The Huff Gravity Model, Explained — and Where It Breaks in 2026

Ask a GIS analyst to model a trade area and there’s a good chance you’ll get some version of the Huff model: a 1963 formula that splits every neighborhood’s shoppers among competing stores based on how attractive each store is and how far away it sits. It’s elegant, it’s intuitive, and it’s still baked into most mapping platforms. It’s also built on assumptions about consumer behavior that 2026 data can now test directly—and several of them don’t survive the test.

This isn’t a takedown. The retail gravity model deserves genuine respect: it turned site selection from gut feel into math, and its core intuition still lives inside modern forecasting. But if you’re underwriting real leases, you should know precisely where the classic ends and where observed data has to take over.

What the Huff Model Actually Says

David Huff’s insight was probabilistic. Earlier gravity models, like Reilly’s law of retail gravitation from the 1930s, drew hard boundaries: everyone on this side of the line shops downtown, everyone on that side shops at the new center. Huff replaced the boundary with a probability. A shopper three miles from a big store and one mile from a small one doesn’t belong to either—she splits her trips between them in proportion to each store’s appeal and inversely with the friction of getting there.

Formally: the probability that a consumer at location i patronizes store j equals store j’s attractiveness divided by travel distance (or time) raised to a decay exponent—divided again by the same quantity summed across every competing store. Three inputs drive everything:

• Attractiveness — in the classic formulation, square footage: bigger store, bigger pull.
• Distance decay — an exponent, typically around 2, controlling how fast willingness to travel falls off.
• The competitive set — every alternative a shopper could choose instead.

Multiply each block’s capture probability by its population and category spending, sum across the map, and you have a demand estimate for a candidate site—plus a genuinely useful picture of trade areas as overlapping probability surfaces rather than tidy rings.

What it got right

Three things, and they still matter. First, competition is relative: a site is only as good as its alternatives, which is why the same intersection can be great for one brand and fatal for another. Second, trade areas are gradients, not boundaries—an idea decades ahead of the data needed to prove it. Third, demand splits: opening near your own store takes trips from it, the founding intuition behind every cannibalization analysis done since. In 1963, with no computers to speak of and no behavioral data at all, this was a remarkable act of structural imagination.

The Fine Print: What Huff Had to Assume

Every model trades realism for tractability. Huff’s trades were reasonable in 1963 and are worth naming plainly in 2026.

Square footage as attractiveness

The model needs a number for “how appealing is this store,” and size was the only one universally available. But square footage says nothing about brand strength, price position, merchandising, drive-thru convenience, or co-tenancy. A 2,000-square-foot cult coffee brand can out-pull a 40,000-square-foot tired grocer, and the classic model has no way to know it.

One decay exponent for everyone

A single exponent assumes all customers weigh distance identically—the symmetric-customer assumption. In reality, willingness to travel varies by category (destination furniture vs. impulse coffee), by trip purpose, by daypart, and by customer segment. Commuters, families, and daytime office populations all decay differently, and averaging them into one number blurs exactly the signal a site decision needs.

Distance as the discovery mechanism

Perhaps the deepest assumption: consumers choose among stores they encounter spatially. The model has no concept of a shopper who finds you on a screen before ever passing your storefront—because in 1963, no such shopper existed.

The Huff model’s genius was its structure. Its weakness is that every parameter inside that structure is a guess.

Three Places Modern Data Supersedes the Classic

1. Observed mobility replaces assumed distance decay

You no longer have to assume how far customers will travel—you can watch how far they actually do. Anonymized mobility data reveals the true, empirical shape of a concept’s trade area: often lumpy, asymmetric, warped by highways, rivers, and commute patterns in ways no smooth exponent can capture. A store’s real capture curve might reach twelve minutes east along a commuter corridor and die at four minutes west across a river. The gravity model would draw a circle through the middle of both errors.

2. Machine learning learns attractiveness from analog stores

Instead of proxying appeal with square footage, modern forecasting learns what actually drives performance for a specific concept by studying its analogs—existing stores of the same brand and comparable brands, across hundreds of markets. The model discovers which combination of co-tenants, access, visibility, demographics, and competitive pressure predicts revenue for thisconcept, rather than assuming one attractiveness formula fits all retail. This is the approach behind Locate’s new-store sales forecasting: the output isn’t a patronage probability, it’s a revenue range you can take to a real estate committee.

3. Digital discovery breaks pure distance logic

The distance-as-discovery assumption is eroding fastest of all. According to Semrush data, “near me” keyword variations now total roughly 7.1 million US searches per month, and “near me” search volume grew 29% between Q1 2025 and Q1 2026, with intent-heavy variants like “near me tonight” (+41%) and “near me open now” (+38%) spiking fastest. That’s tens of millions of monthly store choices made through a ranking algorithm, not a windshield. Proximity still matters in those results—but so do reviews, photos, hours, and category relevance, none of which appear anywhere in a gravity equation. A shopper’s consideration set is now assembled digitally first and spatially second.

Know the Classic, Use the Modern

The right posture toward the Huff model isn’t dismissal—it’s graduation. Its core logic, that demand splits among alternatives in proportion to appeal and inversely with friction, is still the skeleton inside every serious location model. What’s changed is that each parameter can now be observed or learned instead of assumed:

• Assumed distance decay → empirical travel patterns from mobility data.
• Square footage as attractiveness → machine-learned drivers of performance from analog stores.
• Spatial discovery → spatial plus digital demand, measurable down to the trade area.
• One symmetric customer → segment- and daypart-specific behavior.

There’s a practical reason to keep the classic in your head, too. When a modern forecast surprises you—a candidate site scores far higher or lower than intuition suggests—gravity logic is the fastest sanity check: what’s the competitive set, what’s the real friction to reach it, what makes it more or less attractive than it looks? If the modern model can’t explain its departure from the gravity baseline, keep asking questions.

From Probability Surfaces to Revenue Forecasts

Huff gave the industry the right question: not “whose territory is this?” but “what share of this demand can a store here actually capture?” Sixty years later, the honest answer requires data he never had—observed movement, learned attractiveness, and digital demand—translated into the number a growth leader actually needs: projected revenue for a specific site, defensible in front of a board. That’s the standard Locate builds to, with the analysis and the brokerage execution under one roof. If you’re still evaluating sites with square footage and a guessed exponent, talk to us about what the modern version of the same question looks like for your brand.

Common Questions

What is the Huff gravity model in retail?

The Huff model, published by David Huff in 1963, estimates the probability that a consumer at a given location will shop at a given store. It divides a store's attractiveness (classically its square footage) by a power of the travel distance or time to reach it, then normalizes across all competing stores. The result is a probability surface: for every block in a market, the share of demand each store should capture.

How is the Huff model used in site selection?

Analysts place a candidate store on the map, assign it an attractiveness value, and let the model split every neighborhood's demand between the candidate and its competitors based on relative attractiveness and distance. Multiplying captured probability by population and spending per capita produces a rough revenue estimate and a probabilistic trade area, which is useful for screening markets and visualizing competitive pressure.

What are the main limitations of the Huff model?

Its inputs are assumptions, not observations. Square footage stands in for attractiveness, one distance-decay exponent stands in for all customer behavior, and every consumer is treated as identical. It also assumes people discover stores by proximity, which digital search increasingly breaks. Modern approaches replace these assumptions with observed mobility data and machine-learned attractiveness from analog stores.

Is the Huff model still worth learning in 2026?

Yes. It remains the clearest mental model for retail competition: demand splits among alternatives in proportion to appeal and inversely with friction. That intuition underlies even modern ML forecasts. Knowing the classic helps you interrogate any model's outputs; you just shouldn't underwrite a lease with square footage and a guessed exponent when observed data exists.

What has replaced the gravity model for retail forecasting?

Machine-learning revenue forecasting built on observed inputs: anonymized mobility data that shows how far customers actually travel to stores like yours, analog-store analysis that learns what makes a location perform, and demand signals such as local search volume. These models keep the gravity intuition but estimate every parameter from data instead of assuming it.