By Eliot Van Buskirk of Evolver.fm.
We stopped by Next Big Soundâs New York office for a chat with its data scientist Victor Hu, formerly a mathematician for the U.S. Department of Defense and âstats whizâ for the New York Yankees, to see how the company gauges music popularity for clients including Billboard, which relies on Next Big Sound for one of its charts.
I pulled some key nuggets from our interview to try to distill how Hu does what he does, so that we mere mortals (i.e. normal people for whom high school calculus memories is the closest we normally get to this sort of thing) can try to grasp it.
Public and private data mashed together
âI take all of this rich data that we have â itâs essentially three years of any kind of data you would want to know about an artist, both public and private,â said Hu. âIâll look at all the major social media networks combined with private sales data, radio, and concert data â basically for every artist. Weâve been tracking this for a long time, and my job is to take [the data] and glean intelligence from it â turn it into insights that we can recommend to our customersâ¦ for example, the Billboard charts.â
By public sources, Hu explained that he was referring to Next Big Sound accessing APIs from a number of sources (the site lists Facebook Insights, Google Analytics, iTunes Upload, Twitter, Facebook, YouTube, Vevo, Wikipedia, Last.fm, ReverbNation, SoundCloud, Pandora, Vimeo, Rdio, MySpace, and Instagram).
But what private data sources is he talking about? Well, itâs some of the same numbers Billboard relies on for its other charts, just analyzed differently, and mashed against other sources.
âPrivate is something like sales data,â added Hu. âThereâs no way to get access to that unless you have a relationship with them, which we do. Weâre getting it primarily from the labels themselves. The labels who are our customers, they want to see all of their sales numbers in conjunction with the social media numbers, and so they give us this data, so we can put it into our dashboard, and they can see it and slice it in any way they want to.â
âA&R Guysâ Use It
âYeah, that [A&R guys wondering who to sign] is definitely one of the areas that we target,â said Hu. âItâs a sort of a reverse look-up. Instead of taking an artist that you know and finding out what their numbers are [which costs $20 per artist], you say, âI want to find artists with these particular numbers.ââ
It Watches YouTube Replace Radio as a Sales Driver
âWe were asked to do a case study for one of our clients on a particular artist,â remembered Hu. âHe was doing very well with â his sales numbers just took a spike. It wasnât in the presence of strong radio play, so thatâs very unexpected, given how artists normally progress â if you come out with a hot song, thatâs what triggers a lot of your sales. They couldnât figure out why that was, and they had us look into it more. We followed the rabbit hole down, and it turns out it was because he had released a new video on YouTube right around the time of his spike. That in conjunction with his appearance on an award show â you could see the clear shape of his increasing digital sales come right after the release of his music video, which is not, I think, intuitiveâ¦ to see YouTube tied so closely to sales was, I think, very encouraging.â
It measures the acceleration of acceleration of artist popularity
âEric [Czech, Next Big Sound chief architect] came up with a way to measure how quickly artists are accelerating,â said Hu. âFirst we have the Next Big Sound Social 50, and thatâs the top artists for social metrics. The harder question is, âHow do you identify the ones that are up and coming?â Thatâs the Next Big Sound chartâ¦ Weâre looking at the top social media metrics and the acceleration.â
Is that the derivative we remember from calculus?
âItâs the derivative of the derivative,â said Hu.
So itâs the rate of change of the rate of change?
âExactly,â said Hu. âWeâre just fitting a second order polynomial and then seeing what that coefficient is.â
It favors new players
âOne thing thatâs interesting about the chart: We donât want to feature the same people over and over again. If we feature someone one week, and the next week heâs still accelerating, you canât just keep them on there forever, you want to rotate it. Since weâve been doing it for so long, eventually you reach a point where youâre almost running out of artists. Thatâs why we narrowed it down from a long list to a smaller list â because weâre trying to feature new, fresh artists every time, it wouldnât make sense to have the same names over and over again.â
To labels, success has many faces â but not that of the Facebook Like
We wondered what constitutes success for Next Big Soundâs label clients this days. It used to be easy to measure: vinyl (or cassette, or CD) sales. But now, it takes many forms.
âEverythingâs still tied to some sort of tangible sales outcome, but itâs expanded, in that itâs not just physical or digital sales,â said Hu. âI think thereâs a lot more focus on the 360 model now [where the label's contract grants them a piece of all, or most, of an artist's revenue, including tour receipts]. If you can acquire a fan via having a music video, or getting them to listen to your songs on Hype Machine or Spotify or whatever, even though they might not be paying for that as much as you would want, once you convert the fan, you get them to a concert to buy merchandise, so I think thatâs the crux behind a lot of the longer-term thinking: not just focused on the album sale anymore.â
At this point, I mentioned that one problem with the so-called âattention-based economyâ is that you canât pay rent by paying attention to your landlord. So do the labels view tweets and Facebook Likes as wins?
âItâs very much âWhy should we care about Facebook Likes?â Thatâs a lot of what we focus our energy on â this blog post that we did about the actual impact of social media on sales was very well received, because thatâs what people care about: âDo I even care that someone is tweeting about me â what does that translate to in terms of sales?ââ
At this point, I mentioned that Iâd recently heard from someone who knows that âthe kidsâ are Liking and unfollowing bands on Facebook just to seem cool while not actually wanting to hear from that band in their feed.
âWe found consistently that Facebook Likes is not a good indicator of sales,â responded Hu. âItâs much more important to have people coming to your site. Pageviews is a much, much better indicator than Likes. If you think about how you would normally interact with someoneâs Facebook page, it makes sense. If you want to know more about them, youâre going to keep hitting their website over and over again, whereas whether you Like them or not â youâre not going to go right from Liking them to buying stuff.â
He also mentioned that Wikipedia visits âare a big driverâ of music sales â much moreso than Likes.
âRight before sales spikes and after some sort of event,â he said, âthe biggest response is always in Wikipediaâ¦ in many ways, itâs a proxy for a Google search, because thatâs where they go first.â
So, there you have it: If you want to know whoâs popping, check the pageviews on Wikipedia â a non-profit website that does well in Google search results â and not on Facebook, which the stock market says is worth $62 billion, and which is counting in part on those Likes to power its upcoming social search feature.