Equations need to be viewed in Intenet Explorer. Symbol fonts are not available in Firefox.
The main thrust of this book has been the analysis of market-share figures with an explicit objective of improving the marketing manager's understanding of the market and competition. We have presented in Chapters 1 through 3 a framework and models which we consider the best to achieve this end. The data collection and estimation techniques for calibrating these models were discussed in Chapters 4 and 5. A competitive-mapping technique which is useful in interpreting the calibration results was presented in Chapter 6. Yet those chapters have not entirely achieved our objective because the manager is obviously not content with merely analyzing and describing the competitive interactions in the marketplace. Whatever understanding the manager gains through market-share analysis will have to be converted eventually into concrete marketing programs. We turn next to this last stage of market-share analysis.
It has been our position in this book that the competitive conditions in a market may be chiefly described by a set of model parameters, especially the elasticities of market shares with respect to marketing variables. However, we recognize that designing a marketing program based on the knowledge of elasticities is not an automatic process. With only a limited number of brands and marketing variables one may have to deal with a surprisingly complex pattern of competitive interrelationships, indicated by the large number of elasticities and cross elasticities. The manager needs to interpret such a pattern and to select one set of levels for the marketing variables which presumably maximizes the firm's (long-term) profits. But there are no easy rules for converting a given pattern of competition into an implementable marketing program.
Some might think of designing marketing programs as large-scale mathematical-programming problems, but such a conception is unrealistic for several reasons. For one thing, the future environment for a firm's marketing program is full of uncertainties which affect its performance. Economic conditions change unexpectedly; variations in consumer tastes are sometimes illogical; weather and climate substantially impact demand, etc. Statistical decision theory may be employed to cope with future uncertainties, but its application to market-share analysis is complicated by the fact that those factors that cause uncertainties must be explicitly brought into the model and the strength of their influence must be calibrated beforehand. Many important factors cannot be treated in this manner. How, for example, does one calibrate the impact of one-time events, such as new governmental regulations, on one's market share?
For another thing, even if other uncertain factors can be correctly guessed, the best (optimal) marketing program for a firm is still dependent on the willful and often unpredictable actions of the competitors. One may guess competitors' marketing actions and plan one's own program accordingly. But will they guess that our actions are about to be modified and adjust their actions again? As we discuss later, game theory, which is one possible solution technique to decision making under this type of circular reasoning, is not advanced enough to give the marketing manager practical solutions in competitive situations involving many brands and a large number of marketing variables.
Lastly, designing an optimal marketing mix (i.e., the best combination of marketing variable) presumes the existence of a definite objective. In formal theories the objective is assumed to be maximizing either long-run or short-run profits, but in many practical decision situations profit maximization is not always pursued. The real-world managers may have difficulties in conceptualizing long-run profits, yet they are too astute to try to maximize short-run profits. In the context of market-share analysis, it is often a planned level of market share that becomes the main objective and is pursued vigorously. To complicate the matter further, the brand manager at a manufacturing firm may have an entirely different objective than that of a store manager for a supermarket chain. Both the brand manager and the store manager may feel that the objective of the other is wrong. There is no sophisticated theory of mathematical programming which enables one to find an optimal marketing mix when one is not sure of the objective to achieve.
We believe that the present state of the art in decision theory and game theory is such that the marketing manager will find little use for these theories in their planning work. Lacking an easily applicable theory, the manager may resort to the simplest approach, in which he/she forecasts the most likely pattern (or scenario ) of the future environment and the competitors' actions, and designs a single marketing program to meet this pattern. This approach, however, is not a very logical one, especially when the likelihood that the chosen scenario occurs is small. This may sound contradictory, but when the events which constitute the chosen scenario are independent and numerous, the joint probability that all of the events occur simultaneously may indeed be very small - even for the most likely scenario - because it is the product of probabilities for the individual events. It is far more practical for the manager to create several likely scenarios of the future environment and competitors' actions, initially choose one program that corresponds to the most likely scenario, and keep the others as contingency plans. If the initially-chosen plan turns out to be the wrong one in view of the subsequent developments, one may easily move to another plan which fits the new situation best.
This contingency-planning approach appears to us to be the most practical solution to marketing planning under environmental and competitive uncertainties. But if the marketing manager wishes to take this planning approach he/she will need a planning tool which permits him/her to design many marketing programs, each of which corresponds to a likely scenario. Even with only a few brands and several marketing variables, the computation of the optimal marketing mix could be a formidable task. The manager must first collect the data on past market shares and marketing variables, calibrate the model, and forecast the effects of environmental and competitive factors. He/she must then compute, for each likely scenario, the value of profits (or some other objective function) for each combination of marketing variables, and select that combination which maximizes the objective. Unless we supply him/her with an efficient computational tool to perform this task, the manager is likely to revert back to a more naive approach.
In the following sections we present an example market information system, the main purpose of which it to facilitate market-share analysis. Although our example system, which is called CASPER (Competitive Analysis System for Promotional Effectiveness Research), may seem small compared to real-world market information systems (which tend to be immense), it attempts to integrate the data collection, model estimation, interpretation, and the marketing-planning process. Real-world systems can be designed as natural extensions of our proposed system.
The most practical component of a market information system is the decision-support system. The functions of a decision-support system must be broad enough to enable managers to:
CASPER has been developed to illustrate the kinds of functions encompassed by this mandate. CASPER contains a HISTORY file with a year's worth of weekly data, from three grocery chains, summarizing sales in the ground, caffeinated coffee market. These historical data were used to calibrate the market-share and category-volume model.The market-share model used data from all seven grocery chains to estimate parameters.As indicated in the development of the category-volume model (see section 5.12.2), the private-label brands (PL 1-3 and AOPL) were combined into a single APL brand aggregate.
CASPER's Standard Graphic Library summarizes each of the nine brands in this category. Each brand is traced over time, and each week is traced over brands for a comprehensive, visual record of the marketplace. Looking at the data is an indispensable step in building the kind of expertise needed for good decisions. CASPER also contains a menu-driven market simulator. One inquires into the market by specifying the competitive conditions for a particular occasion or series of occasions, with the results being accumulated in an OCCASIONS file. CASPER contains graphing and tabling functions to help summarize simulation results. All basic results are in the form of spreadsheets so that one can use the capabilities of FRAMEWORK to create a wide variety of summary reports.FRAMEWORK is a product of Ashton-Tate, 20101 Hamilton Avenue, Torrance, CA 90502-1319. While FRAMEWORK is necessary to run CASPER, any of the resulting spreadsheets can be imported to other spreadsheet programs for further analysis. The dynamic simulator in CASPER is structured as a game. The GAME provides a way of putting marketing plans to the test. In the GAME module three Brand Teams (for Folgers, Regular Maxwell House, and Chock Full O'Nuts) develop promotion plans and support material to try to convince three Retailer Teams to promote their brands during an eight- or nine-week promotion period. All six teams can compete for profits or the roles of the Retailer Teams can be played by a Game Master. In either case the teams receive results back in the form of three spreadsheets. First is a summary of sales and estimated profits achieved by the real brands and retailers during this period under the profit assumptions of the GAME. Second, each Brand Team receives a summary of how they would have fared against the real brands' actions. And finally all teams receive a summary of the sales and profits in head-to-head competition. By triangulation, each team gets a comprehensive summary of its performance. The complete GAME involves three promotion periods of nine, nine, and eight weeks, respectively. These 26 weeks of data were the ones following the 52 weeks used to calibrate the market-share and category-volume models.
The simulator in CASPER is driven by a high-parameter asymmetric market-share model,The parameter values are listed in Table 5.13. hich incorporates the basic premise that marketing actions must be distinctive to be effective, and by a 31 parameter, category-volume model.The parameters for the category-volume model are given in Table 5.14.This version of CASPER is a brand-management tool for the coffee market, as well as a prototype for the kind of functionality managers should expect in any market covered by optical-scanner data.
CASPER's Standard Graphic Library contains plots which provide an initial summary of the history of all brands. The summary of a brands market share or sales over time is found in Plot Settings.View.Standard Libraries.XY-Plots. If one uses this menu to bring Folgers Sales - Chain 1 to the screen, the result should look like Figure 7.1. The top line reflects the price each week with values recorded on the right-hand vertical axis (from $2.64). The bottom line reflects market shares, with values demarcated on the left-hand vertical axis. Weeks with a feature (F), display (D), and/or store coupon (C) are noted just above the ``Week'' axis at the bottom. Note the large spikes in share or sales (up to 5,238 lbs. in week 49) which occur in weeks with price cuts combined with feature
Figure 7.1: Folgers Market Share - Chain 1
(F), displays (D), and/or store coupons (C). The dramatic pulsing of sales during promotions is something which would be lost in yearly aggregated sales graphs. By looking at a chain-week record one obtains a much clearer sense of market response and the promotion policy of a grocery chain.View the market share or sales figures for Chains 2 and 3. Do grocery chains seem to differ in how frequently they promote Folgers or how they use the marketing instruments? Chain 2 seems more willing to sustain a promotion over several weeks. Chain 3 seems to promote less frequently, using few, if any, store coupons or displays. o see the array of competition a brand faces in any given week use CASPER's Standard Pie Charts. Use Plot Settings.View.Standard Libraries.Pie Charts to look at Chain 2 weeks 10, 13, and 30. These should look like Figures 7.2 - 7.4. Note that at $1.87 per pound (with a newspaper feature and an in-store display) Chock Full O'Nuts acquires 72% of the market in week 10. In week 13 (Figure 7.3) the same offering for Chock Full O'Nuts results in a 44% share. Is there anything in the competitive environment which explains the difference in market response to Chock Full O'Nuts? Obviously the promotion for Maxwell House (low price combined with a newspaper feature, in-store display, and store coupon) has an impact. As was emphasized in prior discussions of the distinctiveness of marketing activities, whatever the value of a particular feature, it is shared by all the brands possessing the feature. In week 30 the shared feature is only the distinctively low prices for Maxwell House and Chock Full O'Nuts, since there is no overlap in the other promotional instruments.
Before beginning to exercise the market simulator, it is natural to want to know how good the simulator is at reproducing history.Up to 52 weeks worth of historical data can be run through the market simulator at a time using the Run.Run Off History menu in CASPER. This was done a grocery chain at a time. The spreadsheets summarizing these simulations are called CH1SIM, CH2SIM, and CH3SIM, respectively. How to build CASPER's Standard Graphs from these spreadsheets is described below. For now we need only note that any CASPER graphics made by the user can be viewed in CASPER by selecting Plot Settings.View.User Libraries and specifying the complete file name.For example, we can view Maxwell House Market Share - Chain 2 from the Standard Graphic Library. It should look like Figure 7.5. Then select file Ch2Maxms from directory \CASPER\CASOCCS\CH2SIM for comparison. It should look like Figure 7.6. The ability of these models to reflect the consequences of competitive actions on a brand's market share make these forecasts valuable as planning aids. Even if univariate time-series were more accurate (which does not appear to be the case),
Figure 7.2: Market Shares - Week 10 Chain 2
Figure 7.3: Market Shares - Week 13 Chain 2
Figure 7.4: Market Shares - Week 30 Chain 2
Figure 7.5: Maxwell House Actual Market Shares - Chain 2
Figure 7.6: Maxwell House Estimated Market Shares - Chain 2
such models do not answer the what-if questions which form the core of any planning exercise.
This section presents a planning exercise assessing market response to a sale for Chock Full O'Nuts. This will be a static simulation in the sense that we first want to reflect market response as if nothing preceded each occasion and nothing followed it. The occasions will be demarcated by price varying from a little lower than is reasonable ($1.60) to a little higher than is reasonable ($2.80). The sale will be supported by a newspaper feature.The own-display and own-coupon parameters for Chock Full O'Nuts are not significant, and are therefore not used in this simulation. he market response will be assessed first against the standard shelf prices for all other brands, then against a sale for Folgers, then Maxwell House, and finally against a simultaneous sale for Folgers and Maxwell House. We will set the background conditions, set the ranges of price for the simulations, and plot the results.
The defaults are the background conditions which don't change in the course of a simulation (a block of occasions). If we are going to vary Chock Full O'Nuts' price it will be set in the next section. Here we are concerned only with setting Feature, Display, and Coupon - the things which will be fixed while prices vary.
First let's look at the existing defaults. We view them in the same pop-up spreadsheet which can alter the defaults. Select the menu options Set Defaults.Set Values.Marketing Instruments.Set.Set Manually/View. A display like Table 7.1 should appear.
Table 7.1: Default Price and Promotion Table
| Brand | Price | Feature | Display | Coupon |
| Folgers | $2.59 | .00 | .00 | .00 |
| MaxHouse | $2.31 | .00 | .00 | .00 |
| MasterBlend | $2.88 | .00 | .00 | .00 |
| Hills Brs. | $2.39 | .00 | .00 | .00 |
| Chock Full | $2.29 | .00 | .00 | .00 |
| Yuban | $3.29 | .00 | .00 | .00 |
| C&S | $2.39 | .00 | .00 | .00 |
| AOB | $2.45 | .00 | .00 | .00 |
| APL | $2.18 | .00 | .00 | .00 |
When CASPER is started, these default background assumptions are made. They simply reflect the average shelf prices for all brands in the HISTORY file. For easy modification, input to this sheet is keystroke filtered so that only sensible values can be entered. Prices must be positive numbers. Promotions are reflected as the proportion (between 0 and 1) of all category volume sold on that kind of a promotion. By using the extremes of 0 and 1, we can simulate what occurs in a single store. By using proportions we can reflect the results for less than a full week of promotion or results aggregated over stores with somewhat differing promotional environments.All changes to the defaults are highlighted in bold-faced type. Any unacceptable characters produce a beep and are not entered. Once any modifications are made, you exit this spreadsheet by pressing the < ESCAPE > key.Entering ``1'' for CFON Feature, < RETURN > , and then < ESCAPE > will set the background conditions for the first block of simulations.The original default values can be restored under Set Defaults.Set Values.Marketing Instruments.Reset CASPER Default Values.All. he background cost and profit-margin assumptions can be reviewed and modified in another pop-up spreadsheet obtained by choosing Set Defaults.Set Values.Profit.Set Costs Manually/View. The default values in this spreadsheet are summarized in Table 7.2.
Table 7.2: Default Costs
| Brand Cost per lb. | Cost per Store | ||||
| Brand | Retailers' | Manufctr's | Feature | Display | Coupon |
| Folgers | $1.98 | $1.39 | $75.00 | $35.00 | $20.00 |
| MaxHouse | $1.89 | $1.32 | $75.00 | $35.00 | $20.00 |
| MasterBlend | $2.07 | $1.45 | $75.00 | $35.00 | $20.00 |
| Hills Brs. | $1.68 | $1.18 | $75.00 | $35.00 | $20.00 |
| Chock Full | $1.70 | $1.19 | $75.00 | $35.00 | $20.00 |
| Yuban | $2.69 | $1.88 | $75.00 | $35.00 | $20.00 |
| C&S | $1.81 | $1.27 | $75.00 | $35.00 | $20.00 |
| AOB | $1.98 | $1.39 | $75.00 | $35.00 | $20.00 |
| APL | $1.53 | $1.07 | $50.00 | $25.00 | $15.00 |
These are only our crude estimates and we are not particularly well informed on the specifics of costs in this market. Firms actually involved in this industry would have access to or interest in developing more accurate estimates of costs for all competitors. There are two points to be emphasized. First, in any simulation these cost assumptions should be made explicitly, not implicitly. Second, a decision-support system should allow one to change these assumptions as better information becomes available. Even in the absence of better data on costs, being able to vary cost assumptions allows one to see how sensitive the final profit results are to the initial cost assumptions.
Note that the cost structure is divided to reflect different roles of agents in the channels of distribution. The grocery stores are characterized as paying a cost per week for newspaper features, in-store displays, or store coupons. The Store Profits associated with each brand's simulated results are therefore a brand's estimated gross revenues (lbs. sold × retail price per pound) minus the wholesale costs (lbs. sold × retailers cost per pound) minus the fixed cost associated with features, displays and store coupons. The Brand Profits associated with each simulation come from the difference between the retailers' and manufacturers' cost per pound × the number of pounds sold. These are obviously elementary computations. The point is that a decision-support system should be able to reflect the profit implications of a brand's plans for the firms and for the channels of distribution. Planning which does not investigate channel profits is unreasonably myopic.
The means a manufacturer has of encouraging stores to promote a brand are in the form of per-pound discounts for features, displays, or coupons. The default assumptions in the CASPER simulator is that a $.05 discount per pound is offered for each promotional element, for each brand. Many different forms of incentives could be employed. But they can be represented on a per-pound basis without loss of generality. While stores are free to promote a brand without these incentives, the assumption here is that a brand offers a per-pound price without support (this is the price reflected in Table 7.2 as Retailers' Cost per lb.), and that the offered discounts per pound shown in Table 7.3 are only received by the retailer if they perform on that particular kind of promotion. A firm which didn't wish to use in-store displays, for example, would simply offer a zero discount per lb. for displays.
Table 7.3: Default Discounts Offered to Retailer by Manufacturer
| Per-Pound Discount for | |||
| Brand | Feature | Display | Coupon |
| Folgers | $.05 | $.05 | $.05 |
| MaxHouse | $.05 | $.05 | $.05 |
| MasterBlend | $.05 | $.05 | $.05 |
| Hills Brs. | $.05 | $.05 | $.05 |
| Chock Full | $.05 | $.05 | $.05 |
| Yuban | $.05 | $.05 | $.05 |
| C&S | $.05 | $.05 | $.05 |
| AOB | $.05 | $.05 | $.05 |
| APL | $.05 | $.05 | $.05 |
The Run menu controls simulations. Since we are going to vary CFON's price, we can Choose Brands.CFON to toggle select Chock Full O'Nuts as the brand on which this simulation focuses.An ``X'' will appear next to CFON on the desktop. Since we will be going back and forth between the Set Defaults and the Run menus we can lock this toggle switch ``on'' by using the Brand Lock option. Use the < END > and < RETURN > keys to move to the next higher level of the menu.The Variables menu is used for specifying the marketing instruments to vary during the simulation. Selecting Price.Set Minimum and typing 1.6 < RETURN > will set the minimum price to $1.60 per pound. Selecting Set Maximum and typing 2.8 < RETURN > will set the maximum price to $2.80 per pound. The number of steps between $1.60 and $2.80 can be set at any interval, but this simulation uses the default increment of $.10.Select the Default increment [.10]. Use the < END > and < RETURN > keys to move to the next higher level of the menu.Selecting Run.Go will begin the first block of occasions.The occasions will be appended to the current Occasions file on the desktop. If there are no Occasion files on the desktop CASPER will bring up a template which will simply be labeled Occasions. After the first block is completed use the File Management.Rename.Occasions menus to change the name of this file to something more mnemonically meaningful. The first three characters of this file name will be used as part of the identification of all plots generated by CASPER off the file, so that a name like C7Sim might be helpful in remembering that these are the simulations developed in Chapter 7 (up to eight characters may be used).We return to the Set Defaults.Set Values.Marketing Instruments.Set.Set Manually/View three times. First we set as background for the CFON simulation, a sale for Folgers using newspaper features and in-store displays to announce a $2.05 (average sale price for Folgers in the HISTORY file), then return to the Run.Run Settings to reset the price ranges for CFON. Second, we reset Folgers to its default values, set a sale for Regular Maxwell House using newspaper features, in-store displays and store coupons to announce a sale price of $2.10 (average sale price for Regular Maxwell House in the HISTORY file), and return to the Run menu to reset the price ranges for CFON. For the last block of occasions we leave the Maxwell House sale in place and add back the Folgers sale values to the default settings. This simulates simultaneous sales for the three leading brands in this market.
We plot simulation results using the Plot Settings.Build.Using Occasions.Standard X-Y Plots menu. Selecting CFON will cause a pop-up menu to appear listing the following options:
Market Share
Sales
Brand Profit
Store Profit
Total Profit
Done .
Consider first the market-share plot (should be the same as Figure 7.7). The saw-tooth pattern at the top portrays the four blocks of simulations, each with its linear increase in price, in $.10 increments from $1.60 to $2.80. The downward-sloping market-share curves in each block seem to have very similar shapes - differing mainly in the maximum value in each block. In each block we can see how the market-share model deals sensibly with extreme values. A linear market-share model might well predict a negative market share for CFON at the higher prices in this simulation, but the MCI formulation shows that market shares have a natural asymptote. With all other brands at shelf price, CFON's market-share response is strongest in the first block. Facing a Folgers sale in the second block dramatically reduces the market-share potential of CFON. Facing a sale for Regular Maxwell House in the third block seems to be less damaging to CFON's market-share position. This may be partly due to the differences in brand-specific effects between Folgers and Regular Maxwell House, but the $.05 difference in average sale prices for these brands also contributes. Against sales for both Folgers and Regular Maxwell House, CFON's market share performs very like it does facing only a sale for Folgers.
Figure 7.7: CFON's Market Shares - Simulation Results
While the market-share model helps us understand how competitive conditions affect how the total pie is shared among the brands, the category-volume model helps us forecast how large the pie is. Table 5.14 reveals that lowering prices for these three brands has the most expansive effect on total category volume. Figure 7.8 plots the total sales volume associated with this simulation.This plot is formed by selecting Plot Settings.Build.Standard X-Y Plots.TCV. The prices on the right-hand vertical axis are a volume-weighted average over brands.Note that in the first block of occasions the variation in CFON's price leads to about 1,000 lbs.' difference in total category volume. With all three major brands on sale, the price variation leads to around three times as much variation in total category sales. What is showing here is that sales for more than one major brand bring a lot of coffee shoppers into the stores. That around 50% of all sales in the category are made on some kind of a trade deal indicates that many shoppers have been trained to look for coffee deals. That the coffee category is purchased by over 90% of household makes coffee a prime category for promotions aimed at bringing shoppers into the stores.Both statistics are from The Marketing Fact Book , Chicago: Information Resources Inc., 1983.Some undoubtedly come in for a particular brand, while others may well switch brands in the store due to differences in prices, displays, and/or store coupons.
Figure 7.9 shows what these promotions imply for the sales of Chock Full O'Nuts. Even though market share declines for CFON over the four blocks in the simulation, sales increase as the total category volume expands. Some shoppers may be drawn in by the features for Folgers or Regular Maxwell House and switch based on price comparisons at the coffee display.
Whenever there is a constant promotional environment the brand profits will be a constant proportion of sales, so that the sales and brand-profit functions will have identical shapes. If plotted, profits for the brand would always be maximized at the lowest price in each block. This is quite different from the plot of profits for the store (should look like Figure 7.10). Note that $2.00 seems to be the price that maximizes CFON's contribution to store profits in each block. The sum of brand profits and the brand's contribution to store profits is plotted as Total Profits in Figure 7.11. The function is similar in shape to brand profits, except that the fixed cost of the features, displays, and coupons cut into profits at the lowest price levels since margins at that point are smallest.
The higher level of total profits associated with sales for all three
Figure 7.8: Total Category Volume - Simulation Results
Figure 7.9: CFON's Sales - Simulation Results
Figure 7.10: CFON's Contribution to Store Profits - Simulation Results
Figure 7.11: Total CFON Profits - Simulation Results
brands stems from the fact that these three brands can expand total category volume when they are on sale. Competition is normally less fierce among brands which can expand markets, but the market expansion we see here is only for a week. People will switch brands, buy earlier, or increase their inventory of coffee when dramatic sales occur, but the evidence is that in the longer term, they don't drink a lot more coffee just because of price promotions. Look at the TCV plot in CASPER's Standard Graphic Library for support of this claim. Because brand managers need to plan promotions over longer periods, it is important that they are not deluded into believing this level of sales can be sustained week after week. The competitive game incorporates the type of constraints which can only be manifest over time. Before turning to the game, however, we will first look at the simulation results from the perspective of the Folgers and Regular Maxwell House.
In this simulation Folgers' market share always increases as CFON's price increases (see Figure 7.12). Comparing the second and fourth blocks in the simulation, it appears that Folgers market share is hurt much more by the joint competition from CFON and Regular Maxwell House than it is by a promotion for CFON alone. Even facing CFON at an unreasonably low price, a promotion for Folgers at $2.05 per lb. using newspaper features and in-store displays is still expected to draw over 40% of the sales. The sales for Folgers in this simulation are shown in Figure 7.13. Note that the maximum occurs in the second block when Folgers is on promotion and CFON is at $2.30 (as close as possible to the average shelf price of $2.29). Brand profits for Folgers in this simulation would be shaped very similarly to the plot in Figure 7.13. Only the decreased margin of $.10 per lb. supporting the retailers' feature and display cost keeps the plots from being exactly proportional. Folgers' contribution to store profits are plotted in Figure 7.14. These are maximized when Folgers' sales are maximized. When Regular Maxwell House is being promoted, the store's profit contribution from Folgers declines.
From the perspective of Regular Maxwell House the results should look like Figures 7.15 - 7.17. Note that market share for Maxwell House first rises as CFON raises its price, and then Maxwell House's share either flattens out or actually drops (when Folgers undercuts it in price). Maxwell House seems to respond differently to being distinctively low priced than does Folgers. The brand and store profit for Maxwell House are maximized in each block when CFON is far below the maximum price in these simulations.
Market asymmetries are being reflected here. Only recently have
Figure 7.12: Folgers' Market Shares - Simulation Results
Figure 7.13: Folgers' Sales - Simulation Results
Figure 7.14: Folgers's Contribution to Store Profits - Simulation Results
Figure 7.15: Maxwell House's Market Shares - Simulation Results
Figure 7.16: Maxwell House's Sales - Simulation Results
market-response models become available which can capture the diversity of asymmetric competition and yet are practical for use. Only recently have the records of retail transactions become available which allow us to see so clearly what drives consumer purchases. The coincidence of these events has created a new opportunity for brand management to develop plans based on comprehensive data and systematic inquiry. But plans are based on assumptions as well as on data. The propriety of such assumptions is the topic of the next section.
It was emphasized earlier that looking at the data is an indispensable part of building the expertise necessary for good decision making. One of the reasons is that looking makes it easier to evaluate the propriety of assumptions.
The marketing literature on optimal pricing, for example, seeks the single price that maximizes profits under certain assumptions concerning competition.cf. Bass & Bultez [1982], Kalish [1983], Ram & Bass [1985].A look at the data and one sees that there are two price distributions: one corresponding to shelf prices and another corresponding to sale prices. The pulsing back and forth between these two distributions seems far too systematic a pulsing between sale prices and shelf prices to be a search for a single, optimal price. Some chains obviously have the policy of promoting a brand for only one week at a time. If a brand is on promotion this week in such a chain, we are very sure it will not be promoted the next.
As we aggregate store-week data, either over stores or over weeks, we know that our ability to see the underlying process diminishes. But it is just such aggregation that makes the assumptions underlying optimal marketing-mix models appear more tenable. Game-theory models are process models, which is why many of the equilibria can be sought numerically by simply running the game . If the process descriptions only seem apt for overaggregated results, then better models are needed.
There needs to be further development of game-theory scenarios tailored to fit what we see in these data. One interesting possibility combines a ``Colonel Blotto'' game with MacQueen'sMacQueen, James [1988], ``Systems of N-Person Time-Variable Games,'' A talk presented to the Jacob Marschak Interdisciplinary Colloquium On Mathematics in the Behavioral Sciences, Western Management Science Institute, UCLA, January 15.development of N-person time-variable games. The Colonel Blotto game is due to Tukey,Tukey, John W. [1949], ``A Problem in Strategy,'' Econometrica, 17, 73.but it was introduced to us by SteckelSteckel, Joel H. [1987], ``On Using Attraction Models to Allocate Resources in a Competitive Environment,'' Columbia University Working Paper, August 23.in his discussion of resource allocation based on attraction models. Steckel's simulation results are based on seven assumptions:Steckel [1987], p.3.
If we think of the planning horizon as a year then these assumptions provide a well-thought-out set of assumptions. The most controversial assumptions would be items 4 and 7. But at the yearly level we would not expect to see wide swings in primary demand that we see in the weekly data. It is even reasonable to assume that firms can make good guesses of competitors' total promotional expenditures; we simply would not be able to anticipate when or where the resources would be spent.
The biggest problem is that if we calibrate an attraction model to correspond to this planning horizon, we would aggregate away all the interesting behavior. The final allocation model could tell us nothing about when or where to allocate resources. We need to decouple the planning horizon from the periodicity of the market-share model. The Colonel Blotto game helps here by proposing a scenario in which ``two players contending on independent battlefields must distribute their entire forces to the battlefields before knowing the opposing deployment. The payoff to each player on the ith battlefield is a function of the opposing forces committed to that battlefield.''Steckel [1987], p. 17.If the battlefields are thought of as regions, the independence between them is probably not a problem. But if battlefields are weeks , then independence is less tenable. The important feature of the Colonel Blotto game is in the relaxation of item 7. Even if competitors' overall resources can be estimated, their deployment is not known. This game also makes it easier to think of a planning period as a collection of subgames, with an attraction model corresponding to the less aggregate subgames.
MacQueen's game enables a generalization to n persons (m brands in our case) from the two-person case, specifically acknowledges the time-varying aspect of subgames. Casting MacQueen's scenario into our context, we have m brands each of which follows a circuit with numerous nodes. The nodes in our case could be regions or time periods, but are probably best though of as promotional conditions (e.g., shelf prices, sale prices, features). Some of the nodes are common points on different circuits. When two brands jointly occupy the same node they play a game, which can be of different durations and payoffs.MacQueen's study of circuit processes and induced fields has helped him derive the stationary distributions representing the expected time a Markov process is in each particular state (node). His efforts make it mathematically easier to characterize the equilibrium solutions to these processes. Interested readers should see MacQueen, James [1979], ``I. Circuit Processes. II. Notes on Markov Particle Systems with Exclusion and Induced Fields,'' Western Management Science Institute, Working Paper No. 294, University of California, Los Angeles, CA 90024-1481.
While generalizations along these lines hold promise, we are a long way from being able to calibrate a model and then just push an optimality button and receive back the best decisions to make. The lure of push-button management is very strong. We encounter it not only in game theory but also in the development of expert systems .
The essential notion of expert systems is that if we could only capture the knowledge of the experts in computer programs, then all we would have to do is to describe our particular context and the program would tell us what to do. There are three basic ingredients. We need the experts, the knowledge engineers to translate expert knowledge into conditioned-act statements, and we need the computer program to act as repository of the knowledge and as the decision-support system to prompt users through the sequence of questions needed to access the stored expertise.
The expert-systems programs (shells) exist although they will not be reviewed here. Marketing academicians with interest in behavioral decision theory are developing the needed experience with knowledge engineering in areas such as advertising decision making. But analytical methods for representing market response based on scanner data are so new that we don't yet see the experts. The management expertise of tomorrow is built from the structured inquiry of today. The development of CASPER as a FRAMEWORK application is designed to provide an open environment in which expert systems can be developed by the real experts - the managers themselves. Getting from here to there requires that we ask the basic question ¼
What experts know is not innate. It must have been learned. So the task is to structure our inquiry into a market so that we learn. We have looked at history and done one planning exercise in earlier sections of this chapter. But much more effort is needed to understand this market. We could perform this same kind of planning exercise varying Folgers' price or Maxwell House's price. Beyond these basic exercises there are still a huge number of simulations which could be run.
We can look at the competitive maps developed in Chapter 6 for guidance in pruning possibly unfruitful branches of inquiry, and provide more focus to our efforts. From looking at the structure over stores and weeks (Figures 6.3 and 6.4) we already know that the major structural changes in price elasticities coincide with shelf-price and promotions for the three major brands: Folgers, Maxwell House, and Chock Full O'Nuts.
Going back to the map of shelf-price competition in Figure 6.5, we can consider what simulations might help determine a robust shelf price or distribution of shelf prices. For Folgers or Maxwell House simulations should include a look at the AOB category, since the AOB brands are aligned to be most vulnerable to Folgers and Maxwell House. Folgers should look at the influence CFON's and Master Blend's shelf prices on Folgers' shelf price, since Folgers is vulnerable to moves by these brands. Regular Maxwell House should look at the influence of its flanker Master Blend, since Master Blend is aligned competitively at shelf prices.
When investigating Folgers' sale price, the map in Figure 6.6 can help reduce the simulations. Folgers is vulnerable to CFON, but also Hills Bros. and Master Blend are well aligned, although they do not appear to have much clout. Simulations could tell if this alignment translates into a substantial threat. Folgers should also look at its influence on CFON and AOB, for although these brands are somewhat distant, they appear to be quite vulnerable. Simulation to help determine Maxwell House's sale price should include how the AOB category is affected, as well as how vulnerable Maxwell House is to Chase & Sanborne and Master Blend (see Figure 6.7). Chock Full O'Nuts (Figure 6.8) on sale is aligned to attack Folgers, Hills Bros., and Maxwell House (although these brands seem to differ greatly in their vulnerability) and, to a lesser extent, it is aligned to attack PL 2 and Master Blend. On the other hand it seems vulnerable to counterattack from Chase & Sanborne and Folgers.
Three things should be remembered. First, the maps from Chapter 6 relate only to prices. While prices are the major instrument in this market, maps for other marketing instruments might help complete the picture. Second, note that the idealized competitive conditions reflected mostly stores in grocery chain 1. The private-label brands might have a more active role in maps corresponding to other chains. And third, the maps are based on market-share elasticities. The brands which are able to expand the category volume in any given week have a somewhat different set of concerns than those who can only compete in a zero-sum fashion.
The simulation exercise showed that a simultaneous sale for the three biggest and most market-expansive brands are estimated to generate over 8,000 lbs. of coffee in a single week. Even if this were an accurate estimate, no one would expect the same market conditions in the following week to produce the same results. The dynamic effects have to be considered in planning promotions.
There is a difference between demand and consumption. All consumers adopt some inventory policy, at least implicitly. If we stimulate primary demand through promotions, we may be filling up the larder, without any influence on consumption. If this is the case, we can expect further stimulation to become less and less effective as inventories build up in the households. While many aspects of consumer behavior in such circumstances have been the basis of rational speculation in economics, we now have the data to answer some basic questions.
Since the release of IRI's Academic Database on the coffee market in 1983, the academic marketing-science community has learned a lot about the behavior of consumer panels.The Academic Database reported on 78,000 transactions in the coffee market by 1,000 households in each of two small cities, Marion, Indiana, and Pittsfield, Massachusetts. The data used in Chapter 5 to develop the market-share and category-volume models are store-level data from about the same time periods, not these consumer-panel data.One statistic we now have available is the average interpurchase time which is 52.5 days in this category.The Marketing Fact Book , Chicago: Information Resources, Inc., 1983.If we wish to know how much we can expect to expand consumption we aggregate sales over some multiple of the average interpurchase time. Since we also know the variance, we could use two standard deviations above the average interpurchase interval as the range.
If we know this range, but have only the store-level data for further analysis, we could plot a centered moving average of sales for this range of weeks. If the plot is relatively flat, we have no evidence of increased consumption.This plot could reveal seasonal variations in consumption, even if there are no variations resulting from efforts to stimulate demand.If there are variations we could relate them to similarly aggregated variations in the promotional conditions.
If we have the panel data for analysis we could do much more. GuptaGupta, Sunil [1988], ``Impact of Sales Promotion on When, What, and How Much to Buy,'' Journal of Marketing Research , forthcoming. eveloped models for assessing when, what, and how much to buy, using the IRI coffee database. He postulates that consumers first decide when to buy, and then decide what and how much . This modeling framework enables Gupta to develop an overall sales elasticity which can be decomposed into the proportion of sales associated with forward buying, the proportion associated with brand switching, and the proportion associated with stockpiling. He found that ``84% of the sales increase due to promotion comes from brand switching (a very small proportion of which may be switching between different sizes of the same brand). Purchase acceleration in time accounts for less than 14% of the sales increase, whereas stockpiling due to promotion is a negligible phenomenon accounting for less than 2% of the sales increase.'' These models are based on the characterization of the coffee category as being a mature product class, with constant long-term consumption rates. But since variations due to brand switching are independent of consumption rates, we can use these results to put an upper limit of 16% on the combined effects of forward buying, stockpiling, and increased consumption.We could also reasonably assume that forward buying is far more substantial an influence than is stockpiling.The brand-switching component agrees closely with the estimates of McAlister and Totten.McAlister, Leigh & John Totten [1985], ``Decomposing the Promotional Bump: Switching, Stockpiling, and Consumption Increase,'' Talk presented to the Atlanta ORSA/TIMS Conference, November.
These percents are useful in putting constraints on a dynamic simulator. CASPER's dynamic simulator has three promotion periods of nine, nine, and eight weeks, respectively, using data from the 26 weeks reserved for cross validation of the market-share model. The total sales of all brands in any promotion period can expand by no more than 16% above the actual sales in the corresponding period. The percentage can be adjusted to reflect that 16% expansion is not expected to occur each and every period.
CASPER's dynamic simulator is structured as a game. To add a note of realism, Brand Teams develop promotional offers which are presented to the Store Teams or a Game Master. Results from the static simulator can be used to entice stores to accept a brand's promotional offer. While each brand must present the same plan to each store, the stores must act independently on the offers.
Using historic data in a dynamic simulator has some obvious advantages. Applying the cost assumptions to the historic data provides a profit baseline for the stores and the brands. This is a particularly valuable baseline if the stores are run by independent teams, rather than by a Game Master. If only the brands are managed by independent teams, the Game Master can rely on actual market decisions to fill in the unknown conditions. In addition to the pure baseline results, it is straightforward to evaluate one brand's plan against the background data, or to evaluate the brands in head-to-head competition.
One basic cycle of inquiry involves:
The full cycle should precede the first promotion period. Cycles for later promotion periods could begin by trying to reconcile results with the competitive maps in item 4.
This seems to be a lot of effort, but if the information potential of scanner databases is to be tapped, this is the kind of effort which needs to be undertaken.
Those of us involved with management research and education face the dual problems of developing relevant management tools and preparing current and future managers to use them. There are obvious tensions involved. If brand managers were captivated by the complexities of choice models or market-share models, they might well have chosen to pursue academic careers. Those of us involved in methods development are rarely intrigued by the pragmatics of brand management. So where do we meet?
The design of market information systems may be as close as we can get to a common ground. This is the arena in which management science can help provide a systematic basis for utilizing data and market-response models, while management practice can use these efforts in decision making.
The emphasis on real data and real brands makes CASPER a prototype decision-support component of a market information system which could be transported to any of the hundreds of categories for which such data are available. While the development effort needed to implement CASPER-style interface in another product area is far from minor, the end result has some obvious benefits. First, brand managers spend their time learning about market response in their own product area. Second, they must make explicit the assumptions about the competition which are too often hidden or implicit in forecasts or simulations. Third, they are forced to consider the revenue and cost implications of their plans, for the firm and for the channels of distribution.
The most obvious benefit for the academicians is that their talents at research and methods development can be used to advance management theory and practice without being judged for how much they know (or don't know) about selling coffee.