Foot traffic data in site selection: how to use it without over-trusting it

Foot traffic data, sometimes called mobility or visitation data, estimates how many people visit a given place, how often, how long they stay, and roughly where they come from. Behind a clean monthly visit count sits a sample of mobile devices that opt into location sharing, then scaled up to stand in for the wider population. The number on the dashboard is a modeled estimate.

That distinction runs through everything below. A reported visit count has already passed through three modeling steps: a panel that samples devices, attribution that ties those devices to a specific place, and weighting that corrects for the sample. Each step adds useful information, and each one can introduce error.

Pointed at the right question, mobility data is among the strongest signals a real estate team has. Its sweet spot is places that already exist and already draw visits, where there is genuine behavior to observe.

• Validating existing and operating sites. For a store that is already open, observed visitation is a real-world check on your assumptions. You can hold modeled demand up against what people actually do and catch a trade area that looked strong on paper but never drew traffic.
• Benchmarking analogs. When you know a handful of stores well, mobility data describes their visitation pattern, so you can look for candidate sites whose surroundings resemble your best performers instead of your weakest.
• Understanding trade-area shape. Visit-origin data shows where customers actually come from. It often contradicts a tidy radius and confirms that catchment follows roads and travel time.
• Reading competitors and co-tenants. Visitation to nearby rivals and anchors shows where demand already concentrates and how a corridor fills and empties across the day and the week.

These limits are reasons to read mobility data carefully, with the same skepticism you would bring to any modeled estimate. Trouble starts when a team treats an estimate as a measurement.

• Panel and sample bias. The device panel behind the numbers is never a clean cross-section of the population. Some ages, incomes, and behaviors show up too often and others too rarely, and the scaling that corrects for this is its own model. Two vendors can report different counts for the same place because they use different panels and weighting methods.
• Point-of-interest error. Visits are attributed to a place through its mapped footprint. When a polygon is wrong, overlaps its neighbors in a dense block, or sits inside a mall, the visits land on the wrong tenant. Small and newly opened locations suffer the most.
• Coverage gaps. Sample density shifts from one geography to the next. Busy urban corridors are covered well. Rural and low-traffic areas can run thin enough that month-to-month swings are mostly noise.
• Visit attribution. Counting a nearby device as a visit, and inferring where that device sleeps at night, depends on dwell-time thresholds and home-location guesses. Pass-through traffic, employees, and stacked retail all blur the line between a customer and an artifact.

The limit that matters most for expansion is structural. Foot traffic data describes places that already have foot traffic. A greenfield pad on a road where your store doesn't yet stand has no visits to observe, because there is nothing there to visit.

This is the central tension in growth-stage site selection. The locations you most need to evaluate are exactly the ones mobility data cannot measure directly. You can read the visitation of nearby anchors, competitors, and the surrounding area, yet you cannot observe visits to a store that doesn't exist. Greenfield decisions force you to model expected demand from the surrounding trade area, then validate it against real behavior once the doors open.

So a serious process treats foot traffic as a calibration and validation layer rather than the prediction itself. The estimate of who can reach the site, who lives and works inside the drive-time trade area, and how much of that demand competitors already absorb has to come from a model. Mobility data then sanity-checks the model wherever observable history exists.

The dependable workflow uses mobility data to tune and test your model. Four steps keep the trustworthy parts of the data doing trustworthy work.

• Start with the analogs you trust. Pull observed visitation for stores you know well. Mark which ones the data describes accurately and which it gets wrong, so you learn where the panel is thin or the polygon is messy before you depend on it.
• Calibrate the model against known sites. Run your demand model on those same operating locations and compare its output against observed visits. Adjust the weights until the model agrees with reality on the cases you already understand.
• Apply the calibrated model to greenfield candidates. Now score the new sites, the ones with no visitation to observe, using a model you have grounded in real behavior.
• Validate after opening. Once a new site goes live, compare its observed visitation back against the forecast. Every opening sharpens the next forecast and shows you where the model still drifts.

A defensible site decision rests on several inputs at once: who can reach the site by drive time and walk time, the demographics inside that trade area, the competition and co-tenants already in place, and the risk of cannibalizing your own nearby units. A weighted score brings them together. Foot traffic feeds several of those inputs. It earns a vote in that score, weighted against everything else.

Geod is built around that idea. It can be used alongside observed foot-traffic data rather than selling a mobility panel of its own, and it produces an explainable score where you can see how much each factor contributed. When you have strong mobility data, you can read it against what the score shows. For a greenfield site with no history, more of the load sits with demographics, drive-time access, and competition, and the brief spells out why. You upgrade your read without surrendering the decision to a black box.

The aim here is narrow and worth stating plainly. Dedicated foot-traffic vendors will out-measure anyone on raw visitation, and Geod is meant to be paired with that signal wherever it helps. The work is turning the inputs that drive a decision into a traceable, committee-ready call.