Data, Data, Everywhere And Not A Drop To Analyze

My sincerest apologies to Coleridge, but this has been a theme for years in Marketing Research. There is an axiom among many researchers, those who perhaps barely passed that one stat class back at school, that more data is better data or that you can never have too much data. Neither statement is a fact, either in the statistical sense or the philosophical sense.  What’s important is to have the right measure to answer your business issue, to have enough data to be statistically sensitive, and to be representative of the population you want to speak to. Beyond those criteria, more data is useless. Yet we don’t seem to understand this point very well. To illustrate, let me take the report by Gary Angel, who forgot those facts and went on Bob Lederer’s RBDR to tell us about all the measures we need to think about regarding displays.

Gary wants to make the claim that visual merchandising and in-store display is at the heart of modern retail. I might argue that price and selection are at the heart of modern retail, but I don’t want to discount the value of a nice looking shelf. He also believes most retailers are clueless about how displays and visual merchandising works. Hogwash is the term we use here in the South. Here’s the problem – Gary doesn’t seem to understand what displays are meant to do.

Gary proposes four measures that are critical – are displays attracting shoppers, are shoppers getting hands-on with the product (are they squeezing the Charmin?), are they getting people to look at the products the way you want, and are they experiencing the products the way you want.  He goes on to tell us that, “Using a combination of video cameras and machine learning you can understand how displays attract shoppers, how many shoppers pass by a display, how many stop and look, how many really engage, whether a display gets people hands-on with the product and whether they take the product away.”  You can then use a combination of these measures to run structured tests to help you optimize the display experience, which lets you optimize the store.

Displays serve one purpose for retailers – they produce revenue, either by sales or rental income, or as a place to store short-term inventory during a promotion (so you can sell more). And contrary to what he thinks, retailers are very good about understanding how much more product each store will sell on display – and that’s how their ordering systems work.  Displays can serve a secondary purpose for manufacturers – it’s a way to introduce a new product whose introduction might otherwise go unnoticed. But again, that goal is to build a franchise (sales).

Of course, we might argue that if you need machine learning to figure out whether a display is good or bad, you have no business being in this line of work. It’s way too simple to know this.

And consider the measures he proposes. How many shoppers see a display is more a function of where the display is in the store than what is on the display- it’s why we try to avoid dead zones in a store. How many people interact with the products on the display is a function of category interest, product interest, price point, number of products on the display, and, finally how interesting the display looks. Put a display for dry yeast and nobody’s interacting with it. Put up a Coke or Pepsi display and the numbers improve dramatically. Unless you’re in fashion, makeup, or media, people don’t much need to interact with your product on a display, short of buying it, and it many cases you probably want minimal interaction – messy displays won’t help.

You can spend a lot of time talking about all the different things one can measure in a display, or you can talk about the only measure that matters – sales. You can collect a whole set of KPIs via video and analysis via machine learning to get a hint of what drives the success of the display. Or you could figure out that there are only so many factors in a display for your product that they are easily testable without all the preamble. And when we test, we’ll look at category and brand sales as your performance measure.