What’s worse? Too little science or too much science?
Ahh, trick question. The answer is that science is not a quantity that can be measured and compared to a standard. Most people would agree with this, but we often hear people say “we need a little more science in our site selection process.”
For some chain operators, “science” is the corrective action for dog stores. I once had a client whose board of directors shut down new store openings until there was evidence of a “scientific process” for approving sites (they had opened about 10 bad stores in a row).
In the past 10 years many companies have increased their investments in computer-based mapping and demographic systems as well as sales forecasting models. This was quite challenging for the chain store industry as a whole because they were the last sector of the economy to spend big money on information technology.
I had the pleasure of rolling out mapping and analysis software to about 50 field development managers at a company’s annual offsite in 1998. Keep in mind that some people didn’t even use email then. They opened many stores each year and the decisions were primarily based on the judgment and experience of the field reps. “Skeptical” didn’t begin to describe their reaction to a computer simulation of their markets! The company now has over 100 users of the software and they can’t get deals approved without completing the site evaluation online. There is far more objectivity and transparency in the process and they have numeric benchmarks to guide their decisions.
However, some folks are wondering if we are starting to place too much emphasis on the “science.” I’m one of them.
The problem is not the “science” itself. The problem occurs when we disconnect the decision maker from the data with “push button” analytical methods that generate go/no go recommendations.
I began thinking about this issue in 2005 when I read Malcom Gladwell’s book Blink (subtitled “The Power of Thinking Without Thinking”). The book is full of examples of experts who make fast, excellent decisions under high pressure without formal analysis such as firefighters, pediatric nurses, and military leaders. Gary Klein, a leader in the field of decision theory, calls this “Recognition Primed Decisions” whereby a person assesses the situation, searches his mind for similar cases, and adapts the knowledge to the situation at hand. In chain store real estate analysis, this is how analog comparisons are done.
What do life and death emergencies have to do with chain store real estate decisions? Although the time pressure is not as great, they both are complex decisions. The brain of a human expert is capable of finding meaning in complex problems that only confound mathematical models. The reason is that a human considers the context of the situation adapts his decision rules accordingly. For example, a burger joint might generate average sales in a suburban location; unless it’s next to the NASCAR track.
Let’s compare the data processing of a sales forecasting model using multiple linear regression to a seasoned real estate dealmaker looking at google maps and some demographic information.
The example site is a for a quick serve restaurant on a pad location in front of a power center in a suburb of a medium sized city.
The regression model has 10 variables including resident population, daytime population, median income, presence of children, competition, and some site characteristics. When the address is entered, the values of these variables are computed and an equation is applied that generates a sales forecast that is close to the average unit volume of the chain.
The dealmaker looks at the map and sees that the interstate highway separates a major office complex from the power center. There are a couple of stoplights on the road that goes under the freeway and it’s likely to get congested at lunchtime. The demographic report shows 30,000 people within 3 miles and 15,000 people within 2 miles. However, the most dense neighborhoods close to the center are lower income according to the shaded map and there are other fast food restaurants between the center and the more affluent neighborhoods. The dealmaker remembers another deal that he did a couple of years ago with a similar situation and the sales were about 20% below the chain average because they simply weren’t capturing the potential business that appeared in the demographic reports.
The current state-of-the-art in site modeling cannot pick up the key insights that this dealmaker easily identified. However, the dealmaker still needs the map and demographic report for the data that he uses to reach his conclusions.
I’m not suggesting that regression models are useless or that humans are infallible! Humans will have biases that color their view of the world and good statistical models can uncover patterns in data that humans can’t see. The key is to have a decision making process that allows the human to see the important data elements in context to build the story that draws on his experience and knowledge.