Blog·

What is a Huff model in retail site selection?

An empty shopping cart standing alone in a parking lot on a sunny day.

The Huff model is a gravity-based probability model that estimates how likely a consumer at a given location is to patronize a particular store, based on the store's size relative to its distance from that consumer and the pull of competing stores in the same trade area. David Huff published the original formulation in the early 1960s, and it's still the backbone of a lot of gravity model work in retail site selection today, even as the inputs analysts feed into it have gotten a lot richer.

The basic mechanics

A store's drawing power goes up with its size (square footage is the usual proxy) and down with the distance a shopper has to travel to reach it, usually with distance raised to some exponent reflecting how sensitive shoppers in that category are to travel time. Run that calculation for every store a household might reasonably choose between, and you get a probability of patronage for each one. Sum those probabilities across a grid of residential points and you've got a spatial estimate of where a given site's customers are likely to come from, and how much of the surrounding demand it should capture versus what leaks to competitors.

Where the model earns its keep is in trade area delineation. Instead of drawing a fixed ring or drive-time polygon around a candidate site and assuming everyone inside it shops there, the Huff model lets the trade area be probabilistic and overlapping. Two competing grocers three miles apart don't each get a clean half of the territory between them. The model assigns shares based on relative pull, which tracks how shoppers behave in practice far more closely than a hard boundary would.

Calibrating a Huff model well depends on getting the demand side right, and that's usually where estimates fall apart, not the gravity math itself. Analysts default to census tract population counts or household totals because that data is free and already in GIS. But a census tract can run half a square mile of single-family rooftops next to a quarter-mile of garden apartments, and those two pockets don't generate the same per-capita demand for a discount grocer or a QSR. Lumping them into one tract-level population figure smooths over exactly the variation that determines whether a candidate site performs once it's built.

That's the gap between a gravity model and a genuinely useful probability of patronage estimate. The model itself doesn't care how you define "demand" at each point in space. It just multiplies whatever weight you give that point by the attraction term and divides by distance. If the weight is a stale tract average, the output is a stale tract average run through a nicer-looking formula. If the weight reflects what's built on the ground, housing type, unit density, and how much of the surrounding land use is residential versus industrial versus already commercial, the output starts to look like something you'd bet a lease on.

Trade Area Profiling reads high-res imagery on demand for the blocks around a candidate site and returns a numbered brief on housing type mix, density, and land use breakdown. That feeds the demand side of the Huff model with a read closer to what's actually there, rather than a tract average that was last accurate when the census was taken.

A note on the exponent and the rest of the calibration

Most of the debate in gravity model literature is about the distance-decay exponent and how to fit it to observed shopping behavior for a given category. That's real and worth getting right, but it's a separate problem from the one above. Even a perfectly calibrated exponent can't fix a demand layer that treats a dense apartment corridor the same as a few scattered ranch houses on acre lots. Get the built-form read right first. The exponent tuning matters more once that foundation is solid.

FAQ

Is the Huff model still used, or has it been replaced by newer methods? It's still widely used, often as one input alongside drive-time analysis and competitive gap analysis rather than as a standalone answer.

What's the difference between a Huff model and a simple trade area ring? A ring assumes a hard boundary and uniform capture inside it. The Huff model produces overlapping, probability-weighted trade areas that better reflect competition between nearby stores.

Does the Huff model need store-level sales data to work? No. It can run on size and distance alone, though analysts often calibrate the exponent against observed patronage data when it's available.

Pull a built-form read on housing type, density, and land use for your next candidate site before you run the numbers.

Get started

← Back to the blog