Welcome to a new corner of our blog, titled Under the Hood! This new series of articles will explore how the mobile location data ecosystem works, and what is required to turn that information into a precise understanding of physical world behavior. We’re going to geek out a bit on technology in this series and share a lot of technical insights, but it won’t require you to have a PHD in Data Science to understand. At NinthDecimal, we’re committed to continuously sharing what’s really possible with the industry, and that’s the goal of this new series. So, are you ready to look Under the Hood?
There has been plenty of discussion about “bad” or inaccurate location data across our industry. Some estimates have this applying to as much as 60% – 80% of location data. The examples are many, with perhaps the most common example being the use of geographic centroids. Rather than passing location data based on a device’s actual location, an app may send a set of latitude and longitude coordinates that are the centroid of a specific geographic area. These can be the centroid of a country, a DMA, a city, or even the center of the consumer’s home zip code. Most of the technology platforms that have worked with mobile location data for several years have developed sophisticated techniques to filter and scrub inbound location data. These filters and techniques are designed to address accuracy. Does this latitude and longitude accurately represent where this device is in the physical world, or is this just an entirely made up location? Examples like the centroid one have become easy to spot, and therefore experienced companies have largely solved the accuracy problem. But what about the Precision Challenge?
NinthDecimal has built a data platform predicated on a people-based approach to managing location data. This is a higher set of precision than alternative place-based approaches which largely rely on tiles or geo-fences to understand the physical world. To that end, we’ve developed Precision Filters to complement Accuracy Filters.
What is a Precision Filter?
I’m sure everyone has had the experience of opening Google Maps on your phone and seeing your location identified by a large blue dot. Over time, that blue dot gets smaller until it represents a pretty precise location of where you are.
This is a natural function of the location services built into your smartphone’s OS “waking up”. As it wakes up, it kicks in multiple location processing tools like A-GPS, Wi-Fi, and Cell tower triangulation to create a more precise location signal. This is now expanding to include the device’s accelerometer, compass, bearing, speed, and other signaling tools.
In late 2014 and early 2015, NinthDecimal worked with several location-based apps to further analyze the horizontal accuracy of location signals. Not surprisingly, the “Blue Dot Effect” above was reflected in the data where some data had a wider range of error. But what was interesting is how quickly that error range shrank over time. Examining billions of impressions with timestamps and horizontal accuracy, we found that 84% of impressions reached a precision of under 10 meters within just 8 seconds.
The interesting element to this is that the vast majority of location data is generated via ad calls today (also known as bid-stream data). Ad calls from mobile apps typically happen every 30 seconds. If the first ad call happens when a smartphone is first waking up, the location data associated with that device will suffer from the “Blue Dot Effect”. However, subsequent ad calls will be roughly 30 seconds into the device’s location services kicking-in and will benefit from a far more precise location signal.
To illustrate this impact, we ran a simple test by keeping a log of device usage. This allowed us to record a device’s actual location at a given point in time, and then compare that against the location generated by an app and recorded in our data platform for that specific device.
The first example shows the device when used at home. The first data point as the device was in wake-up mode placed the location over 175 feet away, essentially three houses down the street from where the device actually was located. However, the second data point generated forty-eight seconds later placed the device exactly at the correct location and within the correct residential parcel.
In this second example, the device was at Disneyland. In the morning, when the device hadn’t been in use,the first location data point was incorrect by 1,000 feet. However, as the device was used continually in the park (for anyone who has used the Wait Time app for Disney rides, you’ll be quite familiar), the device recorded a highly accurate location. In the second image, the device was on the Riverboat Ride around Tom Sawyer’s Island. Being placed in the middle of the water was not an error!
So how does this nuance of mobile devices benefit the mobile location data ecosystem? This is where NinthDecimal’s Precision Filters make a significant difference. When looking at mobile data, periods of activity and periods of inactivity can be seen for each device. These patterns reflect consumer usage and can identify when a device is in wake-up mode or not. The first data points after a period of inactivity reveal the start of a mobile session. And because these are the first data points, they are subject to the “Blue Dot Effect”. By contrast, subsequent data points during that same active mobile session are not subject to this effect.
NinthDecimal developed a unique approach to filter out and remove those first low precision data points. As data comes into our platform and by using a function of time, activity volume, and other factors, we identify whether that data is the start of a new mobile session or is coming from within an already active mobile session. The end result is NinthDecimal is only utilizing precise location data for building audience segments, real measurement results, and accurate insights for marketers.
These are different from the accuracy filters the rest of the industry has built. Those are identifying whether an app is sending good location data. And those inaccuracies tend to be in miles, not tens of meters. Precision filters look at good location data (e.g. it is where the device is located) and then filter on the precision of each impression, not just the accuracy of the source. The combination of Accuracy and Precision filters create a proprietary data profile unique in the industry.
This is just one of the techniques we use to filter for precision, and we continue to develop new approaches in partnership with the industry. For example, the latest mobile operating systems now include horizontal accuracy as a data setting which apps can utilize. NinthDecimal works directly with leading location-based apps that process horizontal accuracy to continuously tune our platform for precision and develop new approaches for the industry. After all, small variations in a location have massive implications on context and therefore marketing services. Stay tuned for more innovation from NinthDecimal.