Getting to the Root of Innovation Part 2 of 2

Constructing the Terrain

To bring the various issues, thoughts, and content elements together to understand how this brand landscape approach to segmentation works, we need to address the analytic components. The starting point is the creation of the brand landscape or the terrain, which is derived as a result of a similarity-sorting task. There are several ways that this can be achieved, depending on whether the research is being conducted online or in person. Typically, consumers are able to sort 100-150 brands or SKUs, usually in the form of pack photographs, based in some way on their similarity of use, and their difference from the products in the other piles. Once this task is completed, new product ideas ranging from close in to further out can be sorted similarly, including provisions for new piles to be created. Essentially, the aim is to establish consumer perceptual distance from one brand to another.

Via the use of proprietary multi-dimensional scaling (MDS), structure maps are generated to show the perceptual brand landscape: how and why the SKUs cluster together (this can be based on usage, occasion, ingredient, pack type, form, brand etc., often in combination). The clusters are defined mathematically, but are interpreted by observation and/or a supplementary description exercise undertaken by respondents. You may not necessarily think of MDS in the context of segmentation: cognitive or perceptual mapping has been available to market researchers for many years, and is probably both underutilized and misinterpreted in many instances. However, it is a very effective method of examining how consumers perceive and group the brand landscape.

There are several types of perceptual mapping correspondence, discriminant analysis (sometimes known as DFA), and multi-dimensional scaling. They achieve similar outcomes, but from different perspectives. At first glance, the resulting outputs look similar, but they are very different. Both correspondence and discriminate maps utilize "internal analysis," in that data is gathered via a brand/attribute association matrix based on preference judgments. Using predefined attributes to achieve brand ratings constrains the outcome. Correspondence analysis is more often applied to brand image mapping, and indeed, at Ipsos we use it for this purpose in conjunction with our segmentation work. It is inherently the easiest approach to comprehend quickly, as points represent positions of both brands and descriptive attributes. However, we also maintain that while in many respects it is the simplest form of mapping to understand, it is open to misinterpretation, particularly when the map is built on all the attributes, regardless of whether they are significant or not. Those that have a significant positive association with a brand add much more power to the interpretation.

The other aspect is that any segmentation or clustering of brand association groups on correspondence maps seems to be done on a qualitative judgment basis, with attributes being circled around a brand to indicate the brand's "territory." There is, however, a more scientific approach, which is to indicate only the items that receive significantly stronger associations with a brand and which brand or brands own or share each attribute, as represented by shapes or colors of the attributes and brands. The relative importance of each attribute can be indicated by creating appropriately sized bubbles.

Discriminant analysis is similarly constrained by attributes, but with emphasis on the attributes that best discriminate among brands as opposed to attributes that belong to individual brands. The resulting map displays vectors representing weighted linear combinations of all attributes that best distinguish among the brands.

As map interpretation is reliant on dropping a perpendicular line from the vector to a brand in order to establish its relevance, it is more difficult for a lay audience to comprehend, and does not permit the identification of brand segments. Like correspondence analysis, it is not easy, at least visually, for discriminant analysis to handle 100+ brands.

Multi-dimensional scaling, on the other hand, offers much greater opportunity to understand how consumer perception plays out by mapping the total brand terrain without any attributes conditioning the way in which the landscape is formed. To stay with a mapping analogy, it is akin to taking one of those city-to-city mileage grids you find at the back of a road atlas. If you take the data from the grid a run it through MDS, you will essentially get a map of the U.S. If you were to take consumer "perception" of those distances, rather than the real miles, the resulting map would probably show the US stretched more East to West as distances between East and West Coast cities are likely to be perceived as much further apart than they actually are, while North-South cities are much closer. Furthermore, the cities can be clustered on their adjacency or dissimilarity to each, and you have a perception-based segmentation.

There is clearly a lot more to the analysis and, particularly, the interpretation than this. At Ipsos, we have re-engineered the process and built in proprietary components. The approach works best when there are many data points (brands or SKUs), with an absolute minimum in the range of seven to 12. While this is usually not a problem at the brand level, it can be if you try to map just categories. With a sorting approach to data collection, respondents can usually cope with over 100 objects (less if online); however, with the alternative of pair-wise ratings, the number of ratings required can quickly escalate. Just 20 objects means 190 pairs, and so partial designs become a necessity and the task becomes tedious. The actual map generated via MDS looks like this:

With the absence of attributes, there is clearly an element of subjectivity in identifying the key landscape dimensions and in defining the brand clusters. Generally, two dimensions account for the vast majority of data variance. Some markets display distinct horizontal axis dominance, creating a "letterbox" that indicates a very linear market. Occasionally, a third dimension might be found, but experience suggests that it is possibly more related to new products and perhaps even connected with white space.

The dimensions describe a series of clusters of brands or SKUs, their irregular shapes defined by products at their edge. Clusters can be formed for a variety of reasons--brand, type, ingredient, occasion, user, price, benefit, etc.--as shown in this stereotypical example of the toothpaste market.

Depending on the market, the number and combination of clusters in which consumers are active varies. Three to eight is a typical range, but in more fragmented markets sometimes 10 to 12 is average. Interestingly, the dimensions seen in the overall landscape dimensions are generally also seen within the clusters themselves. For example, if the horizontal dimension is adult-to-kid from left to right, then the brand or SKU array within the cluster (assuming in has reasonable horizontal extension) will also be from adult on the left to kid on the right.

Observation is the key to identifying and naming the clusters. Often, arraying the visual images of the components can assist. For example, a study of the cookie market in the Philippines revealed a cluster comprising large tins of wafers, a grouping seemingly inexplicable in the context of the rest of the landscape, especially as small packets of wafers were in a cluster much further away. A Philippine-born American researcher was, however, able to identify that this was a gift cluster, consisting of products that city dwellers would take to their families in rural areas when visiting. Beyond observation, there is a vast amount of "surround data" from the structure study itself that can be utilized to further aid interpretation of the terrain, particularly in terms of examining the profiles and needs of consumers who "play" in particular parts of the landscape and brand image/equities, especially in the context of key drivers.

An examination of the combinations of clusters from which consumers use brands and SKUs adds to learning about market dynamics. Visually they can be displayed by what we term the "super Venn."

In this example, 60% of all combinations were represented by these key patterns, leaving the remaining 40% of combinations representing over 200 patterns--almost one per respondent. As the diagram shows, what is interesting are the two distinct groups of landscape clusters. This type of analysis can also be extended via cross-tabulation of the frequency of cluster use by type of cluster.

In this example--with the exception of the cluster on the far right, which comprised SKUs aimed at kids--it became clear that consumers enter this market in clusters with relatively simple products and then step through, moving progressively into more sophisticated or complex clusters as they gain experience. The learning here was important in understanding who might adopt new higher-end products and from what part of the brandscape they would come. Another often-revealing analysis that can be undertaken is to collect satisfaction measures of the existing brands or SKUs in the market and then simply chart this in terms of the derived landscape clusters as shown here. Each major brand's average rating is shown by the colored symbols.

This can be taken a step further by breaking out the individual SKUs and plotting satisfaction against the number of users.

Cluster membership is color-coded here. Opportunities exist in the upper right where satisfaction is high but usage is low. Incidentally, in this example, some of the major brands were in the mid to upper left: high usage but low satisfaction.

Traditional segmentation analysis can also be applied to the landscape clusters. Here is an example of cross-tabulation based on consumer segments and indexed. A question that is often posed in relation to subgroups, and say attitudinal or needs-based segments, is the degree to which the landscape maps differ among groups: they do, but not substantially so. Fundamentally, you are unlikely to see a radically different landscape from one group or segment to the next.

This is because the base landscape is an aggregate of consumers who are usually representative of the market; consumers hold an approximately similar view of the total brand landscape. That doesn't mean that subgroup or segment maps are the same; as can be seen here, the clusters may be somewhat differently configured and hence have different shapes.

To return more specifically to the subject of this paper, where's the white space? Firstly, unlike the brand/SKU clusters, white space is not precisely ring-fenced, nor is it is clearly identified. It is a judgment call, but one that can be made via applied common sense. An obvious, large gap in the brand landscape structure is probably indicative of white space. If that space is within the structure, it is probably going to be more easily defined and assessed simply because the clues come from what is around it and who, in terms of consumer profile, uses these surrounding clusters. As stated earlier, white space on the periphery tends to be more difficult to identify, simply because it may very well be affected by categories, products, or brands outside the scope of the survey.

While it may be possible to identify and define a white space, the question remains as to whether the space is viable. This can be established, at least on a preliminary basis, utilizing the inclusion of a series of new product ideas into the sorting task. As these ideas have to be predefined, they need to be wide ranging. Remember this is not a concept test, merely a way to look at close in to far-out product opportunities, potentially under different branding. For example: a new flavor or whitener with breath-freshening might be a close-in idea, but a small capsule placed in the mouth at night that results in whiter teeth and fresh breath by morning would be a far-out. In addition to current branding, it can be useful to present some of the ideas as competitive brands and as new brands. Analytically, the MDS mapping process allows us to overlay new ideas on the base landscape map. Some will fall into existing clusters, others close by and some, hopefully into white space.

A word of caution: it is extremely unlikely that 15-20 new products would be launched into a market, certainly not simultaneously. At best, it would be one or two at any one time. Therefore, if all the new product ideas are permitted to "actively" impact the map, the landscape would be artificially changed; hence there is a need for them to be overlaid "passively" on the existing base landscape. You might ask, is it not possible to select just one or two ideas and test how they would positively affect the terrain? Here again there would be a distortion, because the ideas were included in each individual respondent's sorting task and clearly had an impact on the way that consumers undertook the exercise.

Nevertheless, displaying all of the new ideas does show where any one individual new idea would likely fall. Additionally, as seen here, new clusters may emerge, demonstrating that allied ideas might converge directly to fill a white space, further demonstrating its viability. A new cluster may overlap an existing one, indicating that there is room for further expansion. Even though a new product idea may indicate an opportunity within the current landscape, there remains the question of establishing its nature and size. The first step is to examine the purchase interest in the ideas. We achieve this via more traditional PI measures and then apply TURF coupled with Shapely Value analysis to determine not only the optimal number of ideas but also the best combination. In this example, there are two principal areas of opportunity: a clearly defined new cluster--essentially, a bridge cluster comprising four products each with reasonable interest levels--as well as a significant extension to an existing cluster.

Here is another example of a newly emergent bridging cluster, which interestingly comprises similar ideas but in different brand guises. Once the level of interest has been established, the data can be examined in terms of which subgroups (in terms of both socio-demographics and behavior, as well as attitude or needs segments) have the highest interest. It is premature to attempt any rigorous sizing, let alone volumetrics, as this is a strategic positioning exercise. However, with the knowledge of who is interested coupled with the nature and extent of the terrain that is potentially being covered, one can make broad approximations as to the relative opportunity.

The MDS mapping approach behind the landscape analysis is rooted in measures of adjacency; thus, each point has a mathematical relationship to every other point. This can be used to determine the degree to which individual brands or SKUs are close to each other. Where this occurs for items from a common manufacturer, this can be used as a first step to portfolio rationalization in crowded clusters. If we return to our earlier U.S. map analogy, it is either Minneapolis or St Paul!

To some extent, we have challenged the notion that segmentation should concentrate only consumers. However, by looking at a market through the eyes of the consumer--how they see the products, brands, and individual SKUs available to them--we have the opportunity to step out of what we might term the "trade classification" of how the marketer and the retailer traditionally look at the product lineup.

By creating a brand landscape, there is the opportunity not only to gain new perspectives of how the market is organized in the consumer's eyes but also to see where white space opportunities exist and how products might be evolved to capitalize on unmet needs. In turn, this can contribute to innovation platform development. The approach is both art and science: it draws on a different application of perceptual mapping, which certainly requires some objective judgment in interpretation, and building market landscapes with market structure studies and their extensive data sets facilitates deep drill-downs, including the integration of more traditional multivariate segmentation tools. The principal aim is to provide the marketer with an actionable roadmap comprising both a thorough view of the current market and a vision of the opportunities that lie ahead for strategic development, innovation, and brand planning.