Imagine that you are looking for a new car or SUV. Because you live 40 minutes away and encounter heavy traffic and winter snow on your way to work, you want high gas mileage and all-wheel drive. If a car or SUV does not have these features, you will not even consider it. This simple rule eliminates over 95% of the alternatives from even being considered. Getting into a customer’s consideration set is most of the battle. But what does it take to be considered?
Applied Marketing Science (AMS) is developing a new choice modeling technique that determines how to “make the cut” – i.e., it identifies the features a product must have to get into consumers’ consideration sets, and the trade-offs consumers make between features of considered products. By explicitly measuring and modeling the formation of consideration sets prior to choice, the new model does a better job of predicting actual customer behavior than traditional conjoint/choice models.
Consumers generally develop their consideration set by reviewing a list of features and finding all the products with their “must have” features and without features that are grounds for elimination, e.g., an unacceptable brand or an unaffordable price. This is a non-compensatory process, because no amount of good performance on other features can make up for the lack of a “must have” feature. Traditional choice and conjoint models, on the other hand, assume a compensatory process, in which a product can be chosen even if it does poorly on some features, as long as it does well on others. This generally works for trade-offs within the consideration set, but not for the formation of the consideration set itself.
With our process, respondents are presented with an array of product alternatives generated from the full set of attributes for the product category. Respondents select the alternatives they would seriously consider buying. A quick mathematical test is done to check whether each respondent used a non-compensatory or compensatory method for deriving his or her consideration set. In initial testing, it appears that only a small minority will use a compensatory method, and for those respondents, we perform a choice-based conjoint (CBC) analysis using the full set of attributes and levels. For the vast majority of respondents, who use a non-compensatory method to form their consideration set, we identify the reduced set of attributes and levels remaining in each respondent’s consideration set, and perform a choice- based conjoint analysis based on alternatives generated from this often greatly reduced set.
Our analysis is based on two methods developed and tested by researchers at MIT. The first method uses “Greedoids” to identify the respondent’s consideration process. The second method uses the FastPace CBC model to adapt questions to individual respondents.
This new tool allows us to closely replicate the actual purchasing process. It also simplifies the survey for respondents by reducing the number of attributes and levels they must deal with during the trade-off exercise. What results is a model that fits and predicts better than current conjoint models.
Better, faster, cheaper – isn’t that just what market researchers have been demanding all these years? This is one rare example where you don’t have to pick any two.
image credit: tuvie.com
Steve Gaskin has conducted and analyzed quantitative primary and secondary research for the consumer packaged goods, automotive, high technology, telecommunications, pharmaceutical and financial services industries for over 20 years. As president of The Delphi Group, his projects included worldwide new vehicle sales forecasting for the Ford Motor Company, litigation research for major corporations, and teaching training courses to consultants on statistics and multivariate analysis techniques.