Applied Statistics for Restaurant Strategic Menu Choice

Theoretical case study using statistical analysis that can be used to determine insights and actionable recommendations for clients.

Statisticians are sexy. At least, that is what Simon Chadwick of Cambiar Consulting said at the recent CASRO Digital Conference. Well, he really said that the new ‘sexy’ job was to be a statistician. When I began my talk the next day, I quoted Simon. Then added, ‘it’s nice to be sexy again. It’s been awhile’.

Simon also said that, with the exponential explosion of data, ‘storytellers’ will be more and more in demand. The C-suite will be searching for those who can sift through the noise and boil down the ‘story’ to a manageable action plan.

I write novels. As a hobby. I’ve written nine so far, four published and none successful. Writing is a terrific hobby for a statistician. As I like to say to my clients, if one can write a novel, one can write a memo.

Creativity is in.

The Nature of Statistics

When my students on the Principles of Market Research Course given by the Georgia Center for Continuing Education (a plug for the course—POMR received the Quirk’s Featured Company Gold Medal for 2013) hit the statistical module section and throw panicked questions my way, I always respond: Statistics is a numerical description of an event. If you understand the event, the methodology follows. For example:

  • A mean is simply an average of a group of numbers;
  • A standard deviation is the average distance of each point from the average. The larger the standard deviation, the more dispersed the numbers. A small standard deviation means a tight group of input;
  • Regression analysis shows which variables are statistically linked to the dependent variable;
  • Segmentation is the statistical grouping of individuals based on behavior or attitudes;
  • Conjoint analysis describes, by statistical methods, what are the greatest variables in the choice process.

A Basic Principle of Marketing Research

When beginning a new project, I ask the client, ‘At the end of the process, what would you like to hold in your hand?’ My golden rule in any marketing research project is: Begin with the end in mind. In other words, solve the problem.

Then work your way backward on methodology.

We are going to illustrate our point using a fictional study that begins with the end in mind, then creatively use numbers—statistics—to give our clients the tools to make their business decision.

Johnny’s American Diner

Johnny’s American Diner is a fictional diner chain located primarily in California (named after the Marlon Brando character in The Wild One). Johnny’s American Diner has begun to open branches in the South and in major malls in the Philadelphia area. Given their expansion, Johnny’s American Diner would like to reboot their menu.

The first step for a menu redesign is to decide which items to remove. That’s where we came in. Johnny’s American Diner hired us to design a study with two goals in mind. The first was to evaluate the existing menu. The second was to test-drive possible menu items.

It was in the first task that creative statistics were employed.

Below is the basic menu at Johnny’s American Diner.

 

Johnny’s American Diner’s Classic Menu

Johnny’s American Diner Study – Item Purchase Intent to TURF
The first section of the Johnny’s American Diner study was to list all of the current menu items. Respondents were asked to rate on a 1-to-5 scale how likely they would be to order each item. This is a typical construction for a TURF analysis. Variables were recoded (1 2 3=0) (4 5=1) into not likely/likely. These values will play an important role in the analysis later even though TURF was not a primary deliverable in the study.

Creative Questionnaire Development – Sequence Weights
The second section of the study—originally developed by Mitch Markel of the Benenson Strategy Group—was constructed specifically to evaluate the sequence of menu choices respondents make. Johnny’s American Diner envisioned a survey where respondents would be isolated within menu categories. Based on their choices in the TURF section and what time of day they said they usually visited Johnny’s (referred to as day-part), respondents were placed into splits where they were to choose among menu items. No respondent was placed into more than two menu categories. Below is a table of how the selection of menu category was performed.

 

 

A respondent would be sent to a menu category and asked, ‘Which of these items would you order first?’ They chose. Then they were asked, ‘If this item were not available, what item would you choose next?’

Sequence weights were created in the following way. If an item was the first in a category, it would be given a sequence weight of 1. For example, there are 12 possible drink items. If one chose ‘Large Lemonade first, his sequence weight would be 12/12=1. If ‘Large Soft Drink’ was chosen second (after Large Lemonade is removed from the choice set), it would receive a sequence weight of 11/12=.92. And so on until the respondent answers ‘No other item.’

Given that different categories had different number of items, the weights are not balanced. In other words, Burgers – Sandwiches had 15 forced choices, then the second choice weight would be 14/15=.93. For Dinner Combos, a category with only 10 items, a second choice would be 9/10=.90. Very few respondents ran through an entire category, but the same problem would hold at the bottom. The lowest weight in Dinner Combos would be 1/10=.10. The lowest in Burgers-Sandwiches would be 1/15=.07.

All items that were not chosen received a sequence weight of 0.

An adjustment was needed in order to view the choice weights across categories equally.

Menu Category Weights

The full-service marketing research client and I brainstormed. We came up with a creative solution. We decided to weight each category by the number of items ordered in TURF section to balance out the sequence weights. Here is how it worked.

Each respondent was likely to order a different number of items from differing categories. So we decided to add up all his menu choices, then divide that number by each category. The individual category weight would be calculated based on the number of items from each category a respondent chose. Table 1 below gives an example.

Table 1 – Calculation of a Category Weight

 

 

Final Item Weights
The final items weights were then calculated using the two weighed dimensions.

 

 

Developing the Picture
We now have weights for all the items on Johnny’s menu (many of them ‘0’, because they were not chosen). But a typical weight might look like this:

 

 

Let’s say that we averaged the Item Weight across all the respondents and found that ‘Large Soft Drink’ has an average weight of .35. What does that mean? Is that good?

The answer, as most statisticians will tell you, is ‘that depends.’ It depends on whether that weight is higher or lower than the weigh for say, ‘Cheeseburger Deluxe’. How much higher? How does Johnny’s American Diner make the decision of which items to cut (and which to promote) based on .35?

Index Modeling

Another rule of thumb: Always present your findings in a manner that the end client can understand in a glance. In this case, the answer is creating percentiles.

A percentile is a measure indicating the value below which a given percentage of item weight in a group of item weight fall. For example, the 20th percentile is the value (or score) below which 20 percent of the item weights may be found. If a score is in the 86th percentile, it is higher than 86% of the other scores. Creating percentiles is easy. The highest value is 100%. The lowest is 0%. The percentages between are the relative distance each item is from the top or bottom.

We reported the percentiles to the Johnny’s management team. The initial report, given that Johnny’s wants to cut items from the menu, is shown below. Table 2 shows the bottom 12 items by item-weight percentile.

Table 2 – Lowest Item Weight Percentiles

 

 

The production of percentiles for item weights is also very flexible. Johnny’s would like us to filter the item weights by only those who visited the restaurant for breakfast. Moreover, they would like to see only breakfast items in the percentile chart. Table 3 shows the percentile output of this filtered request.

Table 3 – Breakfast Item Weights by Breakfast Day-Part

 

 

Conclusion

At the end of the process, Johnny’s American Diner has the tools to make decisions. We have creatively employed the questionnaire and delivered to your client an easy-to-read snapshot of their menu-item demands.

As the creation of data multiplies at an exponential pace, it falls more and more on the marketing researcher not only to design the study, but to creatively produce actionable results.

One does not have to be a statistician/novelist to create a score simulator to Johnny’s American Diner’s choice item weights. One must simply understand the question that needs to be answered, the goal of the study to be reached, and how to use simplicity to deliver sophisticated output.

Please share...

Join the conversation