Forecasting and Business Analysis Notes

Once a new product has passed the initial screening testsdeemed appropriate for strategic fit and acceptable to consumers, firms want to know if it will be profitable enough. Which requires a sale estimate and business analysis. Sales estimates are necessary for estimates of profitability, ROI or NPV, as well as for production, distribution, and promotion planning. In addition to sales estimates, firms need good estimates of fixed and variable costs associated with production and marketing the new product. It is common for firms to conduct breakeven analyses to determine (a) if the product will break even in an acceptable time and (b) how the sale estimate compares with the break-even volume. A simple breakeven equation: BEQ = FC/(P-VC) BEQ is the break even quantity FC is total fixed costs P is the unit price VC is total unit variable costs Consider a simple example. A landscaping firm considers adding a paint-stenciling service in order to repaint house addresses on curb strips or on the sides of buildings. The new service would require purchase of paint spraying equipment and stencils for $1000, but that cost could be spread over three years. The variable cost of the service would be $15 in labor and $3 for materials. Some preliminary research indicated that consumers would be willing to pay $30 for a typical purchase. How many would the firm have to sell to break even in a year? BEQ=$333/(30-18) = 28 So, if the firm could sell more than 28 paint-stencilings per year, it could recoup its investment. That may not be enough. Firms dont want merely to break even. Ideally, a firm should calculate the ROI for a new product and determine that a new venture offers a better return than other options. In this example, the next best option for that $1000 may be a money fund account, which might return only about 5%. Estimation Challenges Forecasting is not a simple matter. Consumers are not very accurate in telling what they will do. Even if people could articulate their current attitudes, which is far from a sure thing, there are significant gaps between attitude and intention and intention and behavior. Among them: attitudes may be general, but purchase of any product is very specific (e.g., attitude toward purchase of a cell phone should not be a good predictor of purchase of a specific model) attitudes change over time

attitudes are individual; purchases may involve other people intentions are formed for specific contexts, which may not match the purchase situation intentions are formed with expectations for income and time, which may change intentions assume availability and substitutes

Estimations are also limited by the usual problems of conducting social research. Data may be collected in ways that introduce error from group effects, experimenter bias, or demand effects. Group effects result from individuals knowledge that their behavior is known by the groupresponses tend to be less extreme, more conservative. Experimenter bias results from suggestions by the researcher to respond in a given way. These suggestions are rarely overt, but even body language and tone of voice may influence respondents. Demand effects are interpretations from the respondents of the purpose of the research. People often attempt to confirm the suspected hypotheses; but, they may also be deliberately subversive. In general, data collection typically suffers from a lack of external validity. The answers to the estimation research do not reflect the behaviors that might occur in the marketplace. Forecasting demand for any product is difficult, but estimating sales of a new product is also limited by uncertain environmental factors. Neither the firm nor the consumers know what demand drivers will bethe ultimate relevant technology, the competitive offerings, regulations, social trends, economy, etc.. The firm may not know who the proper target segment is. And, the firm may not know or be able to communicate the ultimate marketing plan. Thus, no matter how faithfully consumers predict their behavior, they may not have in mind the world that ultimately will exist. Forecasting Techniques Forecasting techniques fall into three categories. First is regression analysis. Regression analysis is the explanation of one variable in terms of other variables. The second is econometric time series analysis, which is regression analysis with the inclusion of the dependent variable from previous time periods. In other words, sales are estimated for future time periods, using sales figures from past time periods. The third category is educated guessing, which may take the sophisticated forms of jury of executive opinion, scenario writing, or Delphi probes. The two techniques described below are examples of regression and time series approaches. For stable and static markets, i.e., when the market can be described fairly well and is unlikely to be changed much by the introduction of the new product or competitive response, most estimation techniques are based on a market ratio model. A large number, an estimate of the total potential market, is successively reduced by factors that represent the ratio effects of various variables. A typical approach is the following: Q = Pot x T x AW x AV x R x Amt x Freq Q=total quantity sold Pot=total potential market

T=trial rate AW=percent of the market that is aware of the product AV=percent of the market for whom the product is available R=steady repeat purchase rate Amt=amount of average purchase Freq=frequency of purchases The potential market is variously defined in terms of consumers, households, businesses, whatever is appropriate; but it is reduced from the total population by whatever known factors determine the target segments. Trial reflects the proportion of the target that is likely to try the product, given their awareness and the products availability. Trial is typically estimated from concept tests. AW is the percentage of the target market that is aware of the product. AW is typically estimated from the marketing plan (levels and types of promotion) and past experience. AV is the percentage of the target market for whom the product is available. In this case, available means distributed in such a way that the consumer could purchase the product under normal shopping conditions. AV is also typically estimated from the marketing plan (intensivity and type of distribution) and past experience. R is the proportion of those who try the product who adopt it by repeatedly buying it. (Of course, not every adopter will buy the product every time, so R reflects some combination of loyalty and switching in.) R is typically estimated from a combination of past research and simulated market testing. The amount of average purchase and frequency of purchasing are typically estimated from past research, concept tests, and simulated market tests. Two popular consulting services for estimating sales of new packaged goods in stable markets are BASES II and ASSESSOR. BASES II is a service provided by Sami/Burke, a marketing research firm. The basic framework is a hierarchy of effects model: Consumers move through awareness, to trial, to repeat purchasing. Awareness in influenced by the level of promotion; trial is influence by the nature of the product concept, the effectiveness of promotion, extent of distribution, and price; repeat rate is influenced by product quality. A consideration of the proposed marketing plan and category norms results in initial estimates of awareness and availability. A concept test is conducted to estimate trial. All the respondents who give a favorable (usually top or second box) purchase intention response are given the product to try at home. These people are later contacted by telephone for measures of reaction, including repeat rate. ASSESSOR also uses the marketing plan to estimate awareness and availability. Rather than a concept test, ASSESSOR presents a group of people with advertising for the new product. After the ad exposure, the group is introduced to a mock store that includes prototypes of the product, along with competitive products. The participants are given money and encouraged to buy (sometimes forced). Trial is measured by purchases in the mock store. Those who buy the product are re-contacted by telephone for measures of repeat rate.

Contadina Pizza Example Sales of a new pizza product were estimated for two versions (pizza with toppings (kit) or pizza only). Responses were very different for consumers who had used a previous Contadina product, fresh pasta, from those who were non-users. Total population was 95.5 million U.S. households. Contadina use share at the time of the pizza introduction was estimated at 25%. Thus, 22.92 million users and 72.58 million non-users. For the pizza kit option, a concept test indicated 30% top box and 57% second box for users, 15% and 59% for non-users. BASES estimated trial as .8 x top box + .3 x second box, yielding estimates of .411 for users and .297 for non-users. (For the pizza only option, the estimates were .320 and .222). Based on the marketing plan, awareness was estimated at 60% for users and 30% for non-users. Availability was estimated at 58% for all. Thus, adjusted trial rates were 14.3% for users and 5.2% for non-users (11.1% and 3.9% for pizza only). The concept test indicated average purchase amounts of 1.2 units for the kit and 1.1 for pizza only. Thus, sales from trial were estimated at 3934 million for users and 4529 million for non-users (2799 million and 3113 million for pizza only). BASES supplied the favorable responders from the concept test with samples of the product and called them back. Repeat rates were estimated at .22, with two purchase occasions. Sales from repeat rates were estimated at 1442 million from users and 1660 million from non-users (1120 and 1246 for pizza only). Total sales were estimated from trial plus repeat for users plus non-users (plus sales of toppings for the kit option), based on estimated retail and factory prices. The estimates were factory sales of $57.763 million for the kit and $35.264 million for pizza only. BASES results indicate accuracy within +/- 20%. Given the estimated breakeven point of $45 million in factory sales, the estimates did not look favorable for the pizza only option, which would have to be on the very high side of error to be profitable. On the other hand, the kit option was estimated to be profitable even with the worst error. (An examination of the sensitivity to the assumed Contadina market share estimated the breakeven point could be reached by the kit option if Contadina reached a 7% market share.) Forecasting Innovations For innovations and their dynamic markets, ratio approaches do not work. An assumption of ratio models is that Trial is primarily a function of the product and attendant marketing. For innovations, however, product trial is a much riskier proposition; and, consumers are influenced both by the product itself and the behavior of other consumers. If other people are buying the product, consumers are reinforced to try it themselves. Diffusion of innovations models estimate the sales of innovative products by considering the effects of both the product and the rate of adoption. These factors are

referred to as the innovation and imitation effects. The most common model is the Bass Model: P(t)=p + q [Y(t-1)/m] P(t)=probability of trial p=initial probability of adoption Y(t-1)=total number of people who have bought by the end of period t-1 m=total number of potential buyers q=estimated diffusion factor This model suggests that likelihood of purchase is influenced by some initial probability (which depends on the product and its marketing) and by the number of people who have already purchased (which effect depends on the types of people and their visibility). With this first step, one can model sales as the probability of purchase times the size of the market: S(t)=P(t)[m-Y(t-1)] Substituting the previous definition of P(t) into the sales equation, yields S(t)=pm + [q-p]Y(t-1) (q/m)Y(t-1)2 or *S(t)=p [m-Y(t-1)]+ q[Y(t-1) Y(t-1)2/m] The model decomposes sales into those that result from the innovation effect and those that result from the imitation effect. Diffusion models estimate sinusoidal (S-shaped) curves, with sales increasing slowly, then rapidly, then slowing, asymptotically. The shape is due to the diffusion effect. At first, there are few adopters to influence others. Then, as the number of adoptions gets large, the imitation effect is also very large. Over time, the number of adopters to imitate remains large, but the number of people who have not yet adopted becomes small, so the marginal gains drop off. The theory of the diffusion of innovations is based on the assumption that a tendency to innovate is normally distributed in the population and that the rate of diffusion is influenced by various aspects of the product: 1. perceived relative advantagethe bigger the advantage, naturally, the faster the diffusion 2. communicabilitythe more effectively the advantage can be communicated... 3. trialabilitythe more easily consumers can try the product, the more confident they will be of its advantages, ... 4. complexitygenerally, technical complexity reduces communicability and increases switching costs that slow diffusion

5. compatibilitythe less diffusions violate norms, values, and beliefs or present inconsistencies with current standards of technology or behavior, the greater the diffusion 6. perceived riskmonetary, social, and physical risk perceptions slow diffusion The Bass Model reflect the theory. Larger relative advantages, less complexity, greater compatibility, and lower perceived risk are reflected in larger p values. Better communicability and easier trial are reflected in higher q values. Research has shown relatively high p values for hybrid corn and b&w tv. Both represent huge advantages. Interesting that color tv had very low p value but relatively high q value. Perhaps due to switching costs and high visibility and easy trial. Cellular phones also had low p value and very high q value, suggesting perhaps less perceived advantage but very high visibility. A study of a range of applications found average values of .03 for p and .38 for q. Assume a market of 16,000 businesses using a digital enhancement technology, which are prospects for a software innovation. A small research project estimates p and q for the product as .01 and .41. Using the *equation above, sales by quarter are estimated as follows: Q1=.01 x (16,000-0) +.41 (0) = 160 Q2= .01 (16,000-160) + .41 (160 1602/16,000) = 223.3 etc. Diffusion models have been more successfully applied to product categories rather than brands. For the innovative product, this means the estimations must be adjusted for competition as it arrives. A firm, might estimate decreasing market shares over the first few periods if competitive entries are expected. The model also does not account for marketing activity. It assumes effortful marketing, but no allowance is made for better or worse. Clearly, marketing can influence the diffusion factors. The product can be designed and promoted to enhance perceived advantages. Its communicability, trialability, risk, and compatibility can all be affected by marketing decisions. Finally, diffusion models have been designed for durables; they do not capture repeat or replacement sales.