag time for household h between the initial lead in week t attributed to source i and the resulting sales appointment; Zlh is the value of lth household descriptor variable for household h; 0i, 1, 2, , 1i, 2i, and l are parameters (to be estimated); and e3h is the error term, which is assumed to be independent across households and to have an extreme value distribution. The effects of seasonality, salesperson, and the set of household-level descriptor variables are modeled in a straightforward manner, with the parameters 1, 2, , and l; note that we do not hypothesize these effects to be media specific. We assume that any household-level lag beyond the corresponding median lag is customer initiated; we capture this by the two terms with the lag variables. If Laghit < Lagt, these two terms reduce to liLaghit; if Laghit > Lagt, they reduce to 1i Lagt + 2i(Laghit Lagt). Thus, 1i measures the impact of the HIR-driven lag, and 2i captures the impact of the part of the lag beyond the median, which is assumed to be customer initiated. The potential order size (given that an order is placed) is influenced by the anticipated number of units of the potential order (ascertained during the initial telephone call), salesperson effectiveness, household characteristics, and the
2In this household-level model, we track the median lag, Lag , by a spet cific day rather than by week.

where Y and Zlh are as we defined previously; Sh|i is the size of the order from household h, given that an order is placed following a lead generated by source i; Vh is the potential size of the order from household i (as ascertained during the initial customer inquiry); i, , and l are parameters (to be estimated); and e4h is the error term, which is assumed to be independently distributed N(0, 4). The source-specific effect on the order size is captured by i, and and l measure the impact of salesperson effectiveness and householdlevel descriptors, respectively. The stagewise models capture the adverse impact of the lag between the initial telephone call (lead) and the sales appointment on the conversion of leads into appointments and then on the closure of appointments to orders. As we noted previously, a key decision for HIR is the timing of communications effort through the various channels. Greater effort generates more leads, but given sales force capacity constraints, larger lead volume will likely translate into a longer delay before the sales visit, potentially causing attrition in conversion and closure rates. Beyond the stagewise response models represented in Equations 14, we need a model to link the leads generated in Stage 1 to the lags that may affect conversion and closure in Stages 2 and 3. Conceptually, the lag in period t would depend on the current backlog, which is captured by the extent of lag in period (t 1), and the new leads generated in that period. If the volume of these new leads exceeds some steady-state level, the backlog may increase from the current level. This conceptualization translates into the following model specification:
(5) N Lag t = 0 + 1 Lag( t 1) + 2 L it L exp ( e5 t ) , i =1

where Lagt and Lit are as defined previously; L is the average volume of weekly leads generated (over the two years of available data) to capture the steady-state level; 0, 1, and 2 are parameters to be estimated from the data; and e5t is the error term, which is assumed to be distributed N(0, 5) with the possibility of serial correlation. ESTIMATION RESULTS We now report on the coefficient estimates obtained for the aforementioned models, based on the data for 2002 2003 described previously; we also discuss the implications. Versions of model equations 14 used for parameter estimation appear in the Appendix. Stage 1: Lead Generation The parameter estimates of the lead generation model (Equation 1) appear in Table 1. Because the error terms may be correlated across submodels in this stage, we estimated the four source (media type [i.e., direct mail, newspaper, exhibitions, and radio advertising]) models jointly using seemingly unrelated regression (SUR). We focus on these four media because they are directly responsible for the

570 Table 1
SUR ESTIMATES FOR LEAD GENERATION MODELS (EQUATION A1) Coefficient Estimate

JOURNAL OF MARKETING RESEARCH, NOVEMBER 2006 times greater lead volume than a smaller show, all else being equal. Any seasonality effect would indeed be subsumed in this big show effect, given the peak season timing of the major shows. In addition, we find a carryover effect only in the case of exhibitions ( = .54), suggesting strong carryover. Next, we observe that current period expenditures have a significant effect on lead volume in each of the four sources. Note that all the i coefficient estimates are less than 1 (.08 for direct mail, .11 for newspaper, .20 for radio, and .32 for exhibitions), suggesting diminishing returns. We also observe some significant interaction effects; expenditures on radio advertising and direct mail in the current period enhance lead generation through newspapers in the current period (newspaper, radio = 1.047, newspaper, direct mail = 1.073). In addition, radio advertising appears to have a positive impact on exhibition leads, though it only approaches statistical significance (exhibition, radio = 1.201, p = .117).4 As a practical matter, this indicates that during a week when a significant event is to be held, radio advertising may be used to boost attendance at the event and persuade attendees to visit the firms booth as opposed to calling the call center. These observations suggest that various media offer different levels of leverage to other media in the same mix. Thus, the expenditure levels on each element within the mix have an important effect on the leads generated. To illustrate, consider a scenario in the high season in which the firm spends a budget of $45,000, with $15,000 on an exhibition event, $5,000 on radio, $10,000 on newspaper, and $10,000 on direct mail in a particular week. Applying the parameter estimates derived for our model, we find that this spending pattern results in a total of 354 leads for the week. However, the same budget reallocated with $10,000 on radio and $15,000 on direct mail, holding expenditures on the exhibition and newspaper advertising unchanged, results in a total lead volume of 381. Although this allocation produces little by way of a direct effect on radio advertising leads (less than one incremental lead), the indirect effects of increased radio spending on the level of response to newspaper advertisements and exhibition are substantial. Newspaper advertising and exhibition leads increase by 4 and 26 leads, respectively, with no change to their budgets (direct mail leads drop with the reduction in its budget, but only by 2.5 leads). From a managerial perspective, this example underscores the value of integrative planning of expenditures at the front end of the process. Next, we describe the critical role of total lead volume in defining the amount of delay that is generated within the system. Delay: Linking Marketing Efforts and Sales Force Capacity Although leads generated in Stage 1 are central to the maintenance of the downstream productivity of sales personnel, they contribute to delays when sales force capacity is exceeded. Such delays may affect conversion and closure in Stages 2 and 3. The parameter estimates of the linking model (Equation 5) appear in Table 2. There is a strong model fit (adjusted R2 = .744) and statistical significance for all parameter estimates. Specifically, 57.5% of the previous period lag (t 1) carries over into the current period, all
4We have a clear prior hypothesis that the interactive effect will be positive; thus, one-tailed tests of statistical significance are appropriate.

Model and Variable

SE

p Value

System-weighted R2 = .448; n = 414 Direct Mail Intercept Seasonality ramp Ln(direct mail) Ln(newspaper) Ln(exhibition) Ln(radio) Newspaper Intercept Seasonality ramp Ln(direct mail) Ln(newspaper) Ln(exhibition) Ln(radio) Exhibition ( = .54) Intercept Seasonality ramp Large show effect Ln(direct mail) Ln(newspaper) Ln(exhibition) Ln(radio) Radio Intercept Seasonality ramp Ln(direct mail) Ln(newspaper) Ln(exhibition) Ln(radio)
aWe

3.143 .144 .082 .001 .0213 .024 .941 .149 .070 .113 .024 .135 2.653 .069 2.688 .066 .018 .323 .183 1.021 .348 .028 .028 .088 .203

.590 .033 .015 .045 .026 .046 .755 .047 .043 .033 .038 .066 2.073 .109 1.41 .099 .150 .049 .154 1.352 .081 .076 .115 .067 .058

.000 .000 .000a .494a .209a .301a .215 .002 .052a .000a .266a .023a .204 .529 .060 .250a .453a .000a .118a .452 .000 .356a .405a .904a .000a

report the p value for one-tailed tests given a clear prior hypothesis that the interactive effect will be positive; otherwise, we report two-tailed tests.

majority of leads generated by the firm and because their expenditures and allocations are actively managed. The fit is good, with a system weighted R-square value of .448.3 In addition, we observe that the preseason buildup and the postseason decay in sales, which is captured through the ramp coefficient 1, is statistically significant in the case of direct mail, newspaper, and radio. However, note that the ramp effect is different across sources. To illustrate, the ramp coefficient in the case of direct mail is .866 (e.144). Thus, six weeks before the high season begins, the leads generated through direct mail are (.866)6, or 42% of the lead volume in the high season, all else being equal. In the case of newspapers and radio, the corresponding fractions are 41% ([.8626]) and 12% ([.706)6]), respectively, all else being equal. In the case of exhibitions, the seasonality effect is not significant; instead, the impact of large exhibitions relative to minor shows is statistically significant. Specifically, we note that a major exhibition generates approximately 14
3We also estimated the submodels independently using ordinary least squares. The R-square values are .486 for the direct mail submodel, .491 for the newspaper submodel, .479 for the exhibition submodel, and .450 for the radio submodel. The ordinary least squares parameter estimates were similar to SUR estimates in magnitude, direction, and significance.

Integrated Marketing Communications at the MarketingSales Interface Table 2

COEFFICIENT ESTIMATES FOR LINKING MODEL (EQUATION 5) Coefficient Estimate 7.8573 .5751 .0538

571 Table 3

COEFFICIENT ESTIMATES FOR APPOINTMENT GENERATION MODEL (EQUATION A2) Coefficient Estimate .008 .049 .077 .285 .081 .140 .073 .145 .137 .073 .447 .849 .941 .994 .899 .982 .967 .967 .978 .911 .039 .111 .079 .086 .081 .042 .056

Variable Intercept Previous week lag, Lag(t 1) Leads steady state, Lit Li
aWe

SE .238 .012 .020

p Value .000a .000a .004a

Variable Intercept Early season Late season Direct mail (DM) Newspaper advertising (NA) Radio advertising (RA) Telephone directories (TD) Referral programs (RP) Repeat customers (RC) Retail (RT) Internet communications (IC) Ln(leads) DM Ln(leads) EX Ln(leads) NA Ln(leads) RA Ln(leads) TD Ln(leads) RP Ln(leads) RC Ln(leads) RT Ln(leads) IC Ln(lag) EX Ln(lag) NA Ln(lag) RA Ln(lag) TD Ln(lag) RP Ln(lag) RC Ln(lag) RT
aWe

SE .082 .021 .024 .109 .115 .115 .127 .105 .123 .128 .118 .024 .017 .024 .039 .036 .034 .043 .039 .041 .056 .043 .051 .047 .048 .048 .046

p Value .921 .021 .001 .009 .478 .224 .565 .169 .265 .565 .000 .000 .000 .000 .000 .000 .000 .000 .000 .000 .244a .005a .059a .033a .047a .140a .110a

report the p value for one-tailed tests given a clear prior directional hypothesis in the positive direction; otherwise, we report two-tailed tests. Notes: Adjusted R2 = .744; n = 104.

else being equal (1 = .575). New leads generated in the current period in excess of the steady-state level (Lit Li) are positively related to the lag existing in the system in any given week (2 = .054). These findings point out the drawbacks of generating too many leads. Lead volume beyond the capacity of the sales organization results in increased lag time between inquiry and sales force follow-up. Moreover, these delays linger in the system beyond the period in which they are created. For example, an effective home show may generate more leads (for a limited time) than the sales force can handle. Leads not serviced during the week of the show are scheduled in later weeks, forcing leads generated in those weeks farther into the future. This cycle continues until the weekly lead volume falls sufficiently below the steady-state level to allow the sales organization to catch up. Stage 2: Appointment Conversion In Stage 2, we begin to notice issues of lead quality and decay. With regard to appointment conversion, some leads are better than others; however, identifying the better leads is not straightforward. Estimation results for the Stage 2 model appear in Table 3. The overall model fit is strong (adjusted R2 = .96). Parameter estimates indicate that leads are converted into appointments at different rates across many of the sources (1i) and that a predominantly negative delay effect (2i) exists at the appointment conversion stage. With regard to the source-specific relationship between leads and appointments, coefficient estimates of 1i ranged from .849 to .994 across sources (all were statistically significant and different from one another [ = .10]). To illustrate the magnitude of these differences, assuming negligible delay in the system (LagNt = 2 days), we would expect to convert nearly all (99%) of the newspaper leads into sales appointments in the high season. In contrast, with the same time lag and seasonality conditions, radio advertising is expected to convert 77% of leads into appointments. This scenario is presented in Figure 3 for selected sources. Before any substantive time-lag effect, leads generated from newspaper advertising, referral programs, and telephone directories are converted into appointments at higher rates. Exhibitions, direct mail, and radio leads are converted at significantly lower rates. It is rare, especially in the high season, for a sales appointment to be scheduled quickly. Therefore, we explore the potentially negative relationship between appointment conversion and time lag (the number of days from lead creation to sales visit). Our results suggest that seven of the nine sources display a directionally negative lag coefficient,

report the p value for one-tailed tests given a clear prior hypothesis that the time-lag effect will be negative; otherwise, we report two-tailed tests. Notes: Adjusted R2 = .960; n = 889.

and four of them (newspaper, radio, referral programs, and telephone directories) are statistically significant at the .10 level or better (for a fifth, retail showrooms, the p value is .11).5 Direct mail and Internet leads produced positive, but not significant, lag coefficients; therefore, we set the lag coefficients to zero for these sources. Figure 3 plots predicted appointment conversion of 30 hypothetical highseason leads from selected sources to highlight differences in decay. As we noted previously, with a delay of two days, almost all high-season newspaper leads are converted into appointments (99%). If the delay increases to seven days, the conversion rate drops to 86%, and at 3 weeks (common during high season), it reduces to 76%. Similar results hold for referral programs and telephone directories; specifically, 4 weeks after the inquiry, these formerly high-quality leads decay to, at best, average levels. Finally, there is a significant early and late season effect (1 = .049, 2 = .077); leads obtained in the first 12 weeks and the last 12 weeks of the calendar year are converted into 4.8% and 7.4% fewer appointments than those obtained in the high season, respectively. Stage 3a: Sales Closure Estimation results for the sales-closure model (Equation 3) appear in Table 4. Unlike previous models, this is an individual-level (household) specification. We estimate the
5See n. 4; in this case, we expect a decline in conversion over time; that is, our prior hypothesis is that these coefficients are negative.

572 Figure 3
THE EFFECT OF INTERCOMMUNICATION TIME LAG ON APPOINTMENT CONVERSION FOR SELECT LEAD-GENERATING MEDIA

JOURNAL OF MARKETING RESEARCH, NOVEMBER 2006 the same rate, all else being equal, direct mail appointments are less likely to be converted into a sale, whereas appointments generated by referral programs, retail showrooms, and repeat business are converted at higher rates. As in Stage 2, we estimate the source-specific time-lag effects with regard to their impact on sales closure. As we noted previously, appointment dates are negotiated with individual prospects on the basis of their needs and sales force availability. The time lag between inquiry and sales visit can be capacity driven or customer driven. Capacitydriven lag is operationalized as the time lag up to the median lag observed in the system at the time the lead was created (min[Laghit, Lagit]). Therefore, any additional time lag (max[Laghit, Lagit] Lagit) is assumed to be customer driven. Controlling for customer-driven lag effects, we observe significant, negative impacts of increased capacitydriven lag among appointments from exhibitions (1(exhibition) = .014), referral programs (1(referral) = .115), and repeat business (1(repeat) = .151). The 1 coefficients for other sources (except retail) are directionally negative but insignificant. Although customer-driven lag is not under direct management control, note that, in general, accommodating additional customer-driven lag has a positive relationship to the likelihood of sales, except among repeat customers (2(repeat) = .032, p = .012). We speculate that customers may hold a positive view of this extended time as an extension of HIRs well-publicized no-pressure sales approach. Sales closure (the rate of conversion of sales visits into orders) is a key sales performance metric. Figure 4 shows predicted probabilities of sales closure by lead-generating source with varying capacity-driven lag. In this illustrative example, we assume that leads are generated in the high season, households are average (in terms of their background characteristics), prospects are called on equally by high- and low-performing salespeople, and customer-driven lag is ignored. For exhibition and direct mail appointments, when the capacity-driven lag is negligible ([min{Laghit, Lagi}] = 1), direct mail appointments are converted at a lower rate (29.6%) than exhibition leads (34.2%). However, when the sales organization is stressed (i.e., incoming leads exceed capacity) to the point at which the inquiry to sales visit time lag approaches three weeks, an exhibition appointment is converted into a sale at the same rate (28.3%) as its once weaker direct mail counterpart. Of potentially greater concern is the impact of capacitydriven delay on apparently unrelated sources. If marketing activities create enough stress on the sales force to generate even minor levels of capacity-driven lag, the predicted sales conversion rates for repeat customer and referral program appointments are adversely affected. A sales visit conducted within a day of a referral lead is expected to be converted into a sale 60.7% of the time, but if that lead waits for 15 days, it is converted only 22.6% of the time, dropping below the rates for exhibition appointments after 11 days and direct mail appointments after 12 days. Similarly, repeat customer appointments close 90% of the time when capacity-driven lag is 1 day, but the closure rate falls to 44.8% when the lag is 15 days. Stage 3b: Order Size Parameter estimates of the order-size model (Equation 4) appear in Table 5; they indicate that, overall, model fit is

30 Number of Appointments (Leads = 30) 28 26 24 22 20 18

8 10 12 14 16 18 20 22 24 26 28
Follow-Up Delay (Days)

Direct mail Newspaper advertisinga Telephone directoriesa Radio advertisinga Referral programsa Exhibition
aSignificant

delay effect ( = .10).

log-likelihood of sales closure as a function of sales representative quality, the season, the source generating the lead, the time lag, and several household-level variables. The overall model fit is reasonable, given its disaggregate nature. The model chi-square (likelihood ratio test) is highly significant (p < .0001), and the corresponding Nagelkerke R2 (a likelihood-based pseudo-R2 measure) is .145. As we expected, the quality of the sales representative has a significant, positive effect on the likelihood of closure. Based on a median split of the previous years sales performance, high-performing sales representatives significantly affect sales conversion (captured by the parameter ), increasing the odds of a sale by 9% (e.087), all else being equal. We find significant seasonality effects in this stage of our model (captured by 1 and 2). The odds of closing a sale are reduced by 8.5% (e.089) in the early season and by 11.2% (e.119) in the late season. Two household-level descriptors (Zlh) also have a significant impact on sales closure: length of residency (positive impact) and age of the home (negative impact). There are significant differences between the leadgenerating sources in this stage. The lead-source dummy coefficients (0i) are significant, indicating deviation from the exhibition reference source (0N). Although exhibitions, newspaper advertising, radio advertising, and telephone directory appointments are converted into a sale at roughly

Integrated Marketing Communications at the MarketingSales Interface Table 4

COEFFICIENT ESTIMATES FOR LOGIT MODEL OF SALES CLOSURE (EQUATION A3) Variable Intercept Early season Late season High-performing sales representative Direct mail (DM) Newspaper advertising (NA) Radio advertising (RA) Telephone directories (TD) Referral programs (RP) Repeat customers (RC) Retail (RT) Internet communications (IC) Capacity-driven lag DM Capacity-driven lag EX Capacity-driven lag NA Capacity-driven lag RA Capacity-driven lag TD Capacity-driven lag RP Capacity-driven lag RC Capacity-driven lag RT Capacity-driven lag IC Customer-driven lag DM Customer-driven lag EX Customer-driven lag NA Customer-driven lag RA Customer-driven lag TD Customer-driven lag RP Customer-driven lag RC Customer-driven lag RT Customer-driven lag IC Length of residency Home age Home value Head of household age Household income Coefficient Estimate .4256 .0894 .1187 .0866 .2174 .1573 .3865 .1862 1.1485 2.6799 .3713 .2325 .0034 .0143 .0018 .0052 .0189 .1154 .1507 .0087 .0354 .0143 .0048 .0046 .0521 .0153 .0096 .0315 .0008 .0020 .0267 .0036 .0172 .0420 .0068 Wald Statistic 3.684 3.241 4.408 5.561 3.012 1.764 2.028 2.304 64.656 393.725 8.959 1.318 .041 3.357 .013 .012 1.466 29.269 68.417 .346 1.109 3.950 2.475 1.204 1.658 3.263 .561 5.637 .041 .203 6.248 9.118 .552 2.136 .466 p Value .055 .072 .036 .018 .083 .184 .154 .129 .000 .000 .003 .251 .420a .033a .455a .457a .113a .000a .000a .278a .146a .047 .116 .272 .198 .071 .454 .018 .839 .652 .012 .003 .458 .144 .495

573

Exp(B) .653 .915 .888 1.090 .805 .854 .679 .830 3.153 14.584 1.450 1.262 .997 .986 .998 .995 .981 .891 .860 1.009 .965 1.014 1.005 1.005 1.054 1.015 .990 .969 1.001 1.002 1.027 .996 .983 .959 .993

aWe report the p value for one-tailed tests given a clear prior hypothesis that the capacity-driven lag effect will be negative; otherwise, we report two-tailed tests. Notes: Log-likelihood test = 1656.24, d.f. = 34, p = .000; Nagelkerke R2 = .145; n = 14,770.

strong, with an adjusted R-square of .449. Potential order size, given the number of units expected to be purchased (ascertained by the call center at the time of initial inquiry), is a major predictor of the realized order size ( = .575, p < .001). High-performing representatives produce 5.5% (e.054) higher levels of revenue from a sale than lowperforming sales staff. The lead-generating source (i) continues to be relevant. Although many sources do not significantly differ from exhibitions (reference source), retail and referral leads tend to create significantly higher levels of revenue, whereas repeat customers generate lower levels of revenue per sale. Finally, sales to households that live in older homes tend to produce lower revenues (homeage = .225), and sales to high-income households tend to produce higher revenue levels (highincome = .207). Model Validation To examine the predictive accuracy of the model, we use data available for the first 21 weeks of 2004 as our holdout sample (we estimated the model on 20022003 data). In Stage 1, 1805 leads were predicted during this holdout period, and 1810 actual leads were generated; this represents a .3% error. Note that Stage 1 submodels that incorporate interaction effects across media perform substantially better than equivalent submodels that exclude media inter-

actions, particularly in the cases of newspaper and exhibition submodels. With regard to predictive accuracy within the holdout sample, 1475 leads are predicted by submodels that are estimated without media interactions included, increasing the error to 22.7%. In Stage 2, our model predicts 2803 appointments from all nine lead-generating sources, whereas actual appointments are 3053 (an error of 8.2%). For sales closure (Stage 3a), we predict a 37.7% closure rate, and we observe 39.9% (5.4% error). Finally, the actual average order size is $8,464, whereas the predicted value is $7,605 (lower by 10.2%). Although the model performs reasonably well in predicting these out-of-sample observations, it underpredicts both the sales-closure rate and the average order size. The 2004 actuals reflect the improved economic conditions over 2003, an aspect not reflected in the model parameters, as estimated on 2002 2003 data. Given the available data, our model does not include the impact of macroeconomic conditions. Such adjustments would need to be made exogenously. In addition to the holdout sample, we also explored potential bias associated with parameter constancy (often referred to as the Lucas critique or endogeneity bias) within model stages and substages that incorporate timesseries data. We generated cumulative sum plots of residuals to detect potential departures from constancy of regression

574

JOURNAL OF MARKETING RESEARCH, NOVEMBER 2006 Figure 4

EFFECT OF CAPACITY-DRIVEN DELAY ON SALES CLOSURE

Table 5
COEFFICIENT ESTIMATES FOR ORDER-SIZE MODEL (EQUATION A4) Coefficient Estimate

endogeneity bias is not a significant concern, and the parameter estimates appear to be acceptably stable. SIMULATION AND NORMATIVE IMPLICATIONS Following discussions with HIR management over a twoyear period, we developed a decision support tool, which was implemented in fall 2004. This user-friendly interface allows managers to assign media expenditures on a weekly basis for the calendar year. It also allows inputs for salesperson allocations, household-level attributes, and gross margin. From a practical standpoint, the tool provides a series of outputs to what-if types of questions, addressing the impact of changes in communications budgets and/or allocations. As an example, Figure 5 shows the salesclosure stage of the tool (Stage 3a). The weekly closure estimates incorporate the complex multimedia interactions and carryover (Stage 1), effects of delay and decay on appointment generation and closure (Stages 2 and 3a), and effects across the stages, such as seasonality. Although this is a single screen shot, other diagnostic capabilities in the tool permit analysis at each of the three stages in the sales process and at varying levels of aggregation. For example, data in Figure 5 are aggregated to provide yearly projections of media expenditures at 5% increments, altered proportionately on the basis of initial inputs. Using the decision support tool, we provide three hypothetical scenarios in Table 6 to illustrate the impact of changes in media allocations on HIRs operations. The results and time-lag diagnostic statistics for weekly budgeted 2004 media expenditures appear in Column 2. Note that the budget allocations were created in fall 2003, before

Variable

SE .238 .035 .033 .080 .036 .041 .081 .034 .062 .012 .020 .078 .024 .108 .098 .082

p Valuea .000 .467 .213 .477 .542 .004 .019 .019 .950 .000 .007 .500 .000 .948 .995 .012

Intercept 7.8573 Direct mail .0257 Newspaper advertising .0409 Radio advertising .0568 Telephone directories .0222 Referrals .1187 Repeat customers .1890 Retail showrooms .0788 Web page .0039 Ln(potential units) .5751 High-performing sales representative .0538 Ln(length of residency) .0523 Ln(home age) .2253 Ln(home value) .0070 Ln(head of household age) .0006 Ln(household income) .2073

aWe report all p values for two-tailed tests of significance. Notes: Adjusted R2 = .449; n = 2980.

relationships over time for model stages that apply timeseries data (Brown, Durbin, and Evans 1975). In each model/submodel we tested, no more than 3.8% of residuals (4 of 104 in the newspaper submodel) exceeded the 1% significance boundary, and these do not appear to signal a behavioral change of economic agents (managers or consumers) or a pattern of instability. Thus, we conclude that

Integrated Marketing Communications at the MarketingSales Interface Figure 5

ILLUSTRATIVE DECISION SUPPORT TOOL OUTPUT: STAGE 3 (SALES-CLOSURE RESULTS)

575

the development of the tool. On the basis of these allocations (for a budget of approximately $2.1 million), the model estimates the number of leads, appointments, and sales orders, as well as estimated sales dollars and total profit (just over $4.1 million, assuming a gross margin of 30%). In addition, the model predicts the average weekly median lag (over 52 weeks of the 2004 budget year) to be 8.3 days, with a minimum and maximum over the year of 2.0 and 18.0 days, respectively (SD = 3.70). Given the complexity of media planning and the external constraints involved (e.g., media availability, price variability, timing of exhibitions), we explore deviations from established media spending patterns to better understand the implications of changes in spending across sources. We develop three stand-alone scenarios by first exploring leadgenerating efficiency of media expenditures in Scenarios 1 and 2 and then examining sales conversion efficiency in Scenario 3. Rather than illustrate the best decision that maximizes firm profits, the scenarios represent alternative strategies that improve profitability by focusing on budgeting and allocation decisions that accentuate the relationships between the stages of our model. In theory, it is possible to employ an optimization approach to develop a zero-based profit-maximizing communications budget. However, radical deviations from current allocations that may result from optimization might imply levels of decision variables well beyond the range of values over which the models were estimated, thus casting

aspersions on the validity of the optimization. Furthermore, radically different allocations can create implementation hurdles as middle-level managers develop misgivings about the value of the exercise. Therefore, we did not suggest an optimization routine to HIR. Instead, we employ a rough heuristic to develop the scenarios to assist managers in improving media budgeting decisions, based on systematic deviations from the status quo baseline (e.q., an incremental rather than a zero-based approach to seeking revenue and profit improvements). First, rather than a cost-savings focus, we recommend media budgets/allocations that predict total sales dollar volume equal to or greater than benchmarked levels to drive profit improvements through media/sales efficiencies, with the objective of improving revenue along with profit levels. Time-sensitive media (e.g., large exhibitions with externally predetermined dates, season-specific media buys) cannot be altered that much. Second, by recognizing positive interactions among media, we can improve budgeting and allocation decisions by deploying multiple media together and reducing spending in weeks with little media activity. For example, this governed our redeployment of radio, which has a relatively modest main effect but reinforces the impact of exhibitions and newspaper advertising (through an interactive effect). Third, as these bursts of multiple-media campaigns are developed, their timing is largely driven by downstream sales force capacity and when the spending occurs on the seasonality ramp. Combined media spend-

576

JOURNAL OF MARKETING RESEARCH, NOVEMBER 2006 Table 6

BUDGET TIMING AND ALLOCATION SIMULATIONS Scenario 1: Media Timing and Lead Focus 821,684 958,991 206,040 141,957 2,128,672 1722 1795 82 2724 4320 10,643 1313 1460 71 1818 3595 8258 374 441 18 541 1448 2823 21,935,446 4,451,962 8.90 3.60 2.017.9 Scenario 2: Media Allocation and Lead Focus 779,354 914,321 287,040 147,957 2,128,672 1683 1945 84 2963 4320 10,994 1285 1583 73 1968 3589 8499 366 478 18 583 1435 2879 22,389,750 4,588,253 9.20 3.73 2.019.7 Scenario 3: Media Reduction and Sales Focus 492,029 1,013,931 269,079 136,732 1,911,771 1550 1758 87 1959 4320 9674 1198 1448 76 1343 3612 7678 342 437 19 407 1494 2700 20,951,843 4,373,782 8.20 2.73 2.012.7

Simulation Stages Communication Budget Direct mail ($) Newspaper advertising ($) Radio advertising ($) Exhibitions ($) Total ($) Leads (Stage 1) Direct mail Newspaper advertising Radio advertising Exhibitions Other source Total Appointments (Stage 2) Direct mail Newspaper advertising Radio advertising Exhibitions Other source Total Orders (Stage 3a) Direct mail Newspaper advertising Radio advertising Exhibitions Other source Total Total sales (Stage 3b) ($) Total profit ($) Delay Statistics M SD Range

2004 HIR Budget Allocation 821,684 958,991 206,040 141,957 2,128,672 1732 1741 83 1877 4320 9752 1320 1427 72 1284 3608 7710 377 433 18 384 1472 2683 20,832,556 4,121,095 8.30 3.70 2.018.0

ing deep in the off-season is less effective in generating leads, but in general, the firm has excess sales force capacity available to service these leads quickly. Scenario 1: Media Timing In Scenario 1, no changes to the overall budget or the allocation across media sources are considered. We simply alter the timing of media spending, shifting spending levels of radio, newspaper, and direct mail efforts from one week to another. We focus on these media because they tend to be most flexible with regard to timing. Exhibitions are largely predetermined events; thus, any changes in timing would be unrealistic. Table 6, Column 3, shows the impact of these timing shifts in each stage of our model. In Stage 1, the new spending pattern results in 891 additional leads, 548 additional appointments, 140 additional sales orders, and $330,867 in incremental profit (an 8.0% increase). Performance improvements are attributed to leveraging interaction effects in Stage 1 of the model. In many cases, HIR alternates the use of media, often spending heavily in one medium in a given week and shifting the emphasis to a different medium the next week. By moving expenditures into common weeks, especially those with heavy exhibition and newspaper spending, we allow the interaction effects of Model 1 to be leveraged better. Specifically, we tried to sup-

port exhibitions by moving more radio spending into the same weeks. Similarly, we tried to leverage radio and direct mail efforts by moving them into weeks with heavy newspaper advertising. Although these shifts have little impact on the number of leads generated directly by either radio or direct mail (the number of direct mail leads decrease slightly), the impact on exhibition and newspaper leads is substantial (890 additional leads from these sources). In other words, by merely shifting the timing of media expenditures, we observe considerable scope for leveraging the potential interactions. Scenario 2: Media Allocation In Scenario 2, we continue to focus on improving lead generation, but here, we examine the existing media allocation. Similar to the first scenario, Column 4 represents a potential allocation to illustrate the stagewise impact associated with changes in media timing and allocation. Again, focusing on the interactive effects of radio advertising on exhibitions and newspaper advertising, we increase the radio budget by $81,000 (39.3%) and spend it in weeks that better leverage exhibitions and newspaper advertisements. The budget for exhibitions has a modest increase (4.25%), whereas direct mail and newspaper budgets were marginally reduced (by 5.2% and 4.7%, respectively) to balance

Integrated Marketing Communications at the MarketingSales Interface the overall budget at current levels. Based on these changes in allocation and timing, the expected annual total leads increase by 12.7%. Although leads from exhibitions increase significantly as a result of more spending and radio support, we note that newspaper leads also increase, even though its allocation was reduced. Expected newspaper leads increase by 11.7% with a budget reduced by nearly $45,000. Following these efficiency improvements through Stages 2 and 3 (appointment generation and sales conversion), overall sales increase by approximately $1.6 million, creating approximately $470,000 of incremental profit (11.3%). However, the improvements on HIRs original allocation (Scenarios 1 and 2) come at a price, namely, significantly increased stresses placed on the sales force and potentially detrimental service levels to prospects and customers. The average time lag between customer inquiry and sales appointment increases from 8.3 days to 8.9 days in Scenario 1 and to 9.2 days in Scenario 2. The expected median lag in some weeks reaches 19.7 days in Scenario 2. The increased delay may have a negative impact on customer sentiment and selling efficiency. The reallocation focuses heavily on creating more exhibition leads, which are converted into sales at a lower rate than other sources. In addition, these leads are created when delay is greatest in the system, further reducing their conversion potential. Thus, not only are the incremental leads less likely to be converted into sales, but the increased lag time also hampers all other leads in the system at that time. The overall conversion rate (orders/ leads) drops by approximately 5% under this scenario. Therefore, although marketing may seem to gain by the reallocated budget (greater leads), the salespeople face serious disadvantages, which is likely to create friction within the organization. Thus, we explore a third scenario to address the integration of lead generation and sales conversion. Scenario 3: Integrating Media and Selling Communications In Scenario 3 (Table 6, Column 5), the overall media budget is reduced by approximately 10%, reallocated across media and redistributed across weeks to improve synergistic effects. The reallocation involves direct mail expenditures being reduced relative to the baseline (HIRs 2004 budget) by 40.1%, radio being increased by 30.6%, and minor changes being made to newspaper (increased by 5.7%) and exhibition (reduced by 3.7%) efforts. These spending patterns generate 9674 leads (expected), 78 fewer than the original 2004 budget; however, slight improvements in the number of closures (17) and the overall sales dollar volume ($119,287) are also expected. Expected profitability improves by $252,687 (6.1%) in this scenario. The improved conversion efficiency is due to two reasons: a slight reduction in time lag (from 8.3 to 8.2 days) and a significant drop in the variation in delay across weeks; as a result, the standard deviation drops by 26%, and the upper bound of the range is reduced from 18.0 days to 12.4 days. These scenarios highlight the relationships among the number of leads, the resulting time lag, and the entire system performance. They also suggest multiple strategies to improve overall performance. The classical perspective stops at lead generation, under the assumption that more

577

leads offer increased opportunities for downstream sales. Scenarios 1 and 2 illustrate how an improved understanding of media synergies can help achieve this objective, but integration is not restricted to advertising or direct communications alone. From an IMC perspective, integration involves an understanding of how all communications influence one another both simultaneously (among media in a given week) and sequentially (between media and selling activity). Scenario 3 illustrates how a better understanding of the impact of media spending (and lead generation) on sales conversion can improve system performance. In this case, fewer leads seem worthwhile because they permit increased service quality (in the form of response time) and selling efficiencies (higher closure rates) downstream. CONCLUSIONS AND FURTHER RESEARCH A report by the Aberdeen Group (Watkins 2002) confirms a crippling disconnect between the marketing and the sales functions within many organizations, resulting in large amounts of wasted expenditures and energy for the firm. In addition, inconsistent customer messaging, poor or delayed sales preparation, and less effective selling dialogues can result. The IMC framework is a useful way to bridge this gap. Despite its intuitive appeal, the concept continues to evolve as researchers and practitioners explore its central principles, namely, that communications through any medium represent an ongoing dialogue with customers and that one mode of communication can contribute to the performance of other elements in the mix. These observations imply that engaging and managing the customers experience requires targeted contacts with effective support from the right elements of the mix (Schmonsees 2005). Marketing and sales expenditures can then be viewed as delivering tangible returns and contributing effectively to the bottom line. Such accountability seems to be the order of the day (Marketing Science Institute research priorities 2004). We modeled the impact of communications in three distinct stages: lead generation, appointment conversion, and sales closure. We offered empirical evidence of interactive and carryover effects across media sources in the lead generation stage. We showed that communications spending can directly contribute to delays in sales force follow-up, thus linking media spending more directly to the sales process. We observed that delays create detrimental effects within the sales process, reducing the likelihood of converting a lead into an appointment and the closure of the appointment into a sale. Based on individual- (household-) level data from a major home improvement retailer, our estimation results indicate that individual media (e.g., print, radio, exhibitions) have different impacts in the various stages of the sales process and are affected differently by the time lag (delay). Through scenarios and using a decision support tool, we showed that the effectiveness of the entire system can improve through the use of two distinct but interrelated mechanisms: interactive effects among leadgenerating media and complementary effects between these media as well as follow-up selling activities influenced by capacity-driven delay. These findings have useful implications for media planning and budgeting. However, our study has several limitations that also serve as useful avenues for further research. First, more detail on the prospects media exposure can be

578

JOURNAL OF MARKETING RESEARCH, NOVEMBER 2006

( A1) ln L it L i ( t 1) = ln 0 i + ln ( 1i ) St + ln ( E ) E t + ii ( ln X it ) +

collected when the inquiry comes into the call center. In our case, prospects indicated the single source that was most responsible for influencing them to inquire about the firms product. Multiple communication sources may be in operation at a given time; thus, further research could examine which other sources the prospect was exposed to and when he or she was exposed to them. Second, our analysis assumes that time lags above the median value were likely driven by the customer; however, it would be desirable to collect more refined data on the extent to which the agreedon appointment date was customer versus firm driven. Third, our efforts centered most directly on media allocation and its impact on the sales system; further development of the theory and measures associated with the sales force within an IMC context is of significant importance to future research. Fourth, our sponsor is a major player in the industry and the market we studied; however, incorporating the effect of competition would be a useful dimension. Furthermore, a replication of these results in other markets would be valuable. Our multistage conceptualization specifies stage-specific models that are estimated separately. An extension may consider a formulation that permits joint estimation of the entire model across the stages. A promising approach in this regard is based on the hidden Markov model framework, for which Smith, Naik, and Tsai (2006) recently developed the model estimation and selection methodology (for its application in the context of a multistage IMC model, see Raman and Naik 2006). Although we noted the practical problems with optimization to recommend profit-maximizing media budgets, the approach may be useful if there are (user-specified) constraints that ensure realistic allocations that avoid expenditures too far beyond the normal range of values. This is an issue that we will consider in the next phase of work with our sponsor. APPENDIX: MODEL SPECIFICATION FOR ESTIMATION For estimation, the models represented in Equations 1, 2, and 4 are linearized by taking logs on both sides. Furthermore, the models represented in Equations 2, 3, and 4 have some parameters that are source specific and others that are common across sources. We can restate the lead generation model (Equations 1a and 1b) as follows:
( A1) ln L it L i ( t 1) = ln 0 i + ln ( 1i ) S + ii ( ln X it ) +

ln (1 + X ) + e
ij jt j = 1, j i

1it .

The sales-appointment, sales-closure, and order-size models represented by Equations 2, 3, and 4, respectively, are estimated across sources. For this purpose, we introduce a dummy variable, Di, that is equal to 1 if the media source is i (i = 1, , N), and 0 if otherwise. Furthermore, the salesclosure and order-size models represented by Equations 3 and 4, both of which we specify at the household level, are estimated across households. We can restate these three models as follows:
N 1

( A 2)

ln A it = ln 0 N +

D ln
i i =1

0i

+ ln ( 1 ) S1 + ln ( 2 ) S2

D {
i i =1

1i

( ln L it ) + 2i ln ( Lagt )} + e2it,

N 1 Ph | it ( A3) ln D i 0 i + 1S1 + 2S2 + Y = 0 N + 1 Ph | it i =1

N 1

D {
i i =1

1i min

( Laghit , Lagt )

+ 2 i max Laghit , Lag t Lag t +

Z
l l =1 N 1 i =1 L

lh

+ e3h , and

( A 4)

ln Sh | i = ln N +

D (ln ) + (ln V ) + (ln ) Y

i i h l lh

(ln Z
l =1

) + e4h ,

ln (1 + X ) + e
ij jt j = 1, j i

1it .

respectively. Finally, the linking model (Equation 5) is already in estimation-ready form. Note that our system of equations representing the sequential process involves a mix of aggregate-level and individual models, including one with a logit specification. Thus, a simultaneous estimation approach would be extremely complex if at all feasible. Therefore, we estimate each of the five models separately. REFERENCES
Acheson, Kerri L. (1993), Integrated Marketing Must Bring 2 Perspectives Together, Marketing News, (August 16), 45. Belch, George E. and Michael A. Belch (2003), Advertising and Promotion: An Integrated Marketing Communications Perspective, 6th ed. Boston: McGraw-Hill. Brown, R.L., J. Durbin, and J.M. Evans (1975), Techniques for Testing the Constancy of Regression Relationships over Time, Journal of the Royal Statistical Society, 37 (2), 14992. Couretas, John (1984), Study to Measure Ads Effect on Sales Force, Business Marketing, 69 (5), 34.

The model may be estimated separately for each source. However, given the possibility of correlated error terms across sources, we recommend estimation with SUR.6 We can restate the model for exhibitions (Equations 1a and 1b) as follows:

6Because ln(L it i, t 1) is not defined when Lit iLi, t 1 < 0, estimation algorithms will understate the value of to satisfy the positivity requirement and, thus, bias l downward.

Integrated Marketing Communications at the MarketingSales Interface

Duncan, Tom (2002), IMC: Using Advertising & Promotion to Build Brands. New York: McGraw-Hill. and Clark Caywood (1996), Concept, Process, and Evolution of IMC, in Integrated Communication: Synergy of Persuasive Voices, E. Thorson and J. Moore, eds. Mahwah, NJ: Lawrence Erlbaum Associates, 1334. Gatignon, Hubert and Dominique M. Hanssens (1987), Modeling Marketing Interactions with Application to Sales Force Effectiveness, Journal of Marketing Research, 24 (August), 24757. Gopalakrishna, Srinath and Rabikar Chatterjee (1992), A Communications Response Model for a Mature Industrial Product: Application and Implications, Journal of Marketing Research, 29 (May), 189200. and Gary L. Lilien (1995), A Three-Stage Model of Industrial Trade Show Performance, Marketing Science, 14 (Winter), 2243. Hanssens, Dominique M. and Barton A. Weitz (1980) The Effectiveness of Industrial Print Advertisements Across Product Categories, Journal of Marketing Research, 17 (August), 294306. Levitt, Theodore (1967), Communications and Industrial Selling, Journal of Marketing, 31 (April), 1521. Lilien Gary L., Philip Kotler, and K. Sridhar Moorthy (1992), Marketing Models. Englewood Cliffs, NJ: Prentice Hall. Morrill, John E. (1970), Industrial Advertising Pays Off, Harvard Business Review, 48 (MarchApril), 414. Naik, Prasad A. and Kalyan Raman (2003), Understanding the Impact of Synergy in Multimedia Communications, Journal of Marketing Research, 40 (November), 37588.

579

, , and Russell S. Winer (2005), Planning Marketing-Mix Strategies in the Presence of Interaction Effects, Marketing Science, 24 (1), 2534. Raman, Kalyan and Prasad A. Naik (2006), Integrated Marketing Communications in Retailing, in Retailing in the 21st Century: Current and Future Trends, Manfred Krafft and Murali K. Mantrala, eds. New York: Springer, 38196. Reid, Mike (2003), IMC-Performance Relationship: Further Insight and Evidence from the Australian Marketplace, International Journal of Advertising, 22 (2), 22748. Schmonsees, Robert J. (2005), Escaping the Black Hole: Minimizing the Damage from the Marketing-Sales Disconnect. Mason, OH: Thomson. Schultz, Don E. (2004), Include SIMM in Modern Media Ad Plans, Marketing News, 38 (9), 6. Smith, Aaron, Prasad A. Naik, and Chih-Ling Tsai (2006), Markov-Switching Model Selection Using KullbackLeibler Divergence, Journal of Econometrics, forthcoming (available at http://www.sciencedirect.com). Smith, Timothy M., Srinath Gopalakrishna, and Paul M. Smith (2004), The Complementary Effect of Trade Shows on Personal Selling, International Journal of Research in Marketing, 21 (1), 6176. Watkins, Harry (2003), Bridging the Great Divide: Process, Technology, and the Marketing/Sales Interface. Wellesley, MA: Aberdeen Group.