Sales Models for Many Items Using Attribute Data

Sales models are mainly used to analyze markets with afairly small number of items, obtained after aggregating to thebrand level. In practice one may require analyses at a moredisaggregate level. For example, brand managers may be interestedin a comparison across product attributes. For such an analysisthe number of relevant items in the product category make commonlyused sales models difficult to use as they would contain too manyparameters.In this paper we propose a new model, which allows for theanalysis of a market with many items while using only a moderatenumber of easily interpretable parameters. This is achieved bywriting the sales model as a Hierarchical Bayes model. In this waywe relate the marketing-mix effectiveness to item characteristicssuch as brand, package size, package type and shelf position. Inthis specification we do not have to impose restrictions on thecompetitive structure, as all items are allowed to have differentown and cross elasticities. The parameters in the model areestimated using Markov Chain Monte Carlo techniques.As a by-product this model allows to make predictions of sales levels and marketing-mix effectiveness of new to introduce itemsor of attribute changes. For example, one can assess the impact of changing the packaging from plastic to glass, on sales and price elasticity. Besides entering and changing products, our model also allows for items to leave the market.We consider the representation, specification and estimation ofthe model. We apply the model to a ketchup scanner data set with 23 items at the chain level. Our results indicate that the modelfits the sales of most items very well