Optimize your optimization

Editor’s note: Jeffrey M. Kirk is managing partner at King Brown Partners Inc., a Sausalito, Calif., research firm. Susan H. Sayre is senior consultant, stakeholder management at TNS, an Atlanta research firm.

Today’s American supermarkets, mass merchandisers, convenience stores and club stores offer more variety in packaged goods to consumers than ever before. Consumers are faced with a myriad of choices between brands and product lines, and also with a myriad of fragrance, flavor and color options within those lines. The relatively modern phenomenon of fast-to-market, continuous innovation has created a marketplace deluged with both first-to-market and fast-follower copycat products that do everything from automatically cleaning your shower to keeping your house smelling like a grove of mangoes.

Operating in this SKU-proliferated environment, in which shelf space is not at quite the premium it once was (because there’s simply more of it in more places), calls into question the appropriateness of traditional product-line optimization research, which generally takes the approach of cannibalization analysis. With the number of variants available for most packaged goods product lines, substantial cannibalization can only be expected. In fact, we posit that for some categories, especially food categories, cannibalization is not only expected, it is desirable.

Demand choices

Consumers have come not only to expect but to demand choices in the products they purchase. Even within a consumer’s chosen brand set, they want options for flavors, colors, fragrances, etc. This expectation is especially important for products with a sensory component, the largest category of which is food. Within a category, even those consumers who have an overall favorite flavor, color or fragrance will likely reach a point when they need or desire alternatives. For example, they may paint the bedroom and demand a different color for their box of facial tissues, or they may tire of their strawberry yogurt for breakfast and opt for some raspberry for a change of pace. Consider what would happen if there were only one variety (the initial favorite) available in each brand: the consumer would look to another brand for the product variety they seek.

The ice cream category provides a perfect example of how offering variety - not minimizing cannibalization - should be the strategy of choice in optimizing a flavor line. Consider the three basic flavors: chocolate, vanilla and strawberry. It is safe to say that with these three flavors alone, virtually all consumers who like ice cream are reached. A cannibalization analysis would inform us that we don’t need other flavors because any new flavors would overlap in appeal (i.e., cannibalize) the core three. However, consider what would happen if Dreyer’s offered only chocolate, vanilla and strawberry: consumers would shop the case and buy alternate brand(s) that offered variety in their flavor lines. Other categories that exemplify the desirability of variety are canned soup, frozen dinners, ready-to-eat cereal, and disposable air fresheners.

Another primary problem with cannibalization analysis in consumer packaged goods (CPG) research is that it fails to take into account a very important purchase dynamic: buying multiples per purchase occasion. The underlying, unstated assumption of cannibalization analysis is that consumers will purchase one product. For many categories (ironically, the categories for which cannibalization analysis is most often conducted), consumers buy multiple “cannibalizing” varieties per occasion. A consumer will often buy two, five, 10 or more individual units across several different varieties in a single purchase occasion, and even more over time. If there isn’t enough variety within one brand, the consumer moves to another brand in his/her product set.

Relied on TURF

The principal, and virtually only, analytic technique on which CPG manufacturers have relied for product line optimization is TURF (total unduplicated reach and frequency). It is often requested as part of the standard deliverable from a concept test. In short, the analysis is used to suggest a line of variants that have minimal consumer overlap with regard to appeal. For all the reasons already mentioned regarding the concept of cannibalization in general, this analysis fails to deliver optimal, actionable insights for most product categories. Understanding the nuts and bolts of the analysis will elucidate its limitations in most situations.

TURF was borrowed for marketing research from the world of advertising. It was developed to measure the breadth and depth of an ad’s household penetration. In that context, reach refers to the percentage of households who have been exposed to a particular advertisement at least once; frequency refers to the average number of times each reached household is exposed to the ad.

Consider the notion of extending this analysis to a flavor line optimization (we will use flavor generically, to indicate product varieties such as color, flavor or fragrance). In this context, reach represents the percentage of consumers who show interest in at least one of the potential varieties of the product being tested; frequency refers to the average number of flavors in any particular line that are of interest among those consumers who are reached by the line. The objective of the line optimization analysis    under the TURF paradigm is to determine what varieties should be included in the product line to maximize its reach. While frequency might help us to understand overlapping appeal of products within the line, the theory driving TURF analytics necessitates minimizing this metric in order to maximize reach.

TURF is essentially a determination of net interest. The analysis usually begins with an investigation into the number of individual flavors that should be introduced to reach a majority of the consumers interested in the product line. For some manufacturers, the number of flavors to introduce is an easy call as they only have the equipment or the budget to produce a limited number of flavors. For others, the number of flavor varieties is open to discussion, in which case it is incumbent upon the research to provide a recommendation for both the optimal number of flavors for the line as well as what those flavors should be.

In executing a TURF analysis, what the computer program does is evaluate the net appeal (reach) of every possible combination for each feasible size of product line and return a list of the “best” lines for each number of products possible. The best lines are defined as those that reach the greatest number of people in total with the least amount of overlap among themselves. Using this model to determine the number of flavors to include in the line presents its own set of problems for, as mentioned, a very few basic flavors (e.g., vanilla, chocolate and strawberry) will generally reach most interested consumers.

For the purpose of argument, let’s say that these three flavors capture 95 percent of interested consumers. Adding an additional flavor might get us to 97 percent, another to 97.5 percent, and another to 97.8 percent. Is it realistic to expect to reach 100 percent of interested consumers? And how much sales volume can we actually expect from reaching an additional 0.5 percent or 0.3 percent of the interested consumers who are participating in a market research study in which the sample may be only 100 or 200 people?

Additionally, when most of the interested population is reached by just a few flavors, yet there is a wide variety of flavors proposed, it becomes very difficult to determine which single flavor to add to increase reach. Most of the time, several flavors present themselves as equally appealing, so we are again reduced to making the decision arbitrarily as opposed to mathematically. Furthermore, basing business decisions on increases in reach of tenths of a percentage point is not a sound practice, as these minute differences in reach are not statistically significant, anyway! Yet manufacturers do it all the time.

Even though chocolate, vanilla and strawberry ice cream covers the bases with respect to reach, the critical component of volume is missing. While in theory, a TURF analysis might recommend a carrot- or beet-flavored ice cream to capture the very small percentage of consumers not reached by chocolate, vanilla or strawberry, would the manufacturer not actually be better off with a “Double Chocolate Explosion” flavor to increase penetration among their chocolate, vanilla and strawberry buyers? While TURF would say no, we would say yes.

Sales volume is more important

At the end of the day, sales volume is much more important than reach. So we need to employ metrics that measure the absolute appeal and volume impact of a flavor vis-à-vis others in a set, not the overlap of its overall appeal with other products in the line. Our recommendation is a simple but robust one, and is consistent with metrics used in concept volume forecasting: the chip allocation. Its execution is simple and its results are far more intuitive and impactful than TURF.

To determine volume impact, consumers are asked in the concept test not only how often they intend to purchase the product but also how many units they would expect to buy at each purchase occasion. From these questions we are able to derive a total number of annual units for every consumer in the study. Consumers are also asked to allocate their next 10 purchases of the product among all possible flavors in the line (or category, as applicable). The chip allocation is used to proportionally allocate each consumer’s annual unit volume across all flavors being evaluated. When summed across all consumers, a share of potential volume can be calculated for each flavor variant.

A key benefit of this approach is that shares are much more intuitive to interpret than TURF results. Beyond that, they convey powerful information about the absolute and relative volume impact of each flavor variety. Because we have taken into account every consumer’s intended purchase volume, the shares that result incorporate the proportional volume impact as well. Furthermore, cannibalization is naturally reflected in the results as those flavors selected were chosen in the context of all possible flavors available. Finally, a share of volume analysis can also suggest the number of SKUs that will be needed to capture the majority of category volume, as diminishing returns result from adding more and more SKUs to the line.

A rethinking

The advent of big-box retailers with acreage to fill, coupled with a proliferation of channel options where consumers can fulfill their shopping needs, has necessitated a rethinking of CPG firms’ traditional approach to line optimization. The belief that it is sufficient only to reach each consumer with a single product in a product line fails to take into account the variety-seeking behavior of our culture and the sky’s-the-limit possibilities that new distribution channels afford. Volume is a key component that must be taken into account to make informed, insightful choices when defining a multi-variety product line. By borrowing a line out of the volume forecasting playbook, we will take a quantum leap forward in actually optimizing our product line optimization research.