Documente Academic
Documente Profesional
Documente Cultură
New Product
Forecasting:
The Bass Model
useful forecasts and insights for executive decision about innovations that References
are opening new and emerging markets. Additional Resources
“The Bass model is the most popular model in the field of marketing.”
Prof. Marnick Dekimpe Claims: The Bass Model...
The model is built on two basic assumptions, that potential adopters of an innovation are influenced by two types of communication
channels. Broadcast media and interpersonal channels. Individuals adopting a new product because of what they see or hear from
mass media advertising messages occur continually, but peak early in the product life cycle. And individual adopting based on what
they see or hear from prior adopters. Bass specified the probability of adoption as a linear function of M the total potential mar-
ket, p the coefficient of innovation (external influence) and q the coefficient of imitation.
Some math leads to the basic equation for predicting sales at time t ,
Extensions of the basic model have been proposed to add the impact of marketing variables (the Generalized Bass Model), account
for repeat purchases of a product (Dodson & Muller), allow M, p and q to vary over time.
2
Tips for Using the Bass Model
Estimating M, Market Potential
It can be estimated with the model if early sales data is available. This is rarely the case. I have generally found it
worthwhile to estimate M independent of the model. In most cases management has a judgment, a strong intui-
tive feel, about the size of the market, even though it may be optimistic. If not this estimate can often be ob-
tained from analyst forecasts, marketing research, or guestimates that can be calibrated by testing the logic and
assumption behind the estimate (E.g. Using the Delphi Method). In some markets these estimates can be rather
precise. Pharmaceutical firms for example, often have rather precise estimates of the incidence of a disease or
ailment. It is often worthwhile to obtain an independent third party estimate to calibrate and minimize the risk of
bias and group think.
I have found it useful to treat M as a variable. Assuming a constant growth rate g, over the modeling time frame
has often produced a more accurate and believable forecast. This is an indirect way to reflect the growth that
occurs in market as the average price drops and the demand for the product/service expands. (Note: Bass, Krish-
nan & Jain insignificant results when trying to estimate g using sales data. If in doubt assume M is fixed.) It is also
probably best to treat M as fixed because numerous studies have shown the simple model to very flexible and
robust.
Estimating p & q
Most applications of the model are used to make plans and decisions before the product/service has been intro-
duced to the market. No sales data exists with which to estimate p or q. Managers do not have an intuitive esti-
mate of p and q. The practical approach is to use the coefficients estimated from the diffusion patterns of analo-
gous products. The average values across a wide range of products is p = .03 and q = .38. Industry specific data is
available for consumer electronics, appliances, medical equipment, pharmaceutical drugs, semiconductors , agri-
cultural equipment, etc. Lilien and Rangaswamy).
The best process in my experience, is to use analogues based on the similarities in expected market reactions
rather than the product category. For example, the adoption of satellite radio is more likely to be similar to cable
TV than the adoption of radio. The first generation of radio had no direct competition and was free. Satellite ra-
dio has adopted a subscription pricing model and faces a direct competitor. The same dynamics that cable TV
faced in converting consumers from free TV to cable service. If necessary, we can consider a weighted average of
p and q values across several categories. Or apply a Bayesian weighting that can be updated as new information is
collected.
Predicting My Sales
Most users are not as interested in industry sales as they are in their product/service sales. This often requires
two adjustments to the Bass model forecasts—an estimate of the addressable market and the firm’s share sales to
the addressable portion of the market.
The addressable market is that portion of M that the firm can serve with its current product offerings. For exam-
ple, the FDA may restrict the sale of a drug to children and women who are pregnant. But later formulations may
expand the addressable market to all patients suffering from an ailment the drug can treat.
It is often more difficult to estimate market share. This can be a subjective assessment based on management
judgment. Or it can be assessed using marketing research. In generating a quarterly or annual sales forecast, esti-
mates are required for each period.
. 3
Additional Resources
A good tutorial with accompanying Dodson & Muller, “Models of New Product Advertising and
Word-of-Mouth.” Management Science, Nov. 1978.
case study is in Lilien & Rangas-
wamy’s Marketing Engineering, Lawrence & Lawton, “Applications of Diffusion Models: Some
Empirical Results,” in New Product Forecasting, Eds.
Revised Second Edition. Reading, Wind, Mahajan, & Cardozo (Lexington Books,, 19810
Mass. Addison-Wesley, 1998.
Libai, Muller & Peres, “The Diffusion of Services,” working
The model has also proved valuable paper, 2007.
in predicting the adoption curves of Mahajan, Muller & Bass, “New Product Growth Models in Marketing: A Review and Directions for
2nd, 3rd and successive generations Research,” Journal of Marketing, 1990.
of technology. P and q are the same Norton & Bass, “Evolution of Technological Generations: The law of Capture,’ Sloan Management
for each generation in many cases. Review, 1992.
Norton & Bass. Norton & Bass, “A Diffusion Theory Model of Adoption and Substitution For Successive Generations of
High Technology Products,” Management Science, 1987
Sample of Topics:
Joe Dodson
UW School of Business
Joe_dodson@hotmail.com