Instrumental: Using AI and Spotify data to find the next big thing in music

Computing speaks to Jonny Woolf, CFO / COO of the service which uses AI and Machine Learning to analyse music and help identify which bands and artists are most likely to hit the big time

In a dingy underground club somewhere on the outskirts of London, a group of late teens in fur coats and straw hats play a guitar-heavy set to a lively crowd. The air is heavy and humid, the floor sticky with spilled beer, the chances of escape should there be a fire probably zero.

It's worth it for the teens on stage because they know there are music industry reps in the audience. It's the responsibility of the Artists and Repertoire (A&R) department at labels of all sizes to get out there and discover the next big thing, wherever that might be.

It's a daunting job. In the 1980s there were scores of music venues in central London alone, with hundreds of unsigned bands playing almost every night. Today, over 20,000 new tracks from unsigned artists are uploaded to streaming service Spotify every day. No A&R department, even at the major labels, can possibly hope to listen to a tenth of it.

Computing's AI and Machine Learning Live! will be held on 19th November 2018 in Central London.

Where AI meets A&R
This is where AI meets A&R, and the company trying to facilitate the introduction is Instrumental. Jonny Woolf, the firm's CFO/COO, describes this as the music industry's 'Moneyball' moment, after the Brad Pitt film about a baseball team which transformed its fortunes through innovative use of data.

"There's no possible way the old-style A&R person can make sense of what's going on today without the help of tools," begins Woolf. "A&R is the R&D of the music industry. They spend around $2.8 billion collectively each year, and it's incredibly inefficient as you have no idea if someone is going to be a success. Most major labels rely on around one in 20 acts becoming successful. It's not a good hit rate."

Today, streaming services like Spotify have transformed the industry, with music lovers walking round with 20 million songs in their pockets, and record labels finally seeing some revenue growth after years of shrinkage.

"Over the last 20 years industry-wide revenue went from $26 billion, to $14 billion. But in the last three years it's climbed up again to $16 billion, because of the license fees paid by streaming companies. So we're seeing more investment, but the difficulty for A&R remains knowing what's good and what isn't.

"That's where we're trying to help. We're starting to make sense of that whole streaming world, and starting to try to predict the point where the artist is seeing traction and starting to be successful."

Instrumental: Using AI and Spotify data to find the next big thing in music

Computing speaks to Jonny Woolf, CFO / COO of the service which uses AI and Machine Learning to analyse music and help identify which bands and artists are most likely to hit the big time

Using Spotify data for fun and profit
Instrumental ingests Spotify's data through its public API, then applies its algorithms to understand that critical point of when an artist is on the cusp of real popularity. According to Woolf, it's all about enabling their clients - a mixture of record labels of varying size - to make contact with the artist or their management before the competition.

The firm is currently looking at playlists with at least 10,000 followers, citing that as the threshold for deciding that a list is getting traction. At the time of writing the firm is tracking 12,000 playlists containing 438,000 different artists, with over five million tracks.

"And we've been doing this for over a year. Even the artists themselves can't see that sort of historical data," says Woolf.

And that history is important, as the firm can use it to try to discern the difference between enduring popularity, or a band being a flash in the pan.

"Our AI classifies the artist into one of 13 genres, because Spotify has 32,000 and we had to limit that. We pull all the relevant data, and see how the top performing artists have changed over the last 30 days."

Britain's Got Data
Woolf cites a famous case study from the early days of the firm. When Instrumental's CEO and founder Conrad Withey was first building the tool's dashboard, he noticed that Britain's Got Talent finalist Calumn Scott was performing well across a range of metrics.

Withey got in touch with Scott, invested in him and agreed a development strategy. Eschewing traditional radio support, they devised a plan revolving around the analytics from the nascent dashboard.

Wilson was soon taken on by Capitol Records, and his stream count went from 200,000 to over 250 million.

It's this ability to spot an act on the cusp of major popularity that's key, says Woolf.

"When an act appears on the ‘Top 100' section of the Instrumental dashboard, that means it's been discovered by people. Usually they're picked up by a label three to six months later. In the music industry, if you can get to an act three to six months before anyone else, that can prove very positive commercially," says Woolf.

So is this the end of A&R departments, the latest in a long list of job functions to be replaced by robots? Not according to Woolf.

"Our belief is what we're doing is helping A&R teams to sort the wheat from the chaff. We make a series of recommendations, it's really a personal recommendation engine. But that doesn't mean the track is any good. We can tell you this is an act that's definitely got traction, but it's still up to A&R to have a listen.

"We wouldn't say to our clients this is to replace A&R, I would classify it more as something which fits into their workflow and makes their life easier."

Instrumental: Using AI and Spotify data to find the next big thing in music

Computing speaks to Jonny Woolf, CFO / COO of the service which uses AI and Machine Learning to analyse music and help identify which bands and artists are most likely to hit the big time

Nuts and Bolts Woolf explains that there's more to it than simply taking data from Spotify and ranking it.

"Anyone is able to pull data in from Spotify's public API. The proprietary piece for us comes from [data science firm] Fospha, who built managed platform for us from numerous DSPs [Demand Side Platforms]. We pull in social data from Instagram, YouTube, the charts and various other sources, and put that into our data lake. Then we apply our algorithms.

"That's kind of the secret sauce, the important bit, around how we classify and make sense of it. Once those algorithms are applied, it's all pushed into data output tables which our clients see."

These tables are viewed on Instrumental's dashboard, but the data is taken from a SQL database housed at parent firm Blenheim Chalcot, a startup accelerator which also manages Fospha.

"They use their own data centres at the moment. We may move the whole thing onto AWS at some point, but not yet, as we're more focused on data protection and security," adds Woolf.

And the technology is far from static, with Fospha's data scientists constantly looking at ways of improving the recommendation engine.

"We're also working on our own API," continues Woolf, adding that it should be ready by early August. "Having that data pipe that our customers can pull into whatever UX design they decide to build will be a great opportunity for us. We show our output of what we think our clefts would like, but it's better for them to choose themselves.

"We're also working on a mobile app, which will push recommendations. Clients will be able to swipe left and right to say whether or not they like it, and that should be out in September."

"Artists are always looking for ways to be more successful on streaming services. We can provide a long-term look at intelligence country by country within Spotify. If we can become known for spotting trends, we can sell that data too," says Woolf.

Our band in the underground venue in London may soon find themselves developing light data analysis skills in tandem with better guitar riffs. Perhaps that will be another topic for discussion when the A&R person approaches them after their set.