Accenture on Analytics

Accenture has an interesting little study published here:

http://newsroom.accenture.com/article_display.cfm?article_id=4777

Short story: everybody thinks they need better analytics because they still make lots of decisions based on intuition. True, that.

Still, I wonder if there is enough attention being paid to some of the reasons for the gap between the need and the practice.

One thing that keeps surprising me is how much data there isn't, especially given the data there is. One firm I worked for quite literally could not identify customers who disconnected their service, moved to a new home, and then reconnected service. There was, then, no way to tell what proportion of "new connects" were actually new and what proportion were old, reliable customers simply re-establishing their service.

But the even more profound problem is the question of what to do with the data that exist. If you attempt to perform analysis without some theory behind it -- and this happens all the time -- you will get entirely subjective results. Basically, you wind up with an argument that says the company ought to substitute the analyst's prior beliefs and intuition for the intuition and prior beliefs of the other managers. Firms are going to find this to be less than compelling.

Metering Price Discrimination

Really interesting paper in Marketing Science this autumn by Gil and Hartmann on metering price discrimination.

Lots of firms sell optional secondary products to customers who purchase some primary one. Movies sell popcorn to those who purchase tickets, your cable TV company sells HBO to those who are otherwise subscribers. And what we frequently observe is that these secondary products are sold with markups that are significantly larger than the principal product.

It isn't obvious why this should be the case.

If the demand of the marginal consumer in the secondary market is less that the average secondary demand of all those who buy the primary good, then the firm will optimally charge a premium. Which is to say that movie theaters who charge larger markups for concessions, and cable companies who charge a larger markup for HBO or Cinemax, are making a specific assumption about their marginal customer. Namely: that the last guy to buy your main product has a lower willingness to pay for the second product.

Now this might -- or might not be -- true. But it's easy to imagine why we might suppose that the last person to buy a ticket to Avatar is just as likely to want popcorn as anybody else. One interesting modeling insight in this paper is to look at the weekly attendance numbers of movie theaters and compare those with the per-customer amount of concession sales. If average concession sales per person are negatively related to attendance, then larger markups are the right policy.

This is because weeks with high attendance are pulling in people who otherwise wouldn't go to the movies -- they are going even though they normally have a low willingness to pay for movies, but some movie of particular interest gets them in.

As it turns out, with the data set they have on movie theaters in Spain for a few years, this is exactly what we observe. A question and an observation.

First, how do these results translate to other industries? Razors and blades, cable TV and premium movie channels, etc?

The observation is that we can't suppose that theater operators have adopted this policy with this analysis in mind. It is a case where the habits of an industry seem to have been adopted by nearly all participants simply because it works. In applying this sort of analysis to other industries, it's a good idea to remember that the current policy might be accidental -- and to point out what assumptions that policy is implicitly making about the behavior of the market.

Role of Analytical Marketing

Looking at marketing problems through the lens of economic theory and statistical methods is what we mean by the term, "analytical marketing." The real benefit to this approach is that it can serve to resolve issues within marketing departments that are otherwise dependent on intuitive judgment or other, less rigorous pieces of analysis like consumer research. Some questions really can't be solved without working through the theoretical implications of utility-maximizing consumers who face a budget constraint.

I suspect that what is happening in marketing is similar -- in some ways -- to what happened when economists started reading Samuelson (who has just died over the weekend at age 94). Of Samuelson, Robert Lucas said, "He'll take these incomprehensible verbal debates that go on and on and just end them; formulate the issue in such a way that the question is answerable, and then get the answer."

A lot of marketing departments will attempt to attack important questions without allowing themselves to build the proper framework for discussing the issue. So we sit in meetings where marketers will say things like:

• I wouldn't stop using our product if it cost a little bit more
• Customers haven't reacted to price increases before, but this could be the straw that makes them all switch producers
• Somebody like my grandmother only cares about [feature X]

Thing is, these statements might all be true. Problem is, everybody shows up at marketing meetings with a pocket full of these prior beliefs and little anecdotes about mythical consumer types, and feels free to put them to work -- whatever the question might be.

One solution is to use data to answer questions. The problem is that data seldom want to cooperate. If you look at a graph of customer churn for the last five years, for example, you are going to be able to fit all sorts of trends into the data. And every shift in the trend that we think we see is going to have lots of explanations -- often the supposed behavior of the same mythical consumers we were trying to ignore in the first place.

What we think we can do with the right analytical approach is not so much simply using the right statistical tools to answer the question, but using theory to frame the question properly.

Here's what I mean: Suppose we sell a primary good as well as a secondary, optional, good. Movie tickets and popcorn, for example, or basic cable and premium movie channels. We care a lot whether reducing the price on the secondary good will get more people to purchase the primary good. And this is precisely the kind of question most likely to be subject to stories about whether someone's grandma never bought popcorn at the movies, or whether the customers will all defect because a competitor is offering HBO for $1 less than we are.

Moreover, this is precisely the kind of question that isn't likely to be settled by looking at some data. Every observed change is going to be over-determined: every change will have a dozen explanations.

Instead, what analytical marketing can do is to reduce the number of variables to something manageable, and define what must be true in order for some policy to be correct. It is one thing to say, "People won't respond to a price change." It is another thing to say, "I think the correlation between the two products is lower than the threshold value." One statement is meaningful (we might be able to estimate the correlation of demand), while the other is not.

So analytical marketing is always asking: what does it mean to be right; what does the right policy look like; how can we test it?

And that's the main benefit, in my estimation. Analytical marketing reduces the scope of disagreements and the power of unexamined prior beliefs.