The Bass diffusion model of technology adoption
The Bass model describes the growth and decline of new-product sales with two coefficients: the rate of independent adoption and the rate of adoption through imitation. The time of peak sales follows from a closed-form expression, which a simulation confirms.
A new technology can work as intended and still fail commercially if adoption proceeds at a different speed than planned: more slowly than investors expected, or faster than production capacity allows. Adoption follows a characteristic pattern, and technology planning has to account for it.
In 1969 Frank Bass proposed a simple model of this pattern that remains the standard introductory model and one of the most widely used. Two kinds of people adopt a new product: those who decide on their own, prompted by advertising or need, and those who adopt because people they know already have. This article examines how the two effects determine the timing of adoption and of peak sales.
A messaging application lies close to the imitation end: it is of little use until one's friends have it, so adoption creeps along at first and then accelerates as each new user brings others. A cinema film lies close to the other end: audiences respond mainly to advertising and reviews, and attendance is usually highest in the opening weeks.
Bass's paper, published under the transposed title A new product growth for model consumer durables (1969), built on the sociology of diffusion that Everett Rogers had summarised in Diffusion of Innovations (1962), with its innovators, early adopters, early majority and laggards. Bass formalised this account as a differential equation and fitted it to sales of consumer durables. Sultan, Farley and Lehmann (1990) pooled 213 published applications and found average coefficients of about for innovation and for imitation, with time measured in years. The logistic substitution model of Fisher and Pry, discussed in estimating the saturation level of an S-curve, is the special case without innovators.
The model
Let be the share of the eventual market that has adopted by time . Among those who have not adopted yet, a fraction adopts each year on their own, and a fraction adopts because of contact with earlier adopters:
Early on, is tiny and adoption is driven by alone. As adopters accumulate, the imitation term takes over, and growth accelerates until the shrinking pool of non-adopters slows it down. The result is the familiar bell-shaped curve of yearly sales and S-shaped curve of cumulative adoption.
Timing of peak sales
Solving the equation gives
With the meta-analysis averages, the peak comes after about six years, when roughly 46% of the eventual market has adopted. If there is no peak at all: sales are highest on launch day and decline from there, which is characteristic of a product adopted mainly through external influence rather than imitation, such as the film above.
The simulation below illustrates two results: the peak time and the share adopted at the peak follow the two formulas, and for typical products most adopters come through imitation rather than innovation.
- The average product. With and , the simulation and the formula both put the peak at 6.2 years, with 46% adopted. Only about a fifth of all adopters came through the innovation term; the rest were imitators.
- Advertising only. Press "Advertising only ()". There is no hump: yearly adoptions fall from the start, half the market takes 11 years and 90% nearly 30. About 70% of adopters are innovators.
- Word of mouth. Press "Word of mouth (, )". Adoption is negligible for five years and then accelerates: from 50% to 90% in under four years. Innovators account for only 4%. Such a product can appear unsuccessful in year three and reach most of its market by year eight.
- Other parameter values. Pick any and compare the two numbers in each readout. They agree to the first decimal.
Derivation of the peak. Write and . The sales rate is . Differentiating the solution, . As a function of this is maximised where , that is at , giving . Substituting into gives . If , then for all , so the rate is decreasing from the start.
For technology planning the model has two implications. For a product adopted mainly through imitation, early sales say little about the eventual market: low sales in year two are consistent with the model and do not indicate the outcome. Once adoption accelerates it does so quickly, so delivery capacity has to be planned before the evidence for it is available. A messaging service that grows mainly by recommendation therefore needs server capacity for its period of rapid growth before that period begins. Both points are easier to discuss with estimates of and than with a single forecast line.
References
- E. M. Rogers, Diffusion of Innovations, Free Press, 1962.
- F. M. Bass, "A new product growth for model consumer durables", Management Science, 15(5), 215-227, 1969.
- J. C. Fisher and R. H. Pry, "A simple substitution model of technological change", Technological Forecasting and Social Change, 3, 75-88, 1971.
- F. Sultan, J. U. Farley and D. R. Lehmann, "A meta-analysis of applications of diffusion models", Journal of Marketing Research, 27(1), 70-77, 1990.