Predicting the Demand for New Products (PDF Available)

HOT NEW RESEARCH COLUMN
PREDICTING THE DEMAND
FOR NEW PRODUCTS
Paul Goodwin
P
INTRODUCTION
redicting demand for new products
poses special problems for forecasters.
By definition, a new product has little
or no demand history. Unless the
demand histories of similar existing products
are available and considered relevant, the use of
statistical approaches to detect and extrapolate
past demand patterns (like exponential smoothing
or ARIMA models) are usually ruled out.
CAN YOU TRUST CUSTOMER INTENTIONS?
One obvious alternative is to simply ask potential
customers whether they intend to buy the new product.
But can you trust their responses?
People may happily tell you that they will be highly
likely to buy your product when it’s launched, and
then fail to make the purchase. In particular, we might
expect prospects to overstate their intention to buy
socially desirable products, like home-gym equipment,
and understate their intentions for products deemed to
be undesirable, like sugary foods. While researchers
have generally found a positive correlation between
stated intentions and actual purchase behavior (those
who indicate they are likely to buy the product do
have a higher probability of purchasing it than those
who say they won’t), this correlation has varied
considerably across different studies. In some cases
it has been very weak.
Vicki Morwitz and her coresearchers (Morwitz et
al., 2007) set out to investigate why this correlation
varies. Their aim was to identify circumstances where
consumers’ indicated intentions can provide a good
basis for sales forecasts, and also situations where
intentions are likely to be unreliable as a guide. The
Paul Goodwin is Foresight’s Research Column editor. His earlier columns covered new approaches to supermarket
forecasting (Summer 2007) and recent studies of forecasting know-how, training, and information sharing
(Spring 2007).
8
FORESIGHT Issue 9 Spring 2008
researchers first conducted a meta-analysis of 40
earlier studies that provided data on more than 65,000
consumers and 200 different products. To verify their
findings, they followed this up with an analysis of data
from 60 separate studies of consumer intentions carried
out by a multinational packaged-goods manufacturer.
Many of the products in these studies were established,
rather than new.
The findings suggest that the faith you can put in the
results of a consumer-intention survey depends on
various factors:
• These surveys are more reliable when the real
purchase decision will be made very soon after the
study. Not surprisingly, consumers can predict their
own behavior better for shorter periods of time into
the future.
a single product variant. It appears that comparing
product variants engages the consumer in explicitly
confronting trade-offs involved in purchasing one
variant rather than another.
• The data from the surveys are better at predicting the
percentage of consumers who will buy the product, as
opposed to total market sales. This is to be expected,
since such studies typically ask prospects about the
likelihood of their buying the product once. Total
market sales will also reflect repeat purchases made
by consumers.
The most important finding of this study, at least from
a new product perspective, is that intentions studies
apparently work far better for existing products than
new ones. Perhaps this is not surprising, as consumers
familiar with existing products are likely to have
experience in using them, so their indicated intentions
are more firmly based. However, as the authors point
out, intentions surveys are used most often in the
very context where they appear to be least reliable:
forecasting the sales of new, nondurable products.
This is a serious concern. More research is needed to
derive improved methods for eliciting the intentions of
consumers in relation to these types of goods.
S-SHAPED CURVES
C
E
S-
U
RV
E
• These surveys work best when consumers are asked
to indicate their intentions about purchasing specific
products within a category (e.g., flavors of ice cream,
different car models), rather than across product
categories (e.g., cars versus notebook computers). They
also work best when consumers are asked to compare
their intentions across different variants of a product,
rather than stating the likelihood that they would buy
H
• The forecasts offered by these surveys are more reliable
for durable than for nondurable goods. Consumers are
more likely to ponder whether to buy a plasma TV, while
a chocolate bar might be bought on a whim.
Of course, intentions studies are not the only method
available for new-product forecasting. A common
finding is that the cumulative percentage of the
population adopting a new product over time follows
an S-shaped curve. Often these curves demonstrate an
increase in the rate of adoption in the early years of the
product’s life, followed by a slowdown in growth as
T
Spring 2008 Issue 9 FORESIGHT
9
the product matures and approaches its market ceiling.
Sometimes the patterns observed in the adoption of
existing products can be used to estimate the patterns
that will apply to analogous new products. A number
of mathematical models are available for representing
adoption patterns, such as the Bass, Gompertz and
logistics models (see Meade and Islam, 2006, for a
recent review of 25 years of research in this area). Once
obtained, these models can be used to predict what the
adoption level will be for a given number of months or
years following the product’s launch.
But arbitrarily choosing a particular model for a given
circumstance can lead to poor forecasts, and methods for
identifying the most appropriate model for a particular
circumstance can be problematical. It is therefore
sometimes advisable to base forecasts on a simple or
weighted average of the forecasts of different models.
In light of these problems, two researchers (Shore and
Benson-Karhi, 2007) have recently proposed and tested
a new method called response modeling methodology
(RMM). This is a general modeling method, which
includes some of the established models as special
cases. The authors argue that the generality of their
method avoids the need for forecasters to identify
which type of model is appropriate for their data. They
say their approach can be used in all circumstances, and
it removes the need to take averages of the forecasts of
different models. In addition, the researchers produce
results which suggest that the method yields more
accurate forecasts than existing methods.
a major retailer, and the game theory component of the
method allowed the relationship between competing
manufacturers and the retailer to be modeled. This
approach enabled the manufacturer to design and price
the product to maximize the chances of its acceptance
by the retailer.
Taken together, these methods show that lack of demand
history does not rule out the use of statistical methods in
new product forecasting. It is likely, however, that new
and innovative methods will continue to be developed
by researchers in the future, given the importance of
these forecasts and the unique challenges they pose.
REFERENCES
Luo, L., Kannan, P.K. & Ratchford, B.T. (2007). New product
development under channel acceptance, Marketing Science, 26,
149-163.
Meade, N. & Islam, T. (2006). Modelling and forecasting the diffusion of innovation – a 25-year review, International Journal of
Forecasting, 22, 519-545.
Morwitz, V.G., Steckel, J.H. & Gupta, A. (2007). When do purchase intentions predict sales? International Journal of Forecasting, 23, 347-364.
Shore, H. & Benson-Karhi, D. (2007). Forecasting S-shaped diffusion processes via response modelling methodology, Journal of
the Operational Research Society, 58, 720-728.
CONJOINT ANALYSIS AND GAME THEORY
Rather than passively hoping that consumers will like
and buy your new product, you can use forecasting
methods at the design stage to increase your product’s
chances of success. Lan Luo and two coresearchers (Luo
et al., 2007) used a combination of conjoint analysis
and game theory models to plan the design of a new
power tool. Conjoint analysis is a statistical approach
designed to predict how customer preference for a
product depends on the product’s different attributes,
like price and packaging. In this case the customer was
10
FORESIGHT Issue 9 Spring 2008
CONTACT
Paul Goodwin
The Management School, University of Bath
[email protected]