Posted by Angela Diaco
Location-based ad targeting creates relevant content for consumers, better reach for brands, and more revenue for publishers. Although latitude/longitude points help advertisers make sense of their audiences on both a macro and micro level, adtech companies are often limited to using the inaccurate readings delivered to them from multiple sources. In fact, our work with leading adtech companies found that up to 90% of the location data appended to ad inventories is inaccurate.
In order to deliver the most actionable location -- whether it’s finding the correct user context for location-based campaigns, or locating a device’s position as the source of an E911 call -- a lat/long must be both accurate and precise if its to be trusted as truly representative of a user’s exact location. But what does a lat/long need in order to be “correct”?
Accuracy is a measure of how close a lat/long data point falls in relation to a user’s true position. With regard to location, Precision means exactness—the number of decimal places in a lat/long. (You can read more about these two terms at the full Accuracy vs. Precision post here)
It's important to never take a single location signal at face value, so a hybrid positioning system is needed. Skyhook scans, processes and acquires a massive volume of data encompassing billions of geolocated access points, cell towers and IP addresses. When examining the correctness of location data, we request a number of fields in the data set. These fields ensure the resulting Personas are as accurate and precise as possible.
Source. In the adtech industry, location data is usually sourced by IP address, but it is often incorrectly positioned by IP location providers. Native location sourced from users’ mobile devices, however, is preferable because it typically uses GPS and Wi-Fi as sources.
Recency. The time at which a lat/long was seen can heavily influence it's accuracy. For example, our system devalues location samples that are more than 5 seconds old, as devices can move significantly in that time.
Reliability. Too many decimal places in a lat/long indicates the location may be fabricated. Our system also devalues lat/longs with 7 or more decimal places, because that level of precision requires differential correction methods not found in consumer mobile devices.
Skyhook generates user Personas by mapping location samples to real-world venues. In doing so, we make use of additional data sets that come with location logs our customers submit. As illustrated in the visual above, the 4th decimal place in a lat/long pinpoint users to an 11.1 meter block. We require four decimal places of precision in lat/longs so that we can determine when a user is within a venue. Any fewer decimal places, and a user could be down the street from a venue, somewhere else in the same city, or in another country altogether.