Breakthroughs by their very nature don't come along very often, but Fader-Hardie probability models clearly qualify. Their general objective is forecasting, but Peter Fader and Bruce Hardie have applied their ideas to a continually expanding range of business applications, including:
- Identifying underlying customer segments
- Calculating customer lifetime value (CLV)
- Forecasting contract renewal rates
- Forecasting trial and repeat sales of new products
- Understanding the drivers behind the frequency of visiting a web site
- Determining the optimal size for a mailing based on a test mailing
- Generating effectiveness measures for a media exposure
Their breakthrough came in developing a way to simultaneously estimate individual processes (like buying processes) and their distribution through a population. The resulting tool forecasts at both the aggregate and, in most cases, at the individual level, and does so with remarkable accuracy. Model setup and estimation can be done in Excel. With the Forecast Wizard's help, the process becomes quite easy.
The forecasts, using real-world data and evaluated using holdout periods, prove robust as the following chart suggests.In this analysis, a time series of unit sales was split in half, and the model estimated using the first half and fitted to the second half of the data. As the chart indicates, the forecast tracks the data well.
Source data for this example come from panelists in a study of purchase behavior.
Scanners captured purchases over 24 weeks. A model built on that data was used to project sales for the next 28 weeks
While not all models take this form, they are all mixture models characterized by a process that describes behavior at the individual level and a mixing distribution that shows how parameters are distributed throughout the population. The sheer breadth of applications is remakable, and we will continue to add more in the future.
For additional screen shots, visit the Forecast Wizard product page.