Wayfair has millions of delivery addresses. It knows exactly which households buy the $1,400 sectional and which never click past the free-shipping banner. So when it started opening physical stores, I expected those eight to trace its own sales heat map.
I mapped all eight and pulled the 16-minute drive-time trade zone around each. Then I went hunting for the number Wayfair screens on.
I found one fast. Retail spend per household, divided by household income: the share of a family's money that turns into store purchases. Call it spend propensity. Across all eight Wayfair zones it sat at 0.40 and barely moved. Measured as coefficient of variation, how far a number swings store to store, it came in at 7%. Population and home value each swung more than three times as hard. This one number held.
I almost wrote it up as the screen.
Then I did the thing I should have done first. I ran it on twelve towns Wayfair would never build in.
The number that wouldn't move is the same in towns Wayfair would never touch
I pulled twelve random US metros as a control. Not cherry-picked comparisons. A deliberate spread from the poorest ground in the country to the richest: Clarksdale, Mississippi, where median household income is $40,000 and more than a third of the population lives below the poverty line. Beverly Hills. The Bronx. Bozeman. Fresno. McAllen. Flint.
Spend propensity across those twelve averaged 0.38, a 7.8% coefficient of variation. The same number. The same tightness.
Clarksdale, the poorest place in the set, posts 0.430. That's higher than six of Wayfair's eight stores. Beverly Hills, the richest, posts 0.40, dead on Wayfair's average. Income triples across that chart. The propensity band doesn't move.
So spend propensity isn't a Wayfair screen. It's a property of the United States. If I'd stopped at the eight stores, I'd have sold you a constant and called it a discovery.
Why the number can't move: the spend figure is built from income
The reason matters for every demographic screen you've ever trusted.
The retail-spend numbers in trade-zone reports, mine included, are modeled. The data vendor doesn't follow families to the register. It estimates each household's spending from income, age, and household size, then adds it up across the zone. Income is the heaviest input.
So when you divide modeled spend by income, you're mostly dividing income by a version of itself. The ratio comes out near-constant. It's arithmetic wearing a demographic costume.
Maybe that's a modeling artifact. Maybe Americans really do turn about the same share of income into retail spend wherever they live, which is roughly what the Consumer Expenditure Survey shows. Either way the conclusion holds. The tightness is real and the signal is zero. A number that reads 0.40 in Beverly Hills and 0.43 in Clarksdale can't tell a site committee where to build.
So what is Wayfair actually screening for?
Throw out propensity and the question reopens. To answer it I stopped asking how tight each number is inside Wayfair's eight stores, and started asking how tightly Wayfair holds it against the country.
That's the test that separates a screen from a coincidence. A real screen is a number Wayfair pins to a narrow, elevated band that the rest of the country scatters all over.
Read the long gray bars. Home value swings 99% across random America, and Wayfair holds it to 25%. Trade-zone population swings 85% across the country, and Wayfair holds it to 25%. Grad-degree share, rent, income, poverty, all the same shape: wide open in the country, clamped down at Wayfair.
Now read the bottom three. Propensity, household size, median age. Wayfair's no tighter than a random spread of US metros. Those are the numbers that look like screens inside the eight stores and evaporate the second you hold them against the country.
And the real screens aren't tight numbers. They're elevated floors.
The screens are floors, and Wayfair clears them high
It's not just that Wayfair holds these numbers tighter than the country does. It holds them higher, and the floor is the screen.
Every Wayfair zone clears about 180,000 people inside a 16-minute drive, and seven of the eight clear 350,000. The random-US control drops to 16,000. Median home value sits at $558,000, double the random-US median and well above the national $340,000 or so. Grad-degree share runs double the control, and poverty runs well below it, even though the control set itself sits high.
So the screen is metro wealth and density, written as floors, not any spend ratio. Read off the eight stores, the floors land around 180,000 people in a 16-minute drive, $67,000 median income, $305,000 median home value, a graduate-degree share above 11%, and poverty below the control set. Dense, affluent, educated, lower-poverty than the country's poorer metros. The numbers don't have to be tight. They have to clear the bar.
The last screen the demographics can't see: an empty box
Wealth and density floors narrow the United States to a few hundred candidate corridors. They don't pick the corner. Something else does, and it never shows up in a demographic table.
Wayfair isn't building. It's moving into boxes that already exist and already failed.
Yonkers is the entire former Lord and Taylor at Ridge Hill, 88,705 square feet, dark since 2021. Denver is a former Macy's at the Shops at Northfield, Atlanta a former Walmart Supercenter, Fort Lauderdale the dead anchor end of the Galleria. These are second-generation big boxes, stores built and run by a prior tenant and now open for the next: parking poured, utilities run, permits and zoning long settled, sitting empty inside metros that already clear the wealth and density floors.
Eight stores is a thin sample, so be careful with the causation. I can't fully separate "Wayfair screens for dead boxes" from "Wayfair takes dead boxes because in metros this wealthy, the only 80,000-square-foot retail space on the market is a former department store." The vacancy may be a true filter or just the cheapest way in. Watch for the counter-example: a qualified metro where Wayfair builds new, or passes for lack of a box.
What I can say is the order, whichever way the causation runs. The demographics qualify the metro. The empty anchor box picks the corner. And what Wayfair buys is clear: 80,000 square feet of finished retail, parking poured and permits long settled, on a second-generation cost structure instead of a new-construction timeline and entitlement fight.
For the rest of us the signal is the same either way. Track the vacancies, not the rooftops. The next Wayfair announcement will be a specific empty department store in a metro that already cleared the floors, not new construction off a fresh interchange.
Run the baseline test on your own model
The mistake I almost made is the one every site model makes. You find a number that holds across your wins and you crown it a screen. You never check whether it holds everywhere else too, including the ground you'd never buy.
So check. Pull your last ten approved sites, ten you rejected, and ten at random, including markets you'd never enter. Run the same variables across all thirty.
Your floors won't look like Wayfair's, and that's the point. A furniture chain pulls a 16-minute drive and needs six-figure home values. A drive-through reads a 5-minute trade zone where 50,000 people is a strong number and Wayfair's wealth tier is beside the point. The specific floors don't transfer across format or radius. The method does.
A real screen does two things at once. It sits tight and high in your approvals, and it scatters or drops in the rejects and the random set. A variable that's tight everywhere, approvals and rejects and random alike, is a passenger. It feels like insight. It tells you nothing. Most models are hauling at least one.
For Wayfair, the real screens are wealth and density floors plus a specific empty box. The number that looked like the secret was the country all along.
Last week I ran the same drill on 71 Buc-ee's and reached a version of this from the opposite end: a chain that screens hard on highway access and lets the trade-zone population scatter 78 to 1. Read that one here. Different retailer, same lesson about which numbers actually carry signal.
Economic Pulse
The headline inflation scare is an energy story. Brent fell 28% off its peak in the same window the CPI and PPI reports posted, so the spike and its reversal are arriving almost at once. Redbook's +10% looks hot until you adjust for inflation, which leaves it far softer in real terms. For site selection: construction and fuel-linked costs spiked and are now easing, while financing stays expensive and isn't getting cheaper this year. Underwrite the rate, not the headline.
The fortnight's closures
Read the top two rows the way a site team does. West Marine's 59 boxes and Saks Off Fifth's 57 are pre-built retail in coastal and urban markets, parking and power already in the ground. They're too small for Wayfair, which needs former anchors in the 80,000-plus square foot range. But for any retailer whose footprint runs 3,000 to 40,000 feet, this is the supply chain: take the dead box, skip the construction, open inside a metro that already cleared the floors. The closure wave restocks the shelf, and if you run a concept that can reuse a big dark box, this fortnight just added more than a hundred.
And if you run a pad instead of a box, the row to watch is On The Border: 27 freestanding restaurants, most of them in Texas, in the 5,000 to 8,000 foot range that fits a drive-through conversion. It's the most time-sensitive opportunity in the table. One caution before you chase it: the closures came with no bankruptcy filing, so confirm whether those leases revert cleanly or are tangled in franchise and assignment disputes. The West Marine and Saks boxes run too big for a pad.
I exported all twenty trade zones, the eight Wayfair sites and the twelve control metros, with income, home value, education, per-capita spend, and the propensity ratio, so you can run the baseline test yourself. Download the CSV. No email gate.
-Andrew
Head of Marketing, GrowthFactor