TURF optimization

TURF (Total Unduplicated Reach and Frequency) is an optimization method that identifies a subset of items, called a portfolio, that maximizes the number of respondents reached.

Originally developed for broader survey research, TURF can also be applied to MaxDiff studies. Rather than selecting the most preferred items on average, TURF identifies combinations that most effectively motivate, satisfy, or appeal to the largest number of respondents.

For example, when deciding which ice cream flavors to stock, TURF can help maximize customer appeal given limited shelf space. Selecting only the most preferred flavors may overlook niche options that appeal to distinct segments, reducing overall reach. If most respondents enjoy vanilla, chocolate, and cookie dough but many are satisfied with any one of these, substituting one with a less common flavor like strawberry or mint chocolate chip could expand reach and boost sales.

For MaxDiff data, we typically recommend Weighted by probability. It makes effective use of potentially sparse data and is more stable — minor changes such as adding a few respondents are less likely to disrupt the rank order of optimal portfolios.

First choice makes less efficient use of detailed preference data and is more sensitive to small data changes, which can significantly shift the rank order of outcomes.

Threshold falls between the two in terms of efficiency, but determining the ideal threshold value for MaxDiff data can be challenging.

Since many TURF solutions deliver similar reach levels, additional criteria such as expert judgment can help select the most suitable portfolio. For instance, if a grocer knows that a frequently appearing flavor spoils quickly, they might prefer an alternative portfolio with comparable reach that avoids that item.

This setting is only available when relevant items MaxDiff is used. When enabled, a missing item provides essentially zero reach for any respondent who did not see it in their MaxDiff questions.

Note: When enabled, a utility value of −30 is assumed for missing items, which effectively zeroes the reach contribution for those respondents.