Understanding if a device visited a place, brand, or type of store can be valuable context to have for your business. Companies use store visit information to build custom audiences for advertising purposes, to better attribute ad campaign spend, and to send contextual push-notifications in real-time. Unfortunately, accurately determining if a device visited a place can be a tough engineering problem to solve.
Dealing with messy GPS data, incomplete business listing information, and limitations in knowing where places exactly are located make visit attribution a complex problem. However, building a visit attribution solution remains a worthwhile endeavor since it enables you to enrich digital data with physical-world context.
Furthermore, building a visit attribution solution in-house allows you to tune the algorithm to your specific input data and specific use case which results in a better end solution for your customers.
We've outlined all the essential topics here, including:
If you’ve got the foundational aspects of this down already, we also have a store visit attribution technical whitepaper you can check out. if you’d rather start with the basics, read on!
Store visit attribution uses GPS location data from mobile phones with POI data to determine if a device visited a place, brand, or type of store. There are two main methods for attributing store visits, but the most accurate way is using precise POI polygons as geofences to truly see which mobile devices passed through a threshold.
The other popular method for store visit attribution is using a centroid radius as the polygon. While this can be easily done with any data point and basic geoprocessing tools, it often contributes to incorrectly attributed visits because a centroid radius is less precise than a building footprint polygon. As a result, GPS pings can be under or over-counted using this method of visit attribution.
Which store visit attribution method you choose will depend on the level of accuracy you need. For some organizations, a centroid radius generated from POI data will suffice, while others need the precision provided by building footprint geometry data.
The other popular method for store visit attribution is using a centroid radius as the polygon. While this can be easily done with any data point and basic geoprocessing tools, it often contributes to incorrectly attributed visits because a centroid radius is less precise than a building footprint polygon. As a result, GPS pings can be under or over-counted using this method of visit attribution.
Build custom audiences - With an accurate view of who is visiting your store (or your competitors), you can strategically plan inventory and marketing campaigns that align to consumer needs and expectations.
Better attribute ad campaign spend - Measuring visits to a specific location enables you to target the right people with the right ads at the right time, so you can optimize ad spending for the correct audience.
Send contextual push notifications in real-time - Geofencing is most effective when it’s accurate, so truly understanding when a person enters or leaves a specific place prevents you from targeting the wrong individuals or missing prime marketing opportunities.
A store centroid radius represents the distance "as the crow flies" from a building or property's centroid. Centroid radii can be generated using simple geoprocessing tools commonly found in GIS or BI programs.
They are ideal for geofencing the general area around a point and conducting proximity analysis, but are not the most accurate or strategic method for attributing exact store visits because they often lead to under- or over-attributing visits.
Building footprints are polygons that denote a structure or property's exact physical boundaries. They are the most precise method of visit attribution because they represent specific places rather than proximity, so GPS pings can be accurately attributed to exact POIs.
Geometry data that contains spatial hierarchy is especially useful for store visit attribution because it includes parent/child relationships (ex. when a store is inside a strip mall). With spatial hierarchy metadata, building footprint polygons can be used to attribute visits to stores within other locations, making them the most accurate method for visit attribution.
To download precise building footprint polygon data that includes spatial hierarchy information, get in touch.
Here is a breakdown of SafeGraph’s store visit attribution method. To read about it in more detail, read the technical whitepaper.
When dealing with GPS data, there are three primary prevalent issues that need to be addressed before correctly attributing store visits: GPS signal drift, spiking horizontal accuracies, and jumpy GPS pings.
To clean the GPS data, remove any non-stationary data and filter all horizontal accuracies above a tuned threshold. For any two points that are close in time, compute a speed between them and if the speed is too high, filter out the pings.
Next, try to determine where pings are coming from without using POI data for context. The key insight here is if you look at a series of GPS pings on a map with no places, you can generally figure out areas that a device could have visited.
After cleaning the data, do a first pass over it, creating clusters out of consecutive pings that are in large POI. Then do a second pass, creating clusters from the remaining blocks of unused pings using a modified DBSCAN. Finally, save the clusters and discard all unused pings to ensure you are using the most relevant pings for your analysis.
With these clustered pings, you can now begin to analyze the data with geospatial context. Simply perform a geospatial join between the clusters and building footprint polygons. You should make sure to add a buffer around the cluster to account for any horizontal accuracy uncertainty of the GPS pings.
The final step in accurate visit attribution is to apply machine learning models that determine the most likely place that ping cluster visited. This is especially helpful for areas where multiple places are located closely together.
Models can be developed using logic related to the time of day visits are recorded, as well as what type of business it is. For example, a cluster of visits between a retail store and a bar at 11 pm would likely indicate they are from the bar, not the store.
Today’s top retailers and advertisers are leveraging geospatial data in their store visit attribution workflows. Here are some examples of ways leading data science teams are thinking about visit attribution.
The outdoor advertising industry has historically lacked good data upon which to select placements that reach ideal audiences. Without quality data, the industry also struggles to help brands measure campaign effectiveness.
With building footprint polygons and anonymous location data derived from mobile phones, Billups was able to complete the customer journey from exposure to in-store visit, transforming OOH into a performance-based media.
When joined with GPS data, the precise geofences increase the accuracy of detecting store visits when compared to using store centroids or geocoded street addresses for the store location.
Media Storm found it challenging to determine from geolocation data whether a mobile advertising ID (MAID) had visited a store without accurate data on where stores were precisely located. Without exact building footprints for the stores, Media Storm’s audiences would include irrelevant MAIDs resulting in wasted ad-spend.
Using brand information and NAICS codes (categories) for a place, Media Storm was able to quickly identify store locations of its clients and those of its clients’ competitors, as well as the exact building footprints of those places, to more accurately create location-based audiences and reduce inefficient ad spend.
When done correctly, visit attribution is a gamechanger for retailers and advertisers looking to optimize ad spend and strategically plan campaigns. To learn more about how SafeGraph data can be used for store visit attribution, contact our data experts for a free demo.
