If you have been pitched foot traffic data, you have probably seen a confident chart and wondered how much of it to believe. That is the right instinct. Mobile location data is useful, but it is an estimate built from a sample, and every step from phone to chart adds some error.
This guide explains where the data comes from, where it goes wrong, and which numbers you can lean on. It ends with a checklist of questions to put to any vendor, including us.
Where mobile location signals come from
A phone works out its own position from several sources, then passes that estimate to apps that have permission to see it.
- GPS (satellite positioning). The most accurate source outdoors. The U.S. government's GPS site says GPS-enabled smartphones are typically accurate to within about 4.9 metres under open sky. Accuracy gets worse near buildings, bridges and trees, and indoors 1.
- Wi-Fi and cell towers. Phones also estimate position from nearby Wi-Fi networks and cell towers. Apple says iPhone location services combine GPS, Bluetooth, crowd-sourced Wi-Fi hotspots and cell towers 2. Android's location service also blends signals such as GPS and Wi-Fi 3.
- Bluetooth beacons. Small transmitters inside a store can place a phone within a few metres, but only where someone has installed and maintained them 4.
- IP addresses. Industry guidance treats mobile IP addresses as generally poor for locating a phone 4.
Most location data used for foot traffic analysis comes from apps on people's phones that have asked for, and been given, permission to use location.
How precise is a single location reading?
Less precise than a map pin suggests. Several things limit it.
The user's settings. Both major phone platforms let people share an approximate location instead of a precise one. Android says approximate location is accurate to within about 3 square kilometres, while precise location is usually within about 50 metres and sometimes within a few metres 3. Apple added a per-app Precise Location switch in iOS 14. When it is off, the app gets only an approximate location 2. Approximate readings are fine for city-level questions. They cannot tell one store from the next.
The source of the reading. The Media Rating Council (MRC), the U.S. body that sets audience measurement standards, published location measurement guidelines with the IAB and the Mobile Marketing Association in March 2017. They describe GPS as capable of placing a user within 10 to 100 metres, and Wi-Fi within 30 to 100 metres, with GPS degrading badly indoors 4. Real phones are often in pockets, cars and buildings, not under open sky.
Precision is not accuracy. A coordinate with six decimal places looks exact. The MRC guidelines warn that the number of decimal places does not, by itself, tell you whether the point is right 4. Ask how accurate a reading is, not how precise it looks.
Typical sources of error
Indoors and dense areas
GPS struggles inside buildings and among tall ones, where signals are blocked or bounce off walls 1. This is a real issue for indoor venues such as arenas and malls. The MRC guidelines flag urban density, meaning many places close together and multi-storey buildings, as a known cause of error. They also note that geofences generally cannot represent multi-storey structures 4.
Small locations and stacked or adjacent stores
A visit is usually assigned by checking whether a device's reported position falls inside a boundary drawn around a place, either a circle or a traced outline 4. If a coffee shop shares a wall with a pharmacy, or sits above a bank, a reading that is off by even 20 metres (an illustrative figure, well inside the GPS range above) can land in the wrong one. Small locations also have fewer readings, so random noise has a bigger effect on the result. The larger and more isolated the place, the more you can trust the visits behind it.
Sample bias
The devices in any dataset are not a random draw of Canadians. They belong to people who own a smartphone, use the particular apps that share location, and opted in. A study presented at a 2021 ACM conference compared U.S. smartphone mobility data with North Carolina voter records. It found older and non-white voters were less likely to be captured 5. That is U.S. data, not Canadian, but it shows how the problem works. The MRC guidelines make the same general point: large app-based samples behave like a panel, and biases exist in who generates the activity and where 4. A good vendor tests for this and adjusts. No vendor can remove it entirely.
Panel changes over time
The mix of apps and devices in a dataset changes. Apps are added or dropped, people change permission settings, and platform rules shift. The MRC lists panel turnover, refresh and calibration among the factors that affect accuracy 4. If the sample grows between two years, raw visit counts can rise even when real traffic did not. Comparisons over time only work if the vendor accounts for this.
Visits are not purchases
Location data can show that a device was at a place. It cannot show what the person bought or why they came. On its own, it also cannot tell an employee or a delivery driver from a customer. Ask any vendor how they handle this. Treat any claim about spending based only on location data with caution.
Why a panel is a sample, not a census
No location dataset sees every person. It sees the subset of devices that share location through participating apps. The MRC guidelines treat this as sample-based measurement. They say vendors should check that the sample is representative, support any weighting with evidence, and disclose the error around projected numbers 4.
In practice, a raw count of devices seen at a venue is not attendance. Turning it into a headcount needs a projection, and that projection carries every bias and gap described above. A vendor who quotes exact visitor numbers without explaining the projection is giving you more certainty than the data holds.
Why indexes and change over time are more reliable
This part is reasoning, not a published finding. Many of the errors above affect a location in roughly the same way from one week to the next. The sample covers the same kinds of people, the building blocks GPS the same way, and the neighbouring store is still next door. When an error stays steady like that, much of it drops out when you compare a place with itself, or with a baseline measured the same way. Errors that shift over time, such as a changing sample, do not drop out, which is why the method has to account for them.
That is why relative outputs tend to be more dependable:
- Indexes. How much more or less a venue's audience shows a behaviour than a baseline, such as the provincial average.
- Shares. The proportion of visitors from within 15 km versus further away, or the share of traffic on each day.
- Change over time. This year's event weekend against ordinary weekends, measured with the same method.
These still carry error, but usually less than an absolute count. They also answer the questions most buyers actually have: did it grow, who came, and how does it compare?
This is how we report at Measured Reporting. Our data is aggregated and anonymized. It comes from a large sample of devices, never everyone, and it is indexed and normalized against a baseline. We report shares and indexes, not headcounts, and we measure visits, not purchases. You can read more on our Data and Privacy page and see what we deliver on our Capabilities page.
Questions to ask any location data vendor
A credible vendor will answer most of these directly.
Where the data comes from
- How are location signals collected, and from what kinds of apps?
- Did users opt in to location sharing, and how?
- What share of readings is precise versus approximate? Do you drop approximate readings for store-level work?
How visits are defined
- How do you decide a device visited a place? What boundary do you draw, and how do you handle multi-storey or adjacent locations?
- How do you separate a visit from someone walking or driving past?
- How do you treat staff and other regular non-customers?
Sample and bias
- How do you test whether your sample represents the population? Which groups are under-represented?
- How has your sample changed over the period I am asking about, and how do you adjust for it?
- Do you weight or project the data? If so, can you show the method and the margin of error?
What you report
- Do you report absolute counts, or indexes and shares? If counts, how are they projected?
- What is the smallest location or time period you will report on, and why?
- How do you check your results against a known source, such as gate counts or ticket scans?
Privacy
- Is the data aggregated and anonymized before it reaches me? What personal information, if any, do you hold?
If a vendor cannot explain their visit rules or how they handle sample change, treat their year-over-year numbers with care.
FAQ
Is mobile location data accurate enough to measure foot traffic?
For trends, shares and comparisons, yes, when the place is large enough and the method stays consistent. For exact headcounts, no. The data is a sample, and single readings can be off by tens of metres, more indoors.
Can location data tell which of two neighbouring stores someone visited?
Sometimes, but this is where error is highest. Stores that share a wall, or sit on different floors of one building, are hard to separate. Ask your vendor how they handle it, and be cautious with results for very small or tightly packed locations.
Why don't vendors just give me attendance numbers?
Because the data covers a sample of devices, not everyone. Turning a sample into a headcount needs assumptions about how many people the sample stands for, and those assumptions add error. Indexes and shares avoid most of it. If you need attendance, pair indexed findings with your own gate or ticket counts.
Does indoor location affect arena or mall data?
Yes. GPS weakens indoors, so phones lean more on other signals such as Wi-Fi. Large indoor venues can still be measured as a whole, but fine detail inside the building is not reliable.
Can location data show what people bought?
No. It shows that a device was at a place. It does not show transactions or what anyone did inside.
See it on your own location
The best test of any location data is a place you already know well. Request a complimentary Sample Insight and we will profile one of your locations, with the method stated, in about a week.