Discrete choice modeling is widely used for estimating the effects of changes in attributes on a given product’s likely market share. However, there are many related effects that discrete choice models alone are less well-suited to address. One of these is the extent to which having more than a single product in a portfolio of offerings can be used to increase the market share for that portfolio. In a market with heterogeneous preferences, the market share will be largest for portfolios that have items that appeal to each of the consumers in that market. However, design and production costs increase with the number of items in the product portfolio and, in most retail environments, producers are limited as to the number of items (stock keeping units or “SKUs”) that they can display.

TURF (Total Unduplicated Reach and Frequency) has been widely used for analogous problems in media placement planning, and, more recently, for consumer product portfolio optimization. In the media placement application, survey data are used to determine consumers’ media consumption patterns and those consumption patterns are in turn used in a sample enumeration form to mechanically calculate reach (number of individuals who are exposed to a given set of media) and frequency (the number of such exposures). The typical product portfolio application uses survey data in an analogous way, asking consumers which of several products they would consider for purchase and which they would not. Again, the survey data can be used in a sample enumeration process to estimate the reach and frequency of a given product portfolio.

This paper describes two applications that use discrete choice methods to provide a more robust metric for use in TURF applications. Both involve products for which there is a high degree of heterogeneity in preferences among consumers: apparel and food products. The apparel application uses maximum difference scaling conjoint to provide the TURF metric while the food product application uses conventional discrete choice methods. Both use individual-level posteriors to reflect the sample heterogeneity.

One of the significant challenges in using TURF is that the basic calculations have a computation dimension that is proportional to n choose m where n is the set of all possible products and m is the allowable number in the portfolio. With a multi-attributed product, n can be quite large; in the food application it is in the range of 57,000 and any portfolio greater than two items becomes computationally intractable. To support these applications, a heuristic method was developed to efficiently sample from the space to create close-to-optimal portfolios.

These approaches have been implemented as part of the product planning process in the client organizations and, however, there are several areas in which alternative methods might be developed to improve this process. The paper concludes with a summary of the challenges in these applications which to date have not been addressed.