An Approach for Determining Optimal Product Sampling for the Diffusion of a New Product

Free samples are an effective means for introducing and promoting a new product. However, product sampling is also expensive. As a result, careful consideration must be given to the question of how many samples should be distributed. To encourage product adoption in any target market, a company needs to determine the “right” amount of sampling. In other words, a firm needs to determine the optimal number of samples that must be available for trial by the innovators, early adopters, and other key consumers who influence the adoption rate of the new product. With too few samples, the product might not reach enough of these key consumers to generate the word-of-mouth recommendations necessary for market success. On the other hand, offering too many free samples is a waste of a company's resources. Dipak Jain, Vijay Mahajan, and Eitan Muller propose a framework for determining the optimal levels of product sampling. In addition to identifying the upper bounds for the sampling levels of both durable and nondurable products, their model identifies the optimal size of product sampling based on such parameters as the coefficients of innovation and imitation, market potential, discount rate, and gross margin. Several observations are made regarding the relationships between the optimal sampling level and the various parameters used in the model. For example, a high sampling level is not appropriate for a product with a high coefficient of innovation. On the other hand, if a product has a high coefficient of imitation, the sampling level should be high because a significant number of trials are necessary for word of mouth to be effective. High sampling levels are also indicated by a high discount rate or gross margin. For durable goods, the optimal level of neutral sampling (i.e., sampling that does not specifically target innovators and early adopters) rarely exceeds 5%, and the maximum level is 7%. The optimal target sampling level is always higher than the corresponding neutral case, but, in most cases, only marginally so. For the parameter ranges chosen in this article, the maximum level for target sampling is approximately 9%. However, it is important to note that the theoretical upper bounds are no more than benchmarks for the maximum possible level of sampling. In practical situations, the optimal level may be considerably lower than these upper bounds. In such cases, the actual values will depend on the values for the various parameters used in the model.