How often do you hear professionals making casual reference to ‘the customer’, as if customers constitute a singular, amorphous whole? They don’t, of course, and whilst few brand and product professionals would openly admit to harbouring such naïve views, routine and unquestioning averaging of customer-generated data would suggest otherwise!
Averaging customer-generated data cannot be justified unless the data is demonstrably homogeneous (i.e. other than a small proportion of outliers, all the customers involved in the study are expressing essentially the same sentiment in their data, albeit to different extents). Violation of this fundamental homogeneity principal is rife across the research industry. As a consequence, mean data that seems to be statistically robust, can be very misleading, with all that this implies for the quality of subsequent decision-making. It was this realisation that persuaded me to found MMR Research Worldwide (sister company to HRW) in 1989, with the primary intention of developing innovative ways of deconstructing large heterogeneous, customer-generated data sets into smaller, homogeneous subsets, which are then analysed separately in order to capture the full diversity of customers’ opinions. This strategy is known as segmentation.
Customer-generated data can be segmented in two ways:
1) Extrinsic segmentation superimposes supplementary secondary (external) information, such as gender, age, socioeconomic status, etc., to divide the primary data into subsets. Summary metrics (e.g. mean ratings) are calculated within each subset and statistical tools determine whether or not any differences across the subsets (e.g. male v female, 18-25 v 26-35 years, etc.) are likely to be significantly different. Whilst extrinsic segmentation can sometimes prove useful for developing targeting strategies for branding and advertising, it’s still entirely possible that the data within each of these subsets (e.g. male v female) is heterogeneous and therefore does not represent a holistic male or female point of view. This being the case, it would be entirely wrong to conclude that males and females (for example) hold opinions about the product that differ in a meaningful way. In fact, statistically significant gender differences could be statistical artefacts that serve only to misdirect marketing decisions. Extrinsic segmentation has been, and still is, widely used in marketing and market research. Whether it has helped, or hindered product design and targeting is definitely a moot point
2) Intrinsic segmentation explores within the primary data set to identify and group people who share common opinions in the context of the question(s) asked. Typically, this would include attitudinal data where customers indicate the extent to which they agree (or disagree) with the sentiment expressed within each of a number of attitudinal statements. Each of the attitudinal clusters (or segments) subsequently identified in the data by the intrinsic segmentation algorithm, is qualitatively homogeneous in terms of the attitudes expressed. This being the case, it’s perfectly legitimate to calculate summary statistics (e.g. mean and standard deviation) within each of the segments and thereafter to compare these metrics across segments. Because the within-segment data is homogeneous, inter-segment differences are meaningful and often prove to be important influencers of customers’ behaviours.
MMR Research Worldwide pioneered the application of intrinsic segmentation to the sensory optimisation of food, beverage and personal care products. By way of example, this would typically involve a set of 6 – 12 comparable products, including prototypes, from within a fairly narrow product category (e.g. milk chocolate), all of which are tasted and rated for liking by a relatively large number of category or target users. The intrinsic segmentation algorithm then searches the data to identify people who share a common view regarding the products within the set that they like most and the products that they like least. This process, known as ‘Liking Segmentation’, typically yields between three and five homogeneous ‘Liking Segments’, all of which are highly differentiated from each other (by definition).
The sensory characteristics of the target products are then optimised either to maximise liking within a particular segment (targeted optimisation) or across several segments when broader appeal is required. Since the inception of ‘Liking Segmentation’ in 1996, MMR has conducted several thousand studies and this process has been used to optimise many of the World’s leading brands. It succeeds where most other optimisation strategies fail because it’s based on the principles of intrinsic segmentation. Sensory optimisation of products based on unsegmented customer data rarely succeeds because customers invariably hold highly contradictory opinions about what they like and dislike. With market leading brands, averaging across such diverse opinions is utterly illogical and likely to result in a supposedly ‘optimised’ product that’s unlikely to delight. In commercial terms, this leaves the door open for entrepreneurial challenger brands, retailer brands and even discounter brands to delight subsets of the target population, if only by luck rather than by design, thereby eroding the brand leader’s market share.
Intrinsic segmentation is one of the pillars upon which MMR Research Worldwide’s spectacular success is based. The principle and the practice of intrinsic segmentation has been adopted by HRW since its inception in 2001, where it has proved to be similarly successful in identifying attitudinal types within target populations and also in optimising OTC products.
What’s the moral of the story?
Don’t be misled by ‘meaningless means’.
Build intrinsic segmentation into your research design whenever it is practical to do so.
By Professor David Thomson