Demand forecasting is a mixture of art and science. It is also a wide topic, and unsurprisingly, there is a large divide between the theory, often complex and difficult to implement, and the level of simplicity required in business practice. The deeper you dig into the forecasting instruments available, the more sophisticated they become, while often adding only marginal improvement to the forecasting quality.
Demand forecasting is closely related to the adoption of innovations within a target population. Everett M. Rogers, who wrote “Diffusion of Innovations” more than 40 years ago, has provided us with a powerful framework. Rogers extensively investigated how innovations diffuse in society, and introduced the notion of S-curve, an S-shaped curve that shows how usage of an innovation evolves in a population over time. To determine the shape of an S-Curve, we need a minimum of three points:
- the number of adopters in the first commercial year, or a later year, for instance the time it takes to reach the critical mass of 10% of the saturation level (the ‘early mass market’)
- the long-term saturation level
- a third point describing how fast the saturation level will be reached, for instance the 50% of saturation point (end of ‘early majority’), or the 90% point (end of ‘late majority’).
We present the five most commonly used diffusion models. All these models assume that the diffusion rate of the innovation is proportional to the remaining number of adopters to reach saturation. The models predominantly differ in the assumptions they make about the coefficient of diffusion. We start with the most widely used model, the logistic curve.