Quantitative Analysis and Cognitive Dissonance in Marketing
Updated: Oct 30, 2018
I love business-related applications of data analysis, and so I want to share a huge shift in my understanding of data analysis frameworks in marketing. I went to the Big Data Roundup last week - moderated by Steven Phillips of Dell - and heard from a lot of interesting speakers on data analytics in their companies, a treat for someone trained in the academic side of data analysis. In particular, a short conversation with one of the speakers resulted in a large paradigm shift in my thinking about the correct questions to ask when doing market research.
While discussing personalized marketing in his original talk, David Shaw from Virally gave the example of My Little Ponies. I understood the need to segment analysis into subgroups for certain products - with My Little Ponies, it was girls 2-9 years of age and men in their forties and fifties (the so-called “Bronies”). However, I did not understand why one would choose to segment the market or personalize the campaign beyond that, once it’s clear that these are your two segments of interest.
I related the question back to chocolate - something we all feel strongly about - and which, aside from a few exceptions, has general support from the entire population. As long as you come up with a convincing campaign that expresses why your chocolate is better than your competitor’s chocolate, why waste time and money collecting and analyzing details about individuals?
Coming from a statistical background rather than a marketing background, the bias-variance trade-off was in the forefront of my mind. For the purposes of the chocolate example, your analysis can focus closer to one of two ends of a spectrum: On one end, you can make conclusions based upon the average data (i.e. what does the average chocolate buyer favor) and, on the other end, you can make conclusions based upon the individuals (i.e. what does each individual chocolate buyer favor). While the second choice sounds wonderful in principle, you must collect a great deal of data (using precious time and money) or be in danger of overfitting your dataset. It uses less initial resources to cater your campaign to an average audience rather than using precious time and money to test a number of different personalized campaigns.
David’s response was short and elegant. To paraphrase him, we are not selling something to you but rather are asking you to go through the process of buying something. Consider your attitudes and decisions after buying the product - are you happy with it? Will you buy it again? Will you recommend it to your friends? If you have personally invested the time to choose the product you need right now, and the product ends up not being quite the right fit, you will blame yourself rather than the company, and your subsequent behaviors will be favorable toward the company.
That made a lot of sense to this cognitive scientist, as cognitive dissonance is a very powerful phenomenon where people will rationalize actions taken and decisions made until the end of time if they have invested the time and resources into making them.
His answer shows that to interpret the data correctly, data analysis methods must also be accompanied with a useful framework. He focused his answer on the journey of the individual customer rather than segmentation of the customers by demographics or by distribution channels.
Before speaking with him I represented the process in my mind with a simple conversion funnel, which I discovered through my personal experience with Google Analytics. You can set up any of a number of models with their tools, but any easy starter model represents aggregate customer experience through a series of funnels - steps taken toward a final goal (e.g. signing up for my weekly newsletter about which chocolates are on special) - with a conversion rate measured at each step. This particular conversion funnel model helps assess where I might have problems on my website (e.g. hard-to-find navigation buttons and links, or poor wording on a page) by making it easy to see the step on which many users drop out before reaching the final goal.
If, instead, we want to emphasize something akin to David’s suggestion above - something like brand empathy in which the consumer and the company ideally have an equal input in the choices the user makes - we might like to visualize data on return customers and brand recommenders.
Additionally, brand empathy goes hand-in-hand with the job-to-be-done view of marketing - the view that a company must deduce the “job” that their customers are “hiring” each product to do. If a brand’s role is to engage customers on what they need to get done right now (e.g. I need to cool off) then they will have more success selling chocolate ice cream to customers who need to cool off than to customers fitting the demographic profile of previous chocolate ice cream customers.
How can we see the impact of this shift in marketing strategy? Traditional market segmentation - by customer demographics or product demographics rather than by the job-to-be-done - makes it hard to see the impact of brand empathy. Segmentation by job-to-be-done thus would allow for quantitative market and sales analysis within the personalization and brand empathy framework.