“Data-Driven Thinking” is written by members of the media community and contains fresh ideas on the digital revolution in media.
Today’s column is written by Khoa Pham, product and data analyst at PlaceIQ.
Location, location, location.
It’s a very complex creature. From the Mercator projection to the development of the first GPS, the representation of people’s movements across space has been a topic of research for hundreds of years. From defense and public transportation to mail delivery routes, our increased understanding of location has created and altered many companies and industries.
Advertising is no different.
I think we can all agree that location is increasingly becoming an important part of the mobile ad space. Understanding location is the first step toward having an in-depth comprehension of audience. As our understanding of how location applies to our immediate use-case matures, we see we still have much more to learn in terms of the most efficient way to use this powerful tool.
Geofences And The Independence Assumption
From the offset, geofences seemed to be a very logical way to represent and expand a place of interest. If I wanted to target a McDonald’s with a geofence, l could just draw a big circle around it, assuming that people around the general area will be more in the “McDonald’s mindset” than those who are farther away.
Though this statement may seem true on its face, we see where the assumption breaks down: Not all McDonald’s are created equally. A stand-alone McDonald’s in the suburbs of New Jersey has a dramatically different radius of influence than a storefront McDonald’s in the middle of Times Square.
Assume we don’t pick all McDonald’s, but pick only a subset that are located in suburbs. The argument for using the geofence again breaks down:
• What other stores are around the McDonald’s? A location centered in a suburban strip mall with a Walmart and Home Depot attached has a much different area of influence than a location that is surrounded by homes.
• What are the demographics of people around the McDonald’s? A location in an old affluent neighborhood is much different than a McDonald’s in a yuppie young suburb filled with apartment complexes.
• What types of transportation are around the McDonald’s? A location close to a busy highway draws different customers than one that is along a local bus line.
There are so many externalities around the way we can subdivide and represent different McDonald’s that we quickly realize the independence assumption breaks down. Locations are not independent, and geofences are not dynamic enough to capture these intricacies. What surrounds a location is just as important as the location itself.
This inherent interconnectedness is what makes representing a “simple” location-based signal very difficult.
Beyond The Geofence
Though geofences are often limiting, there are some instances where they are quite powerful, including concerts and other large events that draw a crowd. Geofences are, however, often misused. Advertisers can fall victim to an oversimplification that does not serve their case.
So if geofences aren’t the answer, what is?
A more comprehensive way of representing location could be to tessellate, or section off, segments of space into mutually exclusive bins and analyze everything that occurs within those bins. This would allow for the analysis of anything of relevance that falls close to a certain point of interest while still allowing for the comparison of two different locations without jeopardizing the frame of reference.
This allows for the incorporation of more metadata around each individual McDonald’s to make a more informed decision around where and where not to serve a particular ad. A very targeted campaign may want to only serve ads within the confines of the store. Others may want to widen it to the store and the parking lots surrounding the store.
Finding The Right Balance
By knowing what else is around each store, advertisers can dynamically choose to expand or contract how wide they want to target using a systematic approach:
• Use a wide radius if there is nothing around the store
• If the store is in a densely populated area, only target the polygon of the actual store
• If the store is close to other dining areas, expand the targeting during lunch and dinner times and shrink the targeting during all other times
• Only target apartment buildings close to a store and not single-unit homes
This way, we can deal with the uniqueness of each store without muddling in the details that aren’t important.
While there is no single, optimal size for targeting, and a solution that works well for one client may not work well for another, location is becoming a driving force within mobile advertising. Understanding location has never been more important.