The Man Behind the Map: 10 Lessons from the Godfather of Demographics

What’s the one data source behind nearly every retail site decision in the past 50 years? If you said “demographics,” you nailed it. Today I had the chance to catch up with one of my role models and an OG in the field: Gary Menger, founder of Applied Geographic Solutions.

Gary has been shaping the demographic data landscape since his early days at CompuSearch in the 1980s. If you’ve ever used demographic estimates to guide a decision, there’s a good chance Gary had a hand in it. What most people don’t know? He also built the Mosaic segmentation for Experian. Not once, but across three U.S. census cycles. That’s 30 years of consumer segmentation evolution.

You can dig into the full conversation here: Youtube | Spotify | Apple Podcasts

Here are my biggest takeaways from our conversation. 👇

#1 Build Models To Avoid Blunders, Not Find Perfection 

There is no such thing as an “optimal” site. Because even if you do find the perfect site, when your competitors move in next year it will not have the same sales. “People are chasing this imaginary optimal site. It doesn’t exist. Avoid the blunders - that’s the job.” I love the practicality here - and the way that this ethos eliminates analysis paralysis and increases speed - a massively underrated trait of great store development teams.

#2 Site Selection Must Evolve From “Site-Centric” To “Network Centric”

Most tools today inherently focus on finding the “top performing site” in a city. This is the wrong way of thinking about it. “Retail is no longer about individual sites. It probably never was… The best site may not be the best for the entity as a whole. You have to have a certain critical mass within a metropolitan area so that people know your brand name - and minimize the number of sites to get the coverage you need in order to drive the sales of a brand as a whole.” You could place a store in the city center to maximize single store forecast — or you can space stores strategically to avoid cannibalization and maximize total sales across the entire city.

#3 Help Crusty Earl Focus On The Good Sites

Gary talks about “Earl”, the crusty real estate guy with 30 years experience in site selection. In reality is Earl is actually legit - he is a walking neural network on what works in real estate. What I didn’t understand at the beginning of my career, is that the point of real estate analytics is not to replace Earl - it is actually to eliminate bad sites so Earl can focus on the good ones.

#4 Somewhere lurking in your demographic data is a site killer

Keep your eyes out for unusual numbers in your demographic site report. Like a blood test - if there is one marker out of range in one direction or another it can kill a site. How can you spot them?  Look at all of your variables relative to the nation. This is where an index can be incredibly helpful. Say you're a high-end furniture retailer and spot a location with affluent, upscale demographics. Looks perfect — until you see the renting index is 300, meaning residents are three times more likely to rent than the national average. They’re not sticking around long enough to buy anything beyond IKEA.

#5 Don’t get stuck on one variable (aka income)

Retailers often fixate on a single metric — like income — and can’t get past it if it doesn’t check the right box. I get it: the CFO wants to see a certain income range because that’s when the model says your sites start performing. But as Gary puts it, “What if that neighborhood hits your target income, but their mortgages are 20% higher than the one next door?” Same income, totally different spending power.

#6 The Truth About Your Customers Might Hurt - But It’s Your Job To Say It

Every analyst knows the feeling: “Damn, my boss is not going to like this.” It’s even worse when the data challenges the company narrative — like when leadership assumes their customers are more affluent than they are, or a different ethnicity altogether. That’s when the real work begins. The best analysts have the conviction to trust what they’ve found, the courage to speak up, and the communication skills to help the company face the truth.

#7 No Peeking - How Experian Mosaic Does & Doesn’t Use Sales Data

There is a misconception about how Experian Marketing uses the Experian Credit data. Because of the Fair Credit Reporting Act and key legal precedent set during TransUnion Vs. FTC - means Experian cannot share or use credit report data (including credit card tradelines and payment history) for third-party marketing. Gary had to use demographic variables to predict people’s income and get a report back from Experian on how close he got. Gary explained this like “playing a video game with a joystick and the TV is in another room.”

# 8 The Snowmobile in Florida Problem: Beware Small Survey Samples

When Gary first met Richard Webber, they were discussing “superclusters” — a 300-segment classification system. It sounded powerful. But here’s the catch: retailers rely on their own customer data or surveys like MRI Simmons to enrich those segments. And when your segments are that small, you often end up with just a handful of data points per group. That leads to wild, misleading indexes. The result? One Bentley owner in the “wrong” segment, and suddenly you’re building an entire marketing strategy on a fluke. Or one demographically similar snowmobile owner the north sends you in the wrong direction in the south. Don’t over-segment. When using survey data, keep your clusters big enough to smooth out the noise, and be cautious of sample size — or risk chasing ghosts in the data.

# 9 Predictive Modeling: There is No “One” Answer Take The Average Of The Top 20 Models

I have seen this approach multiple times now where people will build several different predictive models with several different approaches and take the median of those models. For example you could take the median of a black box model, a subjective model based on agreed upon site criteria, and, and an analog model. “One of the things that I think where we made mistakes was that we imagined that there was one solution to a problem. And these days I would say, look, build 20 different models and take the median value of the results. relying on one model, sooner or later that model is going to spit out something really stupid.”

#10 Bill Goldstein on “batting 300”

Gary shared a story where after a huge failure, his boss at Compusearch (Bill Goldstein) took him aside. Gary thought he was getting fired, instead he got a raise. That moment rewired Gary's entire relationship with failure. This stuck with Gary through all the acquisitions that happened. Whenever a geodemographic company was acquired, the leadership calcified and people quit innovating. Before the acquisitions there was also a spirit of collaboration, where folks like Gary, Jonathan Robbin, Richard Webber, and Bill Goldstein and others would freely share information across companies to make the end product better. This same spirit of desiring collaboration and building new and better things is why Gary started AGS in the first place - other than the fact that he also got fired upon an acquisition and had three kids to feed ;).

Check out the full conversation here:

Youtube | Spotify | Apple Podcasts