GPS

From signal to visits

GPS-based Foot Traffic is produced in two steps: detecting when devices in the panel stay in one place, and estimating from those stays how many people visit each place.

Detecting dwells

Dwells are periods when a device stays in the same place. They are detected as part of the aggregation described below: time spent moving between places, for example driving or walking, is left out, and only stationary periods count. A dwell belongs to the local day on which it starts, and only aggregated dwell observations reach Pine59.

Estimating visitors

For each place, our model uses the dwells on the place's polygon together with context about the place, such as its brand, its type of location and its surroundings. It has learned from a wide range of ground-truth visitation data how dwells relate to real visitor numbers for different kinds of places, and uses this to estimate the number of unique visitors to the place each day. Weekly and monthly figures add up the daily counts, so a person who visits on several days is counted once for each of those days. How the panel is scaled to the full population is described under Extrapolation.

Aggregation starts on the device

The panel signal is aggregated client-side, before it ever reaches Pine59. What we receive are aggregated observations, not raw location traces: no names, no advertising identifiers, no device-level movement histories. There is no personal data in our pipeline to leak, sell, or subpoena, because it never crosses our boundary in the first place.

Every published statistic is then further protected by the methodology below (extrapolation to population estimates, minimum-cohort suppression, rounding), so no output can be traced back to any individual panel member.

Why this matters in restricted states

Several US states have passed legislation restricting the sale of individual location data. Because the Pine59 pipeline never handles individual-level location data, with the aggregation happening at the source, we can continue to produce and sell aggregated visitation statistics in markets where raw location-data products cannot operate. The signal is privacy-safe by construction, not by after-the-fact scrubbing.

Stability over time

The panel of devices that provides the underlying signal naturally changes over time, for a variety of reasons. To keep estimates stable over time, it is essential to compensate for these changes, so that trends reflect real-world trends rather than changes in supply. Our modeling includes several techniques for ensuring panel stability:

• Market share. Extrapolation is updated dynamically to reflect the panel's real market share at any given time, so a growing or shrinking panel does not change visit levels.
• Signal density. Some sources only record a device's whereabouts at certain times of day, and this varies between sources and over time. We measure how much activity is recorded per device each day and correct for changes in it, so a change in reporting frequency does not read as a change in visits.
• Real patterns pass through. The corrections are measured on ordinary weekdays, so real differences between weekdays, weekends, holidays and seasons stay in the data.