Brand Differentiation: A Reality the Dirichlet Model Can’t Account For

The Ehrenberg Bass Institute (EBI) has caused a debate among marketers by claiming that there is no such a thing as brand differentiation, only distinctiveness (e.g. recognizable packaging and logos).  My view: while distinctiveness is certainly important, so is differentiation. Mark Ritson among others agree with my position but that is not the purpose of this blog. 

I want to delve into the underlying reason why EBI makes this bizarre assertion that brand differentiation does not exist. It is because they MUST! EBI is founded on the Dirichlet distribution which is a multivariate version of a beta distribution, however, It has a fatal flaw…it cannot produce similarity patterns between brands! (Technical discussion at the end of this blog.)

In a Dirichlet world, no two brands can be more in competition with one another vs. any other brands because it is derived with that assumption…there is no “competitiveness” parameter to tweak. For example, in a Dirichlet world, Greek yogurt brands cannot be more similar to each other than they are to American-style thin yogurts. BMWs and Toyotas must be just as directly competitive as BMW and Lexus.

There can be no difference in switching patterns between dry dog food brands sold in supermarkets and fresh refrigerated high-priced brands sold in premium pet specialty stores. In other words, in a Dirichlet world, there can be no market structure. As illogical as this sounds, they must defend this position because if the Dirichlet model goes down, so does most or all of their worldview.

Briefly, in a Dirichlet world…

• Consumers cannot be consistently segment-able (or brands would be differentiated)
• Targeting is a destructive exercise in marketing because consumers cannot be segmented
• Loyalty programs are a waste of money
• Innovation teams that try to create something new and differentiated are wasting their time
• Positioning doesn’t matter, just distinctiveness and being “always on” in your marketing to be “mentally available”.
• From an MR point of view…
• perceptual maps must be wrong since no brands are differentiated and no clusters of similar brands could exist.
• Surveys are misleading and of limited value (The father of the institute, Andrew Ehrenberg, was quoted as saying, I never met strong or weak brands, only big or small ones.)

I have already proven that their aversion to targeting would leave significant money on the table (from case studies, we see that Movable Middles…those with a 20-80% probability of choosing your brand…are up to 23 times more responsive to advertising). Since Markov repeat rate is actually the most correlated metric to share growth, we can all reject their idea that loyalty programs are a waste. So yes, their worldview stemming from a flawed model is also flawed.

To be fair, they do present evidence in support of their thesis that brand differentiation does not exist.  It is based on meta-analysis of patterns (or lack of) in aggregate brand uniqueness scores. 

However, is it legitimate to investigate this thesis with aggregate data when individuals have wildly different brand perceptions and preferences? Recently, I worked on a huge study for a bank…10,000 respondents where we had CRM conversion data (and ad serving) merged in with survey results at an ID level. 

I analyzed conversion rates by individual respondent patterns in attribute check offs. A respondent checking off that all brands stand for an attribute but NOT you, is the worst pattern. Checking off that you and only you stand for that attribute is the best pattern (a pattern that reflects brand differentiation). You can imagine the combinations in between. 

For respondents who had differentiated beliefs toward our bank brand, their actual CRM based conversion rates were three times higher. Constant sum data across competitor brands in the study showed similar patterns.  Furthermore, when we segmented consumers from brand preference vectors derived from constant sum, we saw meaningful groupings of brands (online banks, mainstream, those associated with credit cards, etc.)

In research, we need to know how things work because we are delivering insights that direct marketing action based on repeatable principles. The EBI insights about how brands grow have serious flaws as they flow from the faulty premise that brand differentiation cannot exist. 

My advice: keep researching competitive patterns and brand positioning tentpoles.  Keep looking for clues as to which types of consumers will be more responsive to your advertising. Do try to understand the drivers of heaviness of buying because those ARE persistent patterns. Use precision targeting to drive up ROAS. Use AI to direct creative composition.

Embrace heterogeneity of consumers and differentiation of brands!

Technical note on the Dirichlet model.  See: Wikipedia description of the Dirichlet. You’ll see that the covariance formula is based on multiplying the alpha coefficients for any arbitrarily chosen pair of brands.

If you create, say, 5,000 synthetic records in R from a Dirichlet distribution, you will find that the brand probabilities of purchase across brands have a mild negative correlation conforming to the formula but no other pattern. 

The negative correlation comes from the constraint that probabilities of purchase must sum to 1. These patterns cannot be altered in a Dirichlet model. Another sign that brands must be uncorrelated in a Dirichlet model comes from its derivation: a series of Gamma distributions that MUST BE uncorrelated then normalized to sum to 1.

Contrary to the overly restrictive Dirichlet, a VALID multivariate view of purchase probabilities across respondents can be obtained from a constant sum question in your brand research. To statistically model these data, you can use a Gaussian Copula of Beta distributions which DOES allow for systematic covariance patterns across brands…there is a program in R for this.