Billboard placement analysis, Philadelphia
Jun 2021 - Oct 2021 · Delivered
Analyst and developer
A geospatial analysis ranking Philadelphia billboards by proximity to competitor outlets, so advertising spend lands within sight of the right customers.
- Python
- Pandas
- Google Maps API
- Geospatial Analysis
Overview
Which billboards should a fast food chain buy in Philadelphia, if the goal is to place advertising within sight of a competitor’s customers. Published as an article by the client, with the code public.
The data
Philadelphia’s outdoor advertising dataset from OpenDataPhilly, roughly sixty competitor outlet locations from the Google Maps API, and traffic data alongside them.
Method
Plot outlets and billboards on a common coordinate system, compute the distance from every billboard to the nearest outlet using the Haversine formula for great-circle distance, then vary the threshold radius and count how many billboards qualify at each.
Findings
| Radius | Billboards within range |
|---|---|
| 100 m | 15 |
| 250 m | 109 |
| 500 m | 259 |
| 1 km | 688 |
Billboards within 100 metres are the strongest choice, because the target is specifically the competitor’s customers. Widening to 250 metres is defensible since it increases available inventory, though those sites sit further from the outlet. Beyond 250 metres the argument stops working.
Complexities tackled
The analysis is only useful if it says no. Six hundred and eighty-eight billboards qualify at a kilometre, which is a number that flatters the media buyer and helps nobody. The value of the piece is the recommendation to ignore most of them, and that only carries weight because the radius sweep shows what each threshold buys.
Choosing the right distance measure. Straight-line distance on latitude and longitude treated as a plane is wrong at city scale and quietly biases results by direction. The Haversine formula gives great-circle distance on a sphere, which is the correct measure for this and costs nothing to use.
What I would add
The article closes with four extensions worth doing: which way traffic faces the billboard, whether the detour to the advertiser’s own outlet is convenient, whether the billboard is digital and can vary by time of day or weather, and what other advertising shares the location.
Stack
Python, Pandas, Google Maps API, open government data.
Working on something similar?
Tell me what you are building and what is in the way. I will tell you honestly whether I am the right person for it.
Last updated: