I had a conversation the other day that really blew me away. I was with some retail brokers that I respect deeply and trust implicitly.
We were showing them the latest version of our software hoping to turn them into customers and the subject of retail sales forecasting models came up. Their perspective was fascinating, though in the end, shocking.
They explained that most retailers they work with want a sales forecasting model. I made the point that a model would never be as good at picking sites as experienced retail real estate people because people are much better at pattern recognition than mathematical models. They agreed with me but then said something that blew my mind. They claimed that retail real estate professionals like models because if a chosen site ends up performing poorly, they can blame it on the model. Basically, “The model said it would work,” is a valid excuse for poor site selection. My reaction was:
- I never wanted to believe this.
- You’ve got to be kidding me!
Now I am SURE there are retailers where this excuse doesn't cut it. But I trust my friends and so I am equally sure there are lots of places where it does.
Because I am a geek at heart, I’d like to examine this in the context of brain power. Bear with me; this will get us to the sea slugs. Brains are built on a large number of connected cells called neurons and these collections are typically referred to as neural networks. Neural networks are known for being extremely effective at pattern recognition, which is probably why animals and human brains run on neural networks instead of a binary system like a computer. Pattern recognition is a skill useful for finding food, avoiding danger, and finding mates. It is also a good skill to have when trying to determine the merits of a retail site.
I wondered how many neurons it would take to recreate the typical retail regression model. This turns out to be relatively difficult to nail down, but we can still get some pretty powerful insights just by ball-parking. There is software that lets you create neural networks in computers and according to best practices by neural network programmers*, the number of neurons required is the number of inputs (let’s say 10 variables like population, income, number of competitors, etc. for a typical retail regression model) plus the number of outputs, which is one (the sales projection). This is also a middle layer of neurons which is somewhere around the average of the input and output layer, so 4 to 7 neurons. This leads to a total of between 15 to 18 neurons (see the graphic if you like pictures).
But wait. Retail modelers are very sophisticated so I am probably oversimplifying their work. Instead let’s be very generous and assume that I underestimated the neural requirements by a factor of 1,000. So we actually need 18,000 neurons to reproduce a sophisticated retail model.
Now let’s put this in perspective. The human brain has about 85,000,000,000 neurons. That’s 85 BILLION! 18,000 neurons turn out to be the number of neurons in a sea slug. Now, I’ll grant you we don’t use all of our neurons for site selection and I’ll also admit that a sea slug is pretty damn good at... well, being a sea slug. But, given the difference in implicit brain power, the responsibility to recognize the patterns for something as complex as retail has to go to the experienced human, not the sea slug. If your model is outperforming your real estate professionals, perhaps it’s time to get new real estate professionals (or at least some sea slugs since they’re known to work cheap). True real estate professionals, given the proper information in the proper context, will blow the doors off of a model much more often than not.
Please don’t get me wrong here. I like models. I used to make models. I think retailers should have models. Models are an unbiased check on our decisions and can help prevent disaster when someone is biased or missing something basic. And most people who build models are smart and know what they are doing - but only to the limits of what models can do.
I am now going to be very blunt. Allowing people to blame bad retail decisions on something that has the intellectual firepower of a sea slug is simply bad, irresponsible management. Every retailer should have a model but the model should never, ever, ever be allowed to overrule the professionals, neither in favor of, nor against, a site. Therefore the model should not take the fall for a bad decision. The decision, and therefore the blame, needs to placed squarely on the real estate professionals. Anything else is an abdication of responsibility.