Data quality starts with verifying the accuracy of a device’s location signal. There are many factors that we need to check and filter through to ensure that the lat/longs we use are geo-precise.

• Centroids: Centroids are lat/long coordinates that correspond to the exact center of a geographical area including countries, states, DMAs, cities and zips. When publishers are uncertain of where a user is located, they may pass centroid data by deriving it from user IP signals. GroundTruth recognizes centroids and removes them.

• Fraudulent Signals: Through an integration with 3rd-party fraud-protection provider, GroundTruth monitors its entire network and identifies and filters invalid traffic.

• Randomized Lat/Longs: A publisher may generate a randomized lat/long in order to pass any location information because they are rewarded for passing ad requests with location data attached.

• Carrier IP Detection: Carrier IP signals can be inaccurate because the location coming from this signal can be attached to the address of the carrier’s data server instead of the user’s true location.

• Outlier Signals: Sometimes we see the same device appear in two distinct locations like Sao Paulo and Rio de Janeiro within a short timeframe, which we remove.

2) How we accurately map places.

We use all of the most common methods of geo-fencing or mapping locations, each with varying levels of accuracy for measuring visitors to stores.

• Radial Fence by Address The most basic method of location-based geo-fencing is by placing a radius around a place’s address. Since addresses were built to provide directions to the location, many times the physical location is across the parking lot or on the street. (accuracy 14%)

• Radial Fence by Geo-Code This method places a radius around the center of a business or location. Since most locations are not circular, there will inherently be waste. How much waste depends on the size of the radius.(Accuracy 25%)

• Parcel Mapping Also known as property mapping and tax mapping are maps typically built to identify property boundaries and is a popular data source for industries such as real-estate. Parcels can often contain many businesses in the same plot. (Accuracy 50%)

• Polygon Mapping Polygons are traditionally formed by tracing the store or place boundary based on satellite images and are intended to capture the precise boundaries of a location. (Accuracy 100%)

3) How we verify real visits.

Once we’ve verified the location signal and matched it to a place, the final step is determining whether a visit has actually occurred. We look at a number of contingencies that enter into the complex equation of whether we deem a visit to be 100% verified.

• Store Hours: How frequently does the user visit and for how long? What are the store hours and when did the visit take place? We only count visits that occur during open store hours.

• Employee Status: Understanding employee status allows us to include or exclude employees.

• Dwell Time: How long did the user spend in the store? We use dwell time to filter for inaccurate visits.

• Speed: How fast was the user moving when we tracked their visit? If they’re moving faster than our speed threshold, we do not count the visit.

• Horizontal Accurecy: What is the percent accuracy of the lat/long we received? If the horizontal accuracy we see in our systems is not higher than our percent threshold, we do not count the visit.