Behind the Scenes

Stats, Data, and Rock & Roll

Posted on

I think Facebook thinks I’m a bot.


William Steffey and I are putting some serious focus on promoting his music this year. Running ads on Instagram is the first part of that. A Facebook business account comes with a lot of data, so surely I can use my statistical background to get us some sort of advantage?

The standard strategy for promoting music would be to figure out what bands and artists William sounds like and then reach out to them. But he doesn’t really sound like anyone. Really! We ran a survey and the answers were all over the place. If you don’t believe me, head on over to Spotify and then leave your opinion in the comments! Please?

For now, we’re targeting an eclectic mix of music fans based on the survey results. And we’re getting lots of feedback about how the group is responding to the ads. Age, gender, location. But nothing about which artist(s) they liked that put them on our list to begin with. As results come in, I want to be ready with some Bayesian estimation tools to narrow down the best audience for our ads.

Bayesian estimation is all about conditional probability. What is the likelihood of A, given that B is true? In this case, to give an example, What is the likelihood that someone who responded favorably to our ad is a fan of David Bowie or U2 or The Killers or – you get the idea.

Most people running Facebook Ads campaigns on a small scale aren’t former actuaries, so they helpfully provide potential audience information in the form of graphs and summaries. I want everything as granular as possible though. I can get there with a bit of data entry work that I’m happy to do, but I think I ran over my limit today. The page is telling me there’s no data available.

Fingers crossed that I can convince tech support of my humanity.