Problem of GPS Targeting Scale
Solution: Geo Polygon Cookies
Assume that McDonald’s is keen to drive sales to its newly introduced burger “The Grand Mac”, consumed at specific times in a day. Physical stores are better commerce than online buy. Mobile ads using GPS targeting being one of the most effective advertising strategy, how would an ad network (let’s say Facebook Ads) achieve enough scale to muster up engagement?
(1) Of all the types of user identification possible— time of day, day of week, gadgets, demographics, interests, behaviours, current location, intents upto DNA mostly in no particular order— targeting using current GPS location of users co-relate (as one of the highest) to purchase in the ad tech industry, currently. Building behaviors using historically available GPS location come a distant second in this case.
(2) But finding enough buyers who are currently nearby stores, during meal hours, at acceptable accuracy considering an industry conversion rate of under 5%, is hard.
Geo Polygon Cookies
(1) Divide the flat earth of lat longs into polygons. Hexagons have the most coverage and map user real life movements closest. This is in-principle similar to honeycomb hexagons which use the least amount of material to hold most weight.
(2) Treat each polygon as a web page. Like how each web page is dedicated to specific interest, each polygon will represent a point of interest (such as cinema hall, sports stadium, peice of highway or just a no man’s land)
(3) As and when users move in and out of a polygon, drop timestamped cookies against the user
(4) How can all this help in scaling up? In addition to targeting specific polygons, McDonald’s can now get access to:
- users who / just moved out of select polygons,
- users who are expected to enter select polygons,
- users who visit/ visited/ going-to-visit similar polygons,
- users who frequent similar polygons historically,
- users who are currently at their home (polygons) for online ordering,
- users who are currently at their office (polygons) for online ordering,
A Reductionist’s approach.