Propensity models start with known information about tens of thousands to millions or households or people. This raw data may be sourced from credit card company transaction data, automobile dealer sales, retail sales, surveys or other sources. Based on deep demographic knowledge of essentially every U.S. household, regression analyses are performed to determine which known demographic and other characteristics are key factors in predicting a household’s propensities and a model is created. This model is then applied to the tens of millions of households for which raw data is not available in the form of a score, typically 1 to 10, 1 to 20 or 1 to 100. Then this model is back tested against the raw data to validate its predictive value.
The propensity score can now be applied as a data enhancement element to existing name and address data to be used by a company’s marketers and data scientists, or it can be used to select a target audience for postal or email lists.