X₁ = annual spending in marketing research,

X₂ = annual spending for customer surveillance,

X₃ = annual spending on R&D for unmet but known needs,

a_i = relative profitability of a success from i.

1. Examination of demand factors, including analysis of
a. product life cycle and growth potential,
b. product parameters,
c. appropriate promotional efforts,
d. pricing structure, and
e. distribution strategy,

2. Evaluation of match with company resources, including
a. production,
b. marketing (both markets and physical distribution),
c. materials, and
d. management expertise.

3. Development of costing information, including
a. investment in plant and equipment,
b. interdependencies with other products,
c. material and labor costs,

4. Development of profit calculations, including a. demand estimates,
b. revenue flows,
c. constraints on profit.

5. "Decision" factors, including
a. profit, and
b. risk.

Coordinated with this detailed analysis is product R & D and the development of small-scale production lines. Next, a limited-scale market test is started that includes consumer and dealer surveys, respondent concept/placement (product use) tests, panel tests, and test marketing. Each of these data-collection tasks, is intended to generate information that reduces, or at least gives understanding to the risk of investing in and entering a market. Lest you think we have forgotten about marketing models, this prelim>

Transfer interrupted!

Some of these models use preliminary research and test market data as input. Among these are early models by Fourt and Woodlock, and Parfitt and Collins (both discussed in detail in this chapter). They have been selected for detailed treatment in this book because they have good face validity, are simple conceptually and in parameterization, and are appealing. Both are diffusion models -- they predict the rate of acceptance of the new product, given preliminary test market acceptance rates.

Other models also use preliminary market data. Several use test market data to predict awareness, trial, repeat purchase, and brand share. SPRINTER (particularly its third version, MOD III) is a one of the most complex simulation models ever designed, having over 500 equations. It uses input from store audits, panels, surveys, sales force reports, and so on to provide comprehensive diagnostic information, marketing strategy, and sales guidance. Some models use no market data, but rely on management's subjective judgments. A model developed by the N. W. Ayer advertising agency, for example, uses multiple regression techniques to accommodate estimates of 14 "critical factors" to predict awareness, trial, and repeat purchase three months after product introduction. And DEMON, the grandfather of new product models, uses simulation techniques on three sets of estimates (management policies, product category descriptors, and market performance ratios) to provide output including advertising budget, reach, number of impressions, awareness, and sales.

Implementation
Each of the models surveyed above attempts in some way to minimize management's market-entry risk by providing preliminary estimates of the success of a new product idea. Also, each can help management refine and optimize its entry strategy. When a model is used to "try out" a number of alternative strategies, the model, rather than the market, can show the effects each strategy will produce, and many costly and irrevocable mistakes can be avoided.

While many modeling efforts have been undertaken in the area of product strategy, it is still an area that deserves a great deal more attention. We need useful models early in the development process, before heavy expenditures have been made. (Most of the screening models require actual products to generate their input data ... an expensive step to undertake.) We need models to help management recognize the differences between customers -- using segmentation strategies -- rather than treating all consumers alike. We need more attention to competitive reactions in models, and we need models that are appropriate for products other than frequently purchased consumer goods.

The following sections focus on spreadsheet models provided at the website that accompanies this text.

The Parfitt & Collins Model
Structure of the Model.
Parfitt and Collins conceptualized a simple model that has a great deal of intuitive appeal and has greatly influenced the structure and development of other product models. It predicts ultimate market share for new repeat-purchase consumer products using input data from consumer panels. Although the model requires actual market data (which is expensive since it presumes that the new product is at least in a test market), its ability to predict national share prior to national distribution can help management avoid future losses.

Cumulative penetration (the total number trying the brand, over time) and repeat purchasing rates over time from the time each buyer first bought the product (along with a buying-level index) form the basis for predictions of future share. Trial and repeat purchase rate data is typified in Figures 6-1 and 6-2 for a hypothetical new brand.

Parfitt and Collins represent the ultimate brand share as a composite of these three dimensions: Share = T x R x B where,

To illustrate, suppose we had developed a new lemon-lime cake mix and introduced it in test market. As consumers buy it, the number of triers of our product accumulate, growing in number, but at a diminishing rate. A few months after introduction the shape of this growth curve should become fairly well defined, and a (freehand or computer-aided) extrapolation can be made to the ultimate penetration level (illustrated by the dotted line in Figure 6-1). Similarly, the repeat purchase rate for the brand can be examined.

For example, assume that Figures 6-1 and 6-2 represent the cake mix penetration and repeat, and that average repeat level for our product is equal to the product category. Then our ultimate share is projected to be,

That is, if 34 percent of the potential market tries this new product and 25 percent of the triers repurchase it, and they buy neither more nor less than other brands in the product class, the share for the new product will settle at 8.5 percent.

An appealing feature of this model is that the predicted share can be estimated well before stable shares have been reached, and even while the company is in test market with the product. Too, the diagnostic value of the model should not be ignored. Share estimates below expectation may suggest to management that a change in promotional strategy is necessary to increase penetration (trial) rates, or that a change in product strategy is necessary to increase repurchase rates.

Running the Parfitt and Collins Worksheet

Structure of the Model
Intuitively, we can see that repeat purchase behavior is influenced by the relationships (ratios) between the following:

A low awareness-to-trial ratio (i.e., of those aware of the product, few are trying it) suggests, as a starting point, discovering problems in the promotional campaign, positioning, or media; or problems with the pricing or distribution of the product. Each of these problems could be corrected. But a low trial-to-repeat (or repeat-to-repeat) ratio suggests problems with the product itself. This is a more difficult problem to correct since the product may be failing to satisfy the buyer. If the number of consumers at each step did not diminish (a ratio of 1.0), the product would be a remarkable success. But because each step is a necessary condition for the succeeding step, the number of consumers ordinarily does diminish at each step.

This framework is fundamental to both the Fourt & Woodlock and Parfitt & Collins models; both are new product adoption models for frequently purchased products. The

Fourt & Woodlock model is one of the earliest, best-known, and useful market penetration models.

Their model consists of two procedures: a prediction for first-purchase penetration and a prediction for repeat purchase. In developing the first-purchase penetration procedure Fourt and Woodlock, like many model-builders, observed that, as market penetration of a new product accumulates, diminishing returns cause sales to approach but not reach a "ceiling" or saturation level (Figure 6-5). As they stated, "the increments in penetration for equal time periods are proportional to the remaining distance to the limiting 'ceiling' penetration."

They define the additional penetration for any period as,

Q _t = r P ( 1- r ) ^t-1

where, Q _t = the increase in sales at time t, as a proportion of potential sales,
r = the rate of penetration of unrealized potential sales (a constant),
P = the potential sales as a proportion of all buyers, and
t = time period.

Only two values are necessary for the equation: r and P. For example, suppose we believed that our new product would eventually be tried by 50 percent all households (P = 0.5), and that in each period 30 percent of the remaining potential buyers will actually buy (r = 0.3). Using Fourt & Woodlock's equation, the new-buyer penetration of this market for the first period is, Q₁ = r P(1-r)^1-1 = r P = (0.3)(0.5) = 0.15

or, 15 percent of the market. For the second period, the increment in new-buyer penetration is,

Q₂ = rP(1-r)^2-1 = r P(1-r) = (0.3)(0.5)(1-0.3) = 0.105,

or, 10.5 percent of the remaining market. Subsequent periods would continue to penetrate the market at declining rates. For period three:

Q₃ = rP(1-r)^3-1 = r P(1-r)² = (0.3)(0.5)(1-0.3)² = .0735

Naturally, when the product is actually introduced or test-marketed we can watch penetration during the first few periods and update our estimates of r and P.

Fourt and Woodlock's repeat-purchase procedures build on the above. They examined penetration rates for several new products and observed a curious trend. While the absolute number of consumers at each successive step grew smaller, the ratios between steps grew larger. Typical findings (Table 6-1) include 48.5 percent of the new buyers making repeat purchases, a larger percentage (55.9 percent) of those continuing on to buy once again, and so on. These data are calculated from consumer panel data for the new product.

Fourt and Woodlock used this to predict product sales. First they used results from their first-purchase procedure (discussed above) to predict estimated new-buyer sales. (For example, if 100,000 potential buyers comprise the market for a new product, and 15 percent of these are expected to buy in the first period, we would have 15,000 first-period sales.) If the values of Table 6-1 were observed for our product, we would take the product of 15,000 and the first-repeat ratio of .485 to predict 7275 repeat sales for the first period. Second-repeat sales would be the product of 7275 and .559, or 4067, third-repeat would be 2623 (4067 multiplied by .645), and so on for each level of repeat and then for each period.

Running the Fourt and Woodlock Worksheet

Structure of the Model. The Boston Consulting Group (BCG) was founded in the 1960s by Bruce Henderson on the basis of a single idea that the technological learning curve, (which suggests that with practice, people learn to complete repetitive tasks more quickly) applies to company and product marketing. Henderson coined the phrase "experience curve" and suggested that total company costs decline as production experience increases. Specifically, every time company output doubles, total unit costs decline by a constant percentage (usually around 20 to 30 percent). Thus the market leader (largest share, greatest production experience) can be in the enviable position of having lower costs and higher profits than its competitors.

BCG used the concept of the experience curve to develop an approach to strategic analysis: the BCG Market Share/Market Growth Matrix. This is a model based on market growth and market share that represents expected outcomes for a particular marketing strategy in a particular environment. It is a popular and stimulating approach that has intrigued managers for years. (It is, however, not without its critics.)

Using the BCG Growth Matrix, each of the firm's products is identified as falling into one of four quadrants (see Figure 6-9). Each quadrant also suggests something about the cash flows of products. The names associated with each quadrant indicate the attractiveness of each position. BCG defined the following:

1. Cash Cows. Products that generate more cash than they need to maintain market share. Their role is to provide cash to cover overhead, and to finance growth of other brands, and to pay dividends.
2. Stars. Products which need cash to maintain growth. Their objective should be to maintain a dominant market share. While they often generate their own cash needs, the firm is cultivating them to become cash cows.
3. Problem Children. These do not generate enough cash to cover their own needs because their low shares produce low profits. Their market growth makes them attractive, however.
4. Dogs. While these products usually cover their own cash needs, their low share in a slow-growing market hinders them from ever generating much cash.

The BCG Growth Matrix has several implications. Market leaders are capable of continuing to push their costs down and broaden their market base. And they should do so. Also, depending on the position of the firm's products in the matrix, cash needs vary. Given these varying opportunities, management must decide on strategies that will give the best total corporate performance.

Henry Claycamp provided this description of portfolio strategy:

A company uses the BCG model to develop product strategies. The current position of each product is defined by the relative share and market growth. A future position can be estimated either from a linear forecast of the present situation, or a forecast of the results of a change in strategy. In fact, management should do both and then compare the results. Figure 6-10 shows several strategic decisions made by the company:

• Aggressively support product A, newly introduced, to gain dominance,
• Continue present strategies on products B and C to maintain market share,
• Invest in acquisitions for product D to gain market share,
• Engage in a focused segmentation strategy for product E, and
• Divest products F and G.

FIGURE 6-10

STRATEGIC DECISIONS USING BCG GROWTH MATRIX

Building the BCG Worksheet

Non-metric Multidimensional Preference Analysis (MDS)
Structure of the Procedure
Multidimensional preference analysis refers to a group of procedures for changing one-dimensional measures of relationships into multidimensional measures. This is a remarkable set of techniques that represents products or concepts in geometric space to distinguish between them. It can transform ordinal input data into metric output data. Our discussion will be limited to a single type: scaling of preference data, based on perceived similarities between products.

Preference analysis is intended to show the relationships that underlay the one-dimensional measures, and illustrates two aspects of the relationships: (1) a spatial representation of the relationships, and (2) the "goodness of fit" of the spatial model with the original input data.

In marketing, the input data for MDS models can either be evaluations of the brands on a set of attributes, or customer perceptions of similarity (or differences) between pairs of brands or products.

Any product (or any phenomenon, for that matter) can be thought of as having both perceived dimensions and objective dimensions, and these do not necessarily coincide. A distribution manager for toothpaste, for example, might see the product as a green fluoridated gel, packed in a plastic tube. Toothpaste consumers, however, might see it as a decay fighter that reduces the family dental bills. Another consumer might see it as a product that sweetens breath and brightens teeth, thereby making them more attractive and acceptable when on a date. Preference scaling techniques allow the researcher to represent the consumers' perceptions spatially by creating "spatial maps," (see Figure 6-13) which can help the researcher better understand consumers' perceptual reality.

Although a single map can be used to represent the perceptions of many people, different people (even those whose perceptions are captured in a single map) may perceive the product quite differently. Too, they might attach different levels of importance to the product’s characteristics. For example, one person might evaluate a toothpaste in terms of its tooth-whitening and breath-freshening benefits, while another thinks only of its decay prevention powers.

One input data method for preference analysis is collected by presenting consumers (as respondents) with a set of attributes that describe the brand category. Next, each brand to be included in the map is evaluated on each attribute. The ratings scores are averaged over each of the respondents to produce an [ attribute x brand ] matrix, the values of which are arithmetic means. This process works with any product or brand category where there are differences between brands. For example we could conduct a preference analysis as a taste test for 7 donuts.

Iced cake donut with rainbow sprinkles
Apple fritter
Old fashioned donut iced
White cake donut with icing
Chocolate cake donut with chocolate icing
Maple bar
Regular raised donut

Preference for each of the donuts could be evaluated on several attributes or dimensions:

Other forms of preference analysis rely on evaluating all possible pairs of at least seven or eight products. They are then asked to indicate the relative similarity (or dissimilarity) of each pair (using either rank-ordering or scaling methods), from the most similar pair to the least similar pair. The respondents are ordinarily free to use any criteria they want in making these similarity judgments, although it is possible for the researcher to specify criteria. While consumers could possibly evaluate a product on dozens of attributes, it is believed that they actually use only a few.

Preference analysis software converts these judgments into measures of geometric distance for each brand along an attribute vector such that the preference ordering of the original data is preserved. See the software download for the text to download a copy of MDPREF and the donut data. (mdpref.exe, mdpref.out)

Interpretation of the spatial map can be problematic, as can the determination of the appropriate number of dimensions for it. The amount of variance explained by each dimension is useful here, too, when plotted against the number of dimensions. As a practical matter, maps beyond two or three dimensions create significant problems in presentation and understandability because the output is graphically oriented.

One multidimensional scaling program that is particularly useful is the MDPREF program developed at Bell Laboratories by Doug Carroll and G.G. Chang. The MDPREF method of multidimensional scaling uses brand attribute evaluations to map the positions on the attributes that describe the brands.

The brand evaluations may be collected for an individual, but more often are reported as average preferences that are summarized for groups of individuals, such as a market segment.

MDPREF produces perceptual maps similar to that shown in Figure 6-14. In this example, the points represent soft drinks that were evaluated on a set of attributes that include popularity, sweetness, thirst-quenching ability, calories, carbonation, and so on.

MDPREF is a "point-vector model." That is, the products, (represented by the points in Figure 6-14) are interpreted with respect to the attribute vectors. MDPREF permits the perceptual points to be readily interpreted as having more or less of a particular attribute, but does not allow the interpretation of the relative positions between the brand points. The ability to make point-point interpretation is reserved for other multidimensional scaling models. Figure 8-14 shows a sample output for MDPREF in the analysis of a set of soft drinks.

Running the MDPREF Worksheet

Summary
In this chapter we have examined several important product strategy models. This is a critical area for model building, since so much money is invested into this very risky strategic area. And yet, without this investment, the firm would soon fail.

While much has been done in this area, more attention should be given to product models which can be used early in the product development process, before substantial development funds have been invested -- to search for new product ideas, to screen them, to analyze them for potential. Too, product models are needed to identify segmentation strategies more completely. They need to make better use of available input data: behavioral inputs and competitive reactions, in particular. And, most of the modeling for product strategy has been based on frequently purchased consumer non-durables; more attention needs to be given to durables, and to industrial goods.