new product forecasting
Solving the problems of new product forecasting
Kenneth B. Kahn 1
Virginia Commonwealth University, 301 W. Main Street, Richmond, VA 23284-4000, U.S.A.
Business Horizons (2014) 57, 607—615
Available online at www.sciencedirect.com
ScienceDirect www.elsevier.com/locate/bushor
KEYWORDS Business analytics; Forecasting; New product forecasting; New product launch
Abstract An important consideration in solving the problems of new product fore- casting entails distinguishing new product forecasting from the process of forecasting existing products. Particular differences between the two can be identified across the dimensions of data, analytics, forecast, plan, and measurement. For example, new product forecasting features little to no data with which to begin the process, whereas data are available and accessible in forecasting existing products. The minimal data situation requires a qualitative approach that lays out assumptions to provide trans- parency; in contrast, quantitative techniques are predominantly used when forecasting existing products. Different assumptions help construct a range of new product forecast outcomes on which company contingencies can be planned versus a singular point forecast for an existing product. And the measure of forecast accuracy, which is a common metric in forecasting existing products, must give way to meaningfulness so that the new product forecast is actionable. Recognizing new product forecasting as a cross-functional, company-wide process helps resolve the problems of new product forecasting. While incapable of remedying all problems, a properly understood and organized new product forecasting effort can help the company better prepare, execute, and support a new product launch, affording a greater propensity to achieve new product success. # 2014 Kelley School of Business, Indiana University. Published by Elsevier Inc. All rights reserved.
1. The daunting nature of new product forecasting
Amidst the current age of big data, advanced statistical analytics, and computer firepower, new
E-mail address: kbkahn@vcu.edu 1 Professor of Marketing and Director, VCU da Vinci Center
0007-6813/$ — see front matter # 2014 Kelley School of Business, I http://dx.doi.org/10.1016/j.bushor.2014.05.003
product forecasting remains a problematic, inherent- ly error-prone endeavor. A research study conducted over a decade ago revealed the average achieved accuracy for new product forecasts at 52% 1 year after launch (Kahn, 2002). Recent industry reports indicate this figure has changed little (Jain, 2007), which highlights that the problems of new product forecasting persist. Even market-leading companies like Kraft, Microsoft, PepsiCo, and Disney contend
ndiana University. Published by Elsevier Inc. All rights reserved.
608 K.B. Kahn
with the challenges of new product forecasting as the following examples illustrate:
� Kraft Bagel-fuls, a bagel-like product stuffed with Kraft Philadelphia Cream Cheese, were launched in 2008 to much marketing firepower. One report indicated initial sales of $7.7 million in the first 12 weeks and revealed that these sales helped offset volume declines in Kraft’s cheese categories. However, Kraft executives had an- nounced expectations for annual sales in excess of $100 million. Whether due to production problems, negative consumer reaction about fat content, and/or the 2008/2009 economic environment, the product did not meet company benchmarks. Kraft divested itself of the Bagel-ful business in 2010, reporting a loss on the divestiture (York, 2013).
� In debuting its Surface RT tablet, Microsoft stated expectations of at least 2 million units sold by end of fourth quarter 2012. Reports generated in early 2013 indicated that only 1 million units had moved, and consumer appeal and market adoption were not materializing into product purchases. Year-end company financial results included a $900 million charge related to Surface RT inventory adjustments, which analysts saw as a telltale sign that the Windows RT tablet was not living up to Microsoft’s sales expecta- tions. Indeed, it was suggested that Microsoft had up to 6 million Surface RT tablets in unsold stock, not including parts and accessories (Hernandez, 2013).
� PepsiCo launched its Gatorade G-Series Fit line in April 2011 to high expectations. The Fit line–—comprised of a pre-workout protein bar, a fruit-based protein drink, and a low-calorie version of Gatorade sports drink–—was designed to woo fitness fanatics who preferred drinking water. The product line failed to catch on with its intended audience and did not resound with the traditional Gatorade customer, which tended to be teenagers and athletes who play team sports. Price was also an issue, as the G-Series Fit line was more expensive than traditional Gatorade and launched amid an emerging price war with competitors in a saturated energy-drink market. PepsiCo spokes- person Molly Carter stated: ‘‘While we made strides connecting with this athlete, the line did not perform to our high expectations’’ (Ziobro, 2012). Gatorade G-Series Fit was removed from the marketplace in 2012, after just 1 year on the market (Morton, 2012).
� Disney hoped that The Lone Ranger would turn into another Pirates of the Caribbean franchise, which grossed over $3 billion worldwide. Prior to the movie’s launch in July 2013, BoxOffice.com predicted sales revenues of $37 million over open- ing (3-day) weekend, $60—$70 million in North America over the same period, and $135 million in its total domestic theater run. Another forecast by Box Office and Film News suggested that the movie would make $127 million in North America and $185 million outside North America for a worldwide total of $312 million. Actual figures during opening weekend were $29.3 million, with $48.9 million over the first 5-day time frame. A month after its release the film had earned $86.7 million in the United States/Canada and $88.7 million elsewhere for a worldwide total of $175.4 million. The New York Times estimated the film had cost $375 million to produce and market, and would need to earn an estimated $800 million worldwide to break even after accounting for revenue splits with theater owners. Walt Disney Studios Motion Pictures’ vice president Dave Hollis called the results ‘‘very disappointing’’ (Fritz, 2013, p. B1).
Viewed through the lens of contemporary thought regarding big data and advanced analytics, the aforementioned situations would have generated better forecasts had more data and sophisticated analytics been readily available. Such thinking im- plies that forecasting requires one to simply access company databases and/or syndicated datasets, extract and analyze product demand history, and generate a forecast. This portrays forecasting as a strictly data-crunching endeavor. Unfortunately, such a sketch overshadows and misrepresents the task of new product forecasting.
Accessible data are a rare luxury when forecast- ing new products, and whatever data do exist typi- cally are not in immediately usable form. For instance, new product data often include customer interview data, which are more likely to come in the form of audio and video files and/or be summarized in PowerPoint presentations than coded and housed in a computer database. Qualitative analyses serve as the norm for new product forecasting, with judg- ment and intuition as inputs. Complicating and compounding the situation are the biases that these inputs introduce and the inherent uncertainty sur- rounding factors like market size, penetration rate, cannibalization, and competitors. Such factors con- tribute to new product forecasting being a charac- teristically problematic and error-prone endeavor.
As will be discussed, there is a need to distinguish the task of new product forecasting from forecasting
Solving the problems of new product forecasting 609
existing products or services across the dimensions of data, analytics, forecast, plan, and measure- ment. By understanding these differences and rec- ognizing new product forecasting as more than just a data-crunching exercise, companies mitigate the daunting nature of new product forecasting, pro- duce meaningful forecasts, and have the propensity to enjoy greater new product success. Specific rec- ommendations for solving the problems of new product forecasting are provided at the conclusion of this article.
2. Forecasting existing products and new products
The distinct nature of new product forecasting was highlighted by a major benchmarking study on fore- casting practices, whereby 400+ companies were surveyed and 23 company site visits were conducted (Mentzer, Kahn, & Bienstock, 1999). Findings clearly showed that new product forecasting was a pain point for companies. This spurred subsequent research, including a Product Development and Management Association (PDMA) benchmarking study on new product forecasting practices and a survey co- sponsored by the SAS Institute (2012), both of which reaffirmed the unique nature of new product fore- casting and the need to better understand how to manage the endeavor separately from forecasting existing products. Separating new product forecast- ing from the task of forecasting existing products also focuses attention on the key elements necessary to underlie a better launch plan (Haines, 2009).
Using a common set of dimensions to describe the forecasting task, the distinctions between new product forecasting and forecasting existing prod- ucts are highlighted. The dimensions include:
� data gathered to serve as the input for the fore- casting endeavor;
� analytics employed to examine data, toward ex- tracting understanding and meaning from it upon which to project a demand pattern;
Table 1. Comparing forecasting existing products versu
Forecasting Existing Produc
Data History
Analytics Statistical
Forecast Point
Plan Certainties
Measurement Accuracy
� a forecast generated from the analytics con- ducted, to provide a reasonable estimate of what can be attainable under a given set of conditions;
� a plan based on the forecast, to organize the company’s response to the forecast; and
� measurement used to assess forecasting perfor- mance.
As shown in Table 1, new product forecasting data are not necessarily numeric. The analytics used to forecast new products are less about number crunching large data and more about providing key assumptions and business drivers that impact business capabilities during product launch. The new product forecast is not about hitting a specific num- ber, but specifying a range. The new product plan must account for contingencies due to inherent un- certainties. And performance measurement should not fixate on accuracy. Each dimension is now further discussed.
2.1. Data
For existing products, demand history can be amassed over a given time period to create a data time series. Other data that may be amassed and connected with the product demand history include pricing, promotion events, and holidays occurring during the respective time period. Weather and other acts of nature (e.g., blizzards, hurricanes, earthquakes) plus non-typical events (e.g., plant fires, strikes) can be documented, too, and input into the time series dataset to explain anomalies in the demand history.
New product forecasting lacks hard numbers be- cause there is no demand history for the product prior to launch. The demand history for parallel products can sometimes be employed as analogous data for the forthcoming new product; however, there are no assurances that historical demand for a comparative product will correspond equally to the new product. One or more assumptions–— typically, multiple–—are therefore made about the
s new product forecasting
ts New Product Forecasting
Assumptions
Judgmental
Range
Contingencies
Meaningfulness
610 K.B. Kahn
new product such that the new product forecasting endeavor can begin. Assumptions are used in con- junction with any data collected to forecast the new product or are used in lieu of specific data. For example, two simple assumptions might be that new product sales will be 5% higher than previously released products and that a 10% market share should be attainable. The latter assumption illus- trates that new product assumptions may not be sales history-specific but descriptive of the market conditions, which are then translated into a de- mand expectation. When a budget is available, assumptions can be developed and validated using customer/market research.
2.2. Analytics
Because data are available, forecasting existing products emphasizes quantitative analysis using at least one statistical forecasting technique to create a baseline demand forecast. Time series forecasting techniques, which attempt to detect historical de- mand patterns and construct a representative graph or formula to project these patterns into the future, are common for this forecasting task due to the prevalence of time series data. Should there be more than just time series data, regression model- ing techniques can be used by which exogenous or independent variables are correlated to demand history in order to develop formulae and predict demand.
New product forecasting predominantly involves qualitative analysis due to minimal or no data. Judgmental forecasting techniques become the norm in an attempt to turn experience, judgments, and intuition into formal forecasts, with the aim to make sense of these data and corresponding as- sumptions. Jury of Executive Opinion and Sales Force Composite are two of the more popular judg- mental forecasting techniques (Kahn, 2002). Jury of Executive Opinion is a top-down forecasting tech- nique via which the forecast is arrived at through the ad-hoc combination of opinions and predictions made by informed executives and experts. Sales Force Composite is a bottom-up forecasting tech- nique via which individuals, typically salespeople, provide their forecasts; these forecasts are then aggregated to create a higher-level, composite forecast. Assumptions-based modeling is another judgmental technique that attempts to model the behavior of the relevant market environment by breaking the market down into market drivers. Forecasts are generated by assuming values for these drivers. Assumptions-based models are also referred to as chain models, market breakdown models, or waterfall models.
Diffusion models are a set of stochastic modeling techniques that have traditionally been used for new product forecasting. These models have been shown to capture the lifecycle dynamics of a new product; estimate demand in a new product cate- gory; and direct pre-launch, launch, and post- launch strategic choices (Mahajan, Muller, & Wind, 2000). However, diffusion models have two chal- lenges: (1) their complexity and (2) their reliance on a priori information to forecast. Diffusion models also have a poor track record in actually predicting sales more accurately than other methods (Goldner & Tellis, 1998). And while the incorporation of external variables makes these models more prac- tical, adding variables makes the modeling more statistically complex. Consequently, company exec- utives are trending toward the use of simple models that incorporate judgmental techniques, and are finding that simple models are more readily able to be built and implemented (Fader & Hardie, 2005).
2.3. Forecast
The number-crunching orientation of forecasting existing products emphasizes deriving a specific number on which decisions are made. This charac- terizes the forecasting of existing products as point forecasting, the focus of which is to generate and subsequently hit a specific number. The tendency to focus exclusively on the forecast number is prob- lematic when forecasting new products because the variability in model assumptions presents a range of possible forecast outcomes. Focusing on a specific number during the new product forecasting effort also fails to take into account the inherent uncer- tainty that pervades any new product forecast (Joseph & Finney, 2006).
Range forecasting allows new product forecasting to portray the uncertainty surrounding the forecast, such that the company is attuned to pending risk and can determine whether such risk is acceptable. Range forecasting does this by communicating the potential upside–—and downside–—of the opportuni- ty and the relative probability of these occurring. Usually these are presented as pessimistic (worst- case), likely, and optimistic (best-case) scenario outcomes.
2.4. Plan
In the case of both forecasting existing products and new product forecasting, once a forecast is ap- proved, it enters a planning process to help establish operational decisions. When forecasting existing products, the forecast is the basis on which safety stock is calculated. A safety stock policy might be
Solving the problems of new product forecasting 611
one or two standard deviations below and above the forecast, determined by analyzing demand history with the objective to ensure 100% customer service levels. The use of safety stock and other operational policies derived from the forecast intend to guaran- tee certainty around where demand may fall in order to achieve superior customer service.
New product forecasting is less about certainties over a given time period and more about contingen- cies. This is because new product forecasting is inherently uncertain due to the aforementioned range of possible forecast outcomes; therefore, the plan is oriented around risk management (Finney & Joseph, 2011). Use of an assumptions- based model combined with risk analysis can evalu- ate how slight deviations in model assumptions might change the new product forecast and illus- trate the potential for low or high demand based on these deviations. The spread between these two demand situations may be too large for the company to manage. Contingency planning is imperative dur- ing new product forecasting such that the company is prepared to respond to potential outcomes. For ex- ample, what is the company’s response if market awareness is lower than expected? What is the com- pany’s response if trial is stronger than assumed?
2.5. Measurement
Accuracy tends to be the predominant metric for measuring the forecasting endeavor. The relative ease and speed with which accuracy can be calcu- lated when data are plentiful make it a mainstay metric in company reporting systems. However, the problem with using accuracy to evaluate new prod- uct forecasting lies in its expectation of preciseness. The reality for any new product forecast is that it will fall far short of being highly accurate. Consider the findings of the aforementioned benchmarking study that reported the average achieved accuracy for a new product forecast at just 52% 1 year fol- lowing launch (Kahn, 2002). This brings into ques- tion whether accuracy is the best measure in the new product forecasting context.
Companies successful at managing the new prod- uct forecasting endeavor are embracing the notion of meaningfulness rather than accuracy. Meaning- fulness mandates that companies accept error will exist. The focus then shifts to developing various forecast scenarios, understanding the assumptions that underlie these, reaching agreement regarding the most likely scenario, and planning contingencies should other scenarios come to pass. Meaningfulness infers that the forecast is usable and allows for the company to plan properly around expected error. Reflecting this mindset, a supply chain director
commented: ‘‘I know the new product forecast is going to be wrong, but as long as that forecast can get us into the ballpark, we can plan accordingly.’’ In short, an accuracy orientation implies that error will not occur; a meaningfulness orientation expects and plans around error.
3. A case example
A mid-size manufacturer of computer networking equipment was preparing to launch its first software product: a network security solution that would provide comprehensive protection of a networked environment. The key selling points of the software were enhanced operational efficiency through customizable, centrally administered configuration tools and an automated system that isolated and mitigated security threats. The software launch was particularly noteworthy because the manufacturer had strictly sold routers, switches, wireless access devices, and other related hardware products up to this point in time. The target market was senior information technology personnel wishing to better secure their companies’ existing computer server networks.
A cross-functional team meeting was held to discuss the forthcoming product, with product management assuming its usual assigned role of leading the new product forecasting process. Join- ing the meeting were the product manager and technical lead overseeing the software’s develop- ment and launch, the director of product manage- ment, lead product managers for other major product lines, the director of sales, the director of sales support, and the director of the sales and operations planning process who served as the operations representative. A consultant served as team facilitator.
Because the company had never launched a soft- ware product, sales history was nonexistent. Fur- ther, the company sales system only reported quarterly sales data, which limited the use of sta- tistical forecasting techniques due to the aggregate nature of the data. There was no market research department or market research budget, so tradi- tional market research was not an option. The company had purchased a market research report from the Gartner Group to help size the market opportunity.
It was quickly realized that the data situation limited use of any sophisticated statistical tech- nique. A judgmental forecasting technique became the focus, with a preference to construct an as- sumptions-based model. A discussion among the team members led to an iterative model-building
612 K.B. Kahn
session that identified assumptions and their rela- tionships. The final model comprised the five linked assumptions of market size, core use, company market share in that application space, percent of market ready to buy, and company sales coverage. While other model components could have been included, these five model components were viewed as most relevant; more importantly, each of these specified components could be quantified based on existing data sources.
Market size was predicated on the purchased Gartner Group study and supplemented by customer interviews conducted by the product management group. The estimated market size for the software security solution was $3 billion.
Core use was defined as the percent of the mar- ketplace using hardware that would accept the security software on its core system; that is, the software solution could be used as the primary security system or as a peripheral, back-up system. Serving as the primary security system was the company’s focus at the time of launch. A value of 65% was determined through interviews with sales management and product management groups in the company, suggesting that two-thirds of prospec- tive customers would be looking for a primary secu- rity system.
Company market share in the core use segment was estimated based on published industry reports noting competitor market shares, and was supple- mented by sales management and product manage- ment personnel intuition. Market share was estimated to be 20%.
Buying intent was defined as the percent of the market interested and likely ready to migrate to the new core technology. Sales management was the
Table 2. Case example base case and risk analysis
Base Case Analysis
Market Size Core Use Market Share
Buyin Inten
$3 billion 0.65 0.2 0.25
Risk Analysis
Assumptions Base Low Hi
Market Size $3 billion $3 billion $3 bi
Core Use 0.65 0.4 0.8
Market Share 0.2 0.1 0.3
Buying Intent 0.25 0.2 0.3
Coverage 0.8 0.8 0.95
predominant source for this value, and indicated that a follow-up employing its sales pipeline tool would be used to validate the number. Buying intent was estimated to be 25%.
Company market coverage was based on current worldwide sales networks. This value represented the extent of distribution the company held global- ly. An 80% market coverage rate was given, indicat- ing that the company could serve 80% of world markets through its existing distribution system.
The assumptions-based models framework was applied by multiplying the values of the five model components. The forecast for the new software solution in the first year was $78 million ($3 billion x 65% x 20% x 25% x 80%). A risk analysis was then applied to develop a range forecast, whereby the original $78 million calculation represented the likely case. Discussions among the team laid out potential best-case and worst-case scenarios. It was agreed that market size would remain a con- stant. Core Use had the potential to run as high as 80%, but also as low as 40%. History indicated that market share regularly fluctuated between 30% in good months and 10% in difficult months. Buying intent was seen as a variable falling between 20% and 30%. And market coverage, which was viewed as the most certain of the given assumptions, had the potential to increase to 95% based on distributor growth. Holding all values constant to the base case and changing only the assumption under consider- ation determined a worst case of $39 million in sales revenue; the optimistic case was calculated at $117 million (see Table 2).
When these forecasts were compared to manage- ment expectations, concerns arose. The worst-case scenario was particularly troubling. Management
g t
Market Coverage
Base Case
0.8 $78 million
gh Pessimistic Case
Optimistic Case
llion $78 million $78 million
$48 million $96 million
$39 million $117 million
$62.4 million $93.6 million
$78 million $92.625 million
Solving the problems of new product forecasting 613
had presumed sales revenues close to $90 million and preferably closer to $100 million. Further meet- ings led to the resetting of management expectation at $82.5 million and the development of contingen- cy plans to contend with the low and high market scenarios. And although accurately meeting the forecast was a primary point of attention, consensus was reached regarding the addition of measures–— including inventory levels, customer service levels, and launch schedule timing–—as part of a launch scorecard.
Actual sales for the first year came in close to $80 million. While 3% below forecast, the assumptions- based model approach averted a much larger error; had original expectations gone unchallenged and had management remained fixated on the optimistic sales target, the company would have endured a $20 million shortfall versus the $2.5 million it actually experienced. The assumptions-based model ap- proach also afforded company management a better understanding of business drivers, enabled a more robust discussion about what would drive sales, and presented a logic framework to forecast future software solutions. The cross-functional approach affirmed that cross-company expertise is important in forecasting new products, especially as regards launching a new product category. The market ex- pertise of product management and sales was crucial to the modeling exercise and resulted in a more accurate prediction. The broader view surrounding metrics was perceived to have provided better con- trol of the launch situation and was applauded by operations because it accounted for and gave visibil- ity to operational and supply chain factors such that management could observe the entire launch situa- tion, not just a sales figure compared to a forecast.
4. The process of new product forecasting
As shown in the case example, new product fore- casting requires traversing functional units of the company–—including engineering, marketing, oper- ations, and sales–—to best develop assumptions and ensure that focal aspects of the new product launch are incorporated into forecasting and planning ef- forts. Collecting inputs from across the company helps mitigate one mindset or one functional area’s bias. It also helps to temper company politics by allowing all functional domains to have a role in generating the new product forecast and to see those elements that underlie new product-launch decision making. Recognizing the process of new product forecasting underscores a further consider- ation: a mindset that favors big data, advanced
analytics, and data crunching typically implies that forecasting automation–—via a system–—is possible such that an analyst alone can generate the fore- cast. Automation and ‘auto-forecasting’ are not the right course for new product forecasting due to the underlying relational, cross-functional aspects.
Additionally, the notion of meaningfulness plays a role in facilitating the cross-functional nature nec- essary in generating forecasts and demonstrating performance measurement. Meaningful perfor- mance measures should be readily translated and understood by the multiple functional areas in- volved. The measures should connect to issues and problems that these functional areas must ad- dress.
While the case example illustrates a product solution, other contexts equally contend with the problems of new product forecasting. This is be- cause all types of offerings (e.g., product or service, final consumer market or business-to-business mar- ket, small company or large company) face new product forecasting problems across data, analytics, assumptions, planning, and measurement dimen- sions. Most new product forecasting situations also have big decisions connected with them, with these decisions often hinging on the generated forecasts. Various research studies concur by finding that the practice and problems of new product forecasting are similar across different contexts; for example, consumer and industrial contexts (Herbig, Milewicz, & Golden, 1994) and low tech versus high tech (Lynn, Schnaars, & Skov, 1999). This means that the process around new product forecasting should not be taken lightly or disregarded in any context. Even entrepre- neurial ventures should approach new product forecasting in a fashion similar to those of other contexts. The value of the assumptions-based model will be especially beneficial in helping both the entrepreneur and investors understand the business model and envision the underlying relationships that drive the business model.
Two areas of difference appear consistent, though. As one might expect, business-to-business markets tend to rely more heavily on their sales force due to its predominant role in securing sales. Consumer markets tend to rely on consumer re- search studies due to the quantity of customers (Kahn & Mentzer, 1995). A second difference is the time horizon across contexts, which is particu- larly influenced by type of industry. For example, new product forecasts of consumer grocery product companies tend to reflect shorter time horizons than those of business-to-business markets. Apparel and electronics firms tend to have the shortest time horizons; pharmaceutical products have the longest time horizons (Jain, 2007).
614 K.B. Kahn
5. Recommendations for action
When forecasting existing products, one typically accesses and analyzes data using a statistical tech- nique embedded in a computer system to produce a number. Data crunching and statistics are focal points toward generating a number that is accurate and affords certainty. By comparison, new product forecasting starts with little data and–—using a mostly qualitative approach predicated on assump- tions–—attempts to provide understanding, trans- parency, and realistic expectations. A range of possible outcomes are generated as the foundation for company planning. The orientation is to estab- lish meaningful, but not necessarily accurate, fore- casts that help to set realistic expectations via which managers can make the most valid and real- istic new product launch decisions supported by contingency plans. The differences between fore- casting existing products and new product forecast- ing are therefore more than just subtle and can contribute to significant problems if not properly recognized and addressed. The dimensions of data, analytics, forecast, plan, and performance mea- surement represent important means to organize, understand, and put in place action steps to remedy potential problems.
Recognition of the data difference between fore- casting existing products and new product forecast- ing becomes particularly evident when forecasting software is used. Such software is oriented toward the existing product forecasting situation, as nu- merical demand history over time is required for analyses to be conducted. Some forecasting soft- ware may allow the use of analogous data, yet many forecasting software are not designed to allow users to input and apply assumptions to manipulate data, especially those assumptions not directly connected to data time series.
The analytics difference is evident when model transparency becomes a keen interest. Transparen- cy can be defined as clarity regarding how the forecast is derived and the conditions under which the forecast is calculated so potential biases influ- encing the forecast may be identified. Many models used to forecast existing products do not break down assumptions and at times can represent a black box. New product forecasting needs the ability to chal- lenge underlying assumptions in the forecast model such that the inputs and assumptions surrounding their effects may be discussed. Attention is paid to how the forecast happens, not just what the fore- cast is. Such attention to assumptions limits the tendency of senior managers to argue about the numbers in isolation (Joseph & Finney, 2006). Con- sider the example of one company in which a senior
manager insisted that the forecast had to be 10% higher to match up with early management expec- tations. Resisting the normal inclination to adjust the forecast, the forecasting manager presented the forecast model and asked which of the assumptions should be adjusted, deflecting attention from the model’s output. Deliberations among the senior management team could not refute the assump- tions, which brought about the realization that end sales might actually be lower than originally expected. This spurred a necessary discussion over if and when the new product should launch.
In lieu of a single number, it is crucial to empha- size ranges when forecasting new products; senior management will thus be enlightened and better able to judge and accept the risk that will be undertaken. Delineating a range of outcomes fur- ther helps the organization develop a collective understanding of the inherent uncertainties as it envisions the business opportunities connected with the new product. At times, new product forecasts must serve the crucial but unpopular role of pre- senting unpalatable options. A numeric forecast, by itself, does not provide such enlightenment.
Contingency planning during new product fore- casting can be enhanced by making it a standardized practice. For example, some companies use a launch control protocol that lists potential problems that may arise during the new product launch and outlines ways in which these problems can be moni- tored and controlled. Key trigger points are identi- fied to indicate at what point action needs to be taken, as well as what type of action is required. Undertaking such contingency planning avoids un- necessary confusion and ‘firefighting’ during prod- uct launch. Table 3 illustrates a sample launch control protocol for a consumer product. As shown, the problem of customers not making trial purchases of the new product as expected is tracked by point- of-sale data. This problem is signified by a trigger point of less than 100 unit purchases per retail outlet per month. The corresponding remedy is implemen- tation of point-of-purchase displays.
Regarding performance measurement, a single focus on accuracy will only lead to frustration due to the low accuracy problem of new product fore- casting. A predominant focus on accuracy also overlooks superordinate outcomes like improved customer service, reduced costs, and/or improved margins. Use of multiple new product forecasting metrics provides greater visibility to the new prod- uct launch. Reporting revenue, shipments, invento- ry, and customer service, for example, will result in a more robust view of the new product situation and connect to interests and issues facing multiple func- tional areas. Metrics should be presented in usable
Table 3. An example of a launch control protocol
Potential Problem Tracking Contingency Plan
Customers are not making trial purchases of the new product as expected.
Look at POS reports. Minimum, 100 purchases monthly per retail outlet are expected.
Install point-of-purchase displays.
Competitor may have similar new product.
Difficult to track, but conduct surveys with retailers and final consumers.
Offer 2 for 1 program. Consider bundling new product with other products.
Solving the problems of new product forecasting 615
and understandable forms across functional areas, too, such that company-wide decisions may be made. This highlights the need to consider new product forecasting as a process that is integrated into the overall company decision-making process.
6. Conclusion
In sum, new product forecasting needs to be viewed as a different type of effort versus forecasting ex- isting products: the mindset for undertaking the new product forecasting task is not data, techni- ques, and systems, but rather quality assumptions, judgment, and process. Through a concerted effort that extends well beyond data crunching, new prod- uct forecasting will provide insights that enable senior management to better evaluate risk and decide how to act. This broader thinking around solving the problems of new product forecasting may not ensure accuracy, but it increases the pro- pensity for new product success.
References
Fader, P. S., & Hardie, B. G. (2005). The value of simple models in new product forecasting and customer-base analysis. Applied Stochastic Models in Business and Industry, 21(4/5), 461—473.
Finney, A., & Joseph, M. (2011). Getting your forecasting and planning fundamentals right. Foresight, 20(1), 29—36.
Fritz, B. (2013, August 6). Disney warns of ‘Lone Ranger’ loss; ESPN, theme parks offset studio weakness. The Wall Street Journal, p. B1.
Goldner, P. N., & Tellis, G. J. (1998). Beyond diffusion: An affordability model of the growth of new consumer durables. Journal of Forecasting, 17(3/4), 259—280.
Haines, S. (2009). The product manager’s desk reference. New York: McGraw Hill.
Herbig, P., Milewicz, J., & Golden, J. E. (1994). Differences in forecasting behavior between industrial product firms and consumer product firms. Journal of Business and Industrial Marketing, 9(1), 60—69.
Hernandez, P. (2013, July 29). Microsoft Surface RT’s weak sales: Ballmer’s take. eWeek, p. 4.
Jain, C. (2007). Benchmarking new product forecasting. Journal of Business Forecasting, 25(4), 22—23.
Joseph, M., & Finney, A. (2006). Foreword. In K. B. Kahn (Ed.), New product forecasting: An applied approach. Armonk, NY: M. E. Sharpe, (pp. xi–—xiii).
Kahn, K. B. (2002). An exploratory investigation of new product forecasting practices. Journal of Product Innovation Manage- ment, 19(2), 133—143.
Kahn, K. B., & Mentzer, J. T. (1995). Forecasting in consumer and industrial markets. Journal of Business Forecasting, 14(2), 21—28.
Lynn, G. S., Schnaars, S. P., & Skov, R. B. (1999). A survey of new product forecasting practices in industrial high technology and low technology businesses. Industrial Marketing Manage- ment, 28(6), 565—571.
Mahajan, V., Muller, E., & Wind, Y. (Eds.). (2000). New-product diffusion models. New York: Springer.
Mentzer, J. T., Kahn, K. B., & Bienstock, C. (1999). Benchmarking sales forecasting management. Business Horizons, 42(3), 48—56.
Morton, A. (2012, August 13). PepsiCo to pull G Series Fit after poor performance. Just-Drinks Global News. Retrieved from http://www.just-drinks.com/news/pepsico-to-pull-g-series- fit-after-poor-performance_id107856.aspx
SAS Institute. (2012). New product forecasting survey highlights. Cary, NC: SAS Institute Inc.
York, E. (2013, February 20). Kraft back on successful flight path for new product launches. Chicago Tribune. Retrieved from http://articles.chicagotribune.com/2013-02-20/business/ ct-biz-0220-kraft-20130220_1_new-products-barry-calpino- kraft-foods-group
Ziobro, P. (2012, August 10). PepsiCo pulls Gatorade ‘Fit’ line over poor sales. MarketWatch. Available at http://www. marketwatch.com/story/pepsico-pulls-gatorade-fit-line- over-poor-sales-2012-08-10