Geofencing location data analysis
Top problems we solve:
- Precision & accuracy
- Preventing App publisher malicious tactics
- Pre-bid real-time analysis
Target your ideal customers accurately and report results with confidence and visuals that back them up.
The goal of this article is to share with you the methodology Meteora uses to understand and deploy successful geofencing campaigns, explain the differentiators, and expose false realities you’ve most likely been pitched in the past.
Understanding the core challenges of location data:
For our purposes, we will not be discussing beacon mobile pings as this method is severely limited by whether a device owner has your beacon app installed on their phone. Instead, we will focus purely on device locations coming from the Adtech ecosystem (Publishers). The flow of control is shown below:
We live in a world filled with countless types of devices all of which have technical limitations of their own as it relates to sending accurate data into the ad eco-system.
The speed of obtaining maximum precision is as long as 30 seconds from the time an app is launched or a webpage is rendered. Data entering the ecosystem in sub-16 seconds generates a risk of polluting your campaign with often inaccurate coordinate samplings. GPS signals depend heavily on a clear line of site and cellular signal. Devices struggle with interference generated by dense buildings, concrete, or “Hive” blocking generated by too many devices in close proximity to one another. (I.E. Stadiums, Downtown skyscrapers, Subway stations).
It must also be noted that GPS strength can also be hindered by each publisher’s specific app settings. Since each of our campaigns has access to win ads on hundreds of thousands of apps/ millions of websites, it is extremely important to blacklist or optimize bid weight for imperfect GPS consensus. This does not mean that said app developers are “shady” by any means, as they might lower accuracy settings to preserve the battery life of a device that is accessing their service or for other various reasons.
In an OpenRTB protocol-based exchange, a Demand Side Platform only has optional fields to further their understanding of each individual location a device is seen from. Each of the following fields translates critical information to Geofencing success “Estimated Accuracy”, “Acceleration/ Speed”, and “Timestamp for when the coordinate was sampled”. The movement speed of a device is critical to differentiating between drive-by freeway traffic and someone standing at your cash register. Estimated accuracy is often submitted in feet and not meters by ill-informed publishers!
Is your geofencing platform accounting for these nuances?
A second look at cleaning pre-bid location data:
Latitude and longitude data is depicted by decimal values such as:
33.812092, -117.918974 (Disneyland). These Disneyland coordinates are precise down to 6 decimal places OR 4 inches. This is considered heavily favorable precision. For fun, let’s explore how far coordinate precision can actually go:
|Decimal Places||Precision||In English|
|1||10 kilometers||6.2 miles|
|2||1 kilometer||0.62 miles|
|3||100 meters||Roughly 328 feet|
|4||10 meters||Roughly 33 feet|
|5||1 meters||Roughly 3 feet|
|6||10 centimeters||Roughly 4 inches|
|10||10 microns||A speck of pollen|
|12||0.1 micron||Flew virus size|
|15||0.1 nanometers||You can send ads to an atom|
It’s critical to distinguish the accuracy of each devices coordinates vs the precision of said coordinates. Showing false precision can lead others into believing accuracy is great than it really is. You can visualize accuracy vs precision here:
Info: Accuracy vs Precision visualization
Scoring: Keeping track of apps, publishers, and historical accuracy.
Improbable locations often surface in the bid stream and thus need to be removed for optimum campaign ROI. Over time a historical score can be applied to each app/ publisher, while each score can and should manipulate your bid multipliers or suppression of said publisher. Invalid coordinates such as 0,0 tell us that an app has been designed poorly or an error in GPS processing by the device has occurred. This location points to the middle of the ocean, an unlikely target for your mobile ads. Out-of-bounds coordinates (Uninhabitable land/ water mass) should be removed pre-bid.
Another example includes publishers sending latitude data variables where an exchange request is meant to have longitude variables. By flipping the variables you can visualize the intended location that was meant to be sent to the exchange. Algorithms can easily pick up flip-flop variables when looked at on the publisher level. It is possible to look at these publishers in a way that relates them to “misspelled searches” on Google and buy those keywords in your campaigns to lower overall costs while maintaining target accuracy. If your geofencing platform cannot identify flip-flop publishers and use them to buy valuable device impressions at cheaper rates, you’re missing out.
Lastly, coordinates should be frequently analyzed using machine learning to discover and de-value bad publishers who bend or break their methodology over time. Often times bad actors will slightly manipulate their coordinates. The hope is for them to yield a higher frequency of impression sales when devices creep closer to highly populated areas often targeted by campaigns. Analyzing coordinates with extremely high volumes of devices in a very short time frequency either tells us that a significant event is underway (Super Bowl), or a publisher is hoping to trick your campaign targeting functions. Often times we see publishers showing hundreds of devices standing on exactly the same 10-meter coordinate.
Info: Un-natural city center traffic
Represent the true ROI of geofencing:
At Meteora we are as transparent as possible while leading buyers toward lucrative and measurable ROI. This is how geofencing reporting should be:
Using our reporting you are able to make both marketing and business intelligence decisions without the burden of radical scale or spending. True ROI can be calculated with even low numbers of impressions and conversions! Statistical relevance depends on your use case. The greater the data a campaign generates, the greater the accuracy of a model a data scientist might use. I.E: Creating AI that decides how likely a customer is to convert at a physical location if they clicked on your banner or not.
Final notes on geofencing and mobile ad units:
Both post-view and post-click methodologies apply to geofencing as it is a heavily branding-based channel. Mobile users are often not location static. Creatives need to be mindful of font size and calls to action to ensure quick and assertive hand-holding toward engaging with your ad. Mobile ads are often misclicked due to screen space constraints so it is extremely common for sessions to show up blank in non-log-level analytics software packages such as Google Analytics. Use HTTP level tracking to measure drop-off! This result is also common due to page load times being slow for landing pages that are not fully optimized by the advertiser. This causes the analytics tag to not fully load before a session spanning as long as even 10 seconds to be counted. 3rd party analytics tags are extremely dependent on load times/ internet connection and where they are placed in your landing page code. Please ask us for advice if you wish for us to offer tips to help your landing page perform better in a mobile advertising environment.