The Art of Noise
Method’s Brand Context Research Technique
Our Brand Context quantitative research methodology was developed as the result of acknowledging two harsh realities: clients have more data about their own businesses than their agencies ever will, and target audience research respondents don't really understand why they do what they do. So much consumer research gets done while conveniently ignoring these facts, and that's what we were trying to fix.
So what is a well-meaning creative agency to do when it wants to provide new insights into their clients' businesses and brands, insights that can drive better understanding of their customers, and in turn better experience design? Utilize a research methodology that seeks to identify perceptual patterns that consumers themselves can't articulate, while providing the out-of-category context clients have no obvious reason for seeking via their own performance-oriented metrics. As my 10x10 article explains, one way of doing this is to deliberately inject what would normally be considered noise into a quantitative survey.
With the Brand Context approach, we do this by asking respondents for very quick, gut reaction ratings of several hundred brands at once, both within, but mostly outside, the client's industry category. The goal is not to ask respondents directly why they like a particular brand, only that they do or do not like it. The key is in realizing that strong brands themselves already have meaning; instead of trying to understand the meaning of brands, we flip the equation backwards and use the meaning of brands to understand how the target audience thinks.
A single click yields both awareness and an appeal rating at once.
This understanding is generated by examining the very complex patterns generated by the repetition of very simple metrics across a large-enough quantitative sample. In the case of the Brand Context research, we present a long list of 200 or more brand names, each shown in plain helvetica text, with no logos, proprietary colors, or other trade dress. This limits the respondents' reaction to their mental perception of the brand, not the other elements of a brand's presentation. In addition, the order of the brand list is randomized for each respondent, so that the brands are not arranged in their proper industry categories. That means Nike might follow Mercedes-Benz, Tampax, Banana Republic, and Burger King. Creating that kind of chaos deliberately might sound silly, but that's how brands tend to be encountered in the real world.
The survey instructions are simple, but not typical. Instead of asking respondents to think carefully about why they like brand A instead of brand B, we tell them run through the list of brands as quickly as they can, skipping those brands they have never heard of, and rating the rest on appeal using a 7-point scale. This approach is highly efficient, because in an instant a single click yields both awareness and an appeal rating at once, and the lack of a click, where a respondent skips a brand by being unaware of it, is itself a data point.
More importantly, collect 200 of these instants, and a very rich pattern of data is created by the sheer scale of the potential variance across that many responses. Each brand rating has 8 potential states, and with each respondent considering (at least) 200 brands, 8200 creates a number of potential combinations that is too big to have a name. So, with about 10 minutes of a single respondent's time, we've generated a perceptual pattern with tremendous complexity. Combine those across several hundred respondents, and you have a very rich set of data that yields a view of target audience perceptions that's impossible to obtain via qualitative methods, yet is somehow very qualitative in nature.
Those 199 brands play the role of the essential noise in this system - the perceptual benchmarks surrounding the actual location of the client brand in the audience’s mind
The reason for that is that much like other forms of high-level perception, like our distance perception example in the 10x10 article, the value of the perceptual patterns created by this technique are in the eye of the beholder. There are no statistically significant differences or normative thresholds to calculate once the results have been calculated, only a subjective interpretation of what the patterns indicate about the target audience in question.
What we do is start with the brand we're interested in, and create three different sample populations from the total data set: those unaware of the brand, those aware of it who either don't like the brand or feel indifferently about it, and those who really like the brand. Comparing the demographics of those three groups is an obvious place to start, but that usually serves only to confirm what the client already knew about their brand.
What's really interesting is what happens when we quantitatively compare the perceptions of the other 199 brands in the survey that are not the client's brand, from the perspective of those three states of perception about the “target” brand itself (1 - ignorance of it, 2 - disdain/indifference towards it, 3 - enthusiasm for it). Even though our goal is to understand how a client's brand is perceived by various audiences, the function of the respondent's rating of that client brand is only to separate the sample into three buckets, so we can determine the differences in how all three groups perceive the other 199 brands. Those 199 brands play the role of the essential noise in this system - the perceptual benchmarks surrounding the actual location of the client brand in the audience's mind. Think of it as a kind of branding sonar; bouncing signals off of the ocean floor to see how the echoes trace out the outline of a sunken ship you're trying to find.
One of the keys to this working is that the respondent is not aware that she is self-selecting herself into one of these three samples — the quantity of the brands being rated is too high to keep track of what you're doing. That's how we avoid the pervasive tendency of research respondents to want to appear rational and self-consistent in their responses. When you have no idea of what the topic of the research really is, you don't worry about justifying your thoughts about it.
While we never really know what will happen with this technique, there are some general tendencies worth noting. First, the most powerful, well-known brands are of the least analytical value in this approach. That's because by definition, the most powerful, well-known brands (like Google, Apple, Target, Amazon, etc.) get that way by appealing to almost everyone, so the perceptual differences about them across all three sample groups tends towards zero. The same is true of generally disdained brands (e.g. cigarettes, oil companies, health insurance brands), so those cancel out too. The real action is in the middle; those brands that create differences in opinion across the three samples, differences that are generated by the only source of difference between the three samples: their perception about the client's brand.
The “action” that results is a small set of brands, say 10 to 25 of them, for each of the three samples that bubble to the surface as being disproportionately favored by that group of people when compare to the other two. This collection of brands are connected somehow; they represent a pattern of perception that's unique to the point of view of each sample group, and that's where the useful insights come from.
Over the years, using this approach has revealed such useful nuggets as the unspoken mid-life crisis behind 40 year-old men's desire for guitars, how powerful the connection between athletic apparel and jewelry is for brides-to-be, or how the appeal of the Cadillac Escalade occupies the same mental territory as that of a weekend in Las Vegas.
Creative teams relate better to being shown these kind of relationships more readily than they do to column and pie charts, no matter how pristine the data behind the charts might be. The results are still purely quantitatively driven; it's just that the inclusion of the randomized noise involved creates a pattern-based output, not a collection of discrete data points. More importantly, the pattern of brands that results allows designers to use their own unique expertise to determine what the meaning of that pattern is for them. It provides a means for them to empathize with a target audience, and then understand what that audience is ready for, not just what they might say they want. Ultimately, that's what has to happen anyway for something interesting and human to be designed.