How Do External Factors Affect Ad Performance
Editor’s Note: This is the first in a series of posts about Flite’s most recent hackday. The hackday theme was “Big Open Data Sets”, and several different teams had one day to create a demo incorporating that theme.
At Flite, we have no idea what type of content is displayed alongside our ads, which run millions of impressions all over the web. The engagement and interaction rates are mostly based on the creative and messaging. But this may not always be the complete picture. The context of the real world in which the ad runs also may have a great amount of influence on these rates.
For our hackday project, we wanted to pull in data sets from many sources and measure the numbers against the engagement and interaction rates to see if there are any correlations within our Report Studio.
We started the process by looking for the apporiate data sets that would be apllicable to our project. Some of our challenges included the following:
- Free or public data is hard to find, especially daily or historical data
- Data in various formats made it hard to parse
- Google APIs are inaccessible, or not available/rate-limited
The stock index (S&P 500) ended up providing us with the cleanest data set. We added a UI in our Report Studio that would enable users to input a stock symbol and see whether we could correlate that data to our metrics on the ad impression.
We also explored how we could correlate open data if we had more granular geolocation data available in the Studio. We collect geolocation data at the ad level and that can currently be found under the Measure tab in the Ad Studio; however, that data is not available in Report Studio (yet). If it were available, we could tell the user the geographic location where the most impressions were served and information about that location, like headlines, events or weather that happened there on that day.
Impression data could also be represented on a map using geolocation to see if local news events could possibly have a major influence on the numbers.
Other data sets we considered included
- Trends (google, twitter),
- News Articles (page views for news articles),
- Weather (by region),
- Geolocation info could also correlate with that
- Map against custom data (like your own KPI or internal data sets)
This was a very informative project and expect more soon.
Team Scallop: Grant Lee, Joel Antipuesto, Matt Thomas, Mekuria Getinet, Paul Krohn