How To Think About Machine Learning & Marketing in the Music Industry
By Kristin Westcott Grant
One of the goals of the industrial revolution was to have machines simulate physical tasks in order to build cars and machinery more efficiently. Fast forward a few decades. The goal of artificial intelligence, also known as AI, is to simulate any mental task. Machine learning is arguably one of the most important subsets of AI, because it affects all other fields within AI. In any industry, you have a pattern or a model that you know to be true, you make a prediction, and then you update your pattern based on the result. This represents the learning process of machine learning. The introduction of this technology into industries like music, online dating, online publications, video-sharing and sports is becoming vital to each organization's competitive sustainability.
The more data you have, the better the accuracy of your machine learning algorithm. In the music industry access to data presents a big challenge, but once you have it, the question becomes, how can you use and manipulate it using machine learning?
Let us quickly recap why access to data is such a big challenge in the music industry.
In this metaphor, Sony, Universal and Warner Music Group are represented as “The Majors”. Online streaming platforms like spotify and apple are represented as The Streams. Consumer data from streaming platforms is represented as “gold”. The Majors and the Streams sit knighted at the round table. The Streams rule the island of consumption and as a result control access to the gold. The Majors are granted access to the gold because they own most of the island of content and a share of land on the island of consumption. The independents own a smaller portion in the land of content, and as a result they have to get their gold from Robin Hood. Also known as direct-to-fan platforms that provide consumer data to artists such as Pledge Music, Hive or Superphone.
The question is, once you have access to the gold, what do you do with it and how can you maximize its value?
Have you ever been hunting for gold? How do you go about it? You get a sieve, you dip the pan into the water and you pull up a whole bunch of dirt, mud, rocks and stuff that you do not need. However, somewhere in there is your gold, otherwise known as your actionable data. Actionable data can be rate of collections, follower change, plays per user, or collections from a playlist. The term, collection, on Spotify refers to when someone listens to a track, presses the add sign and adds the song to their library. Tapping into engagement metrics increase the likelihood of reaching potential super fans. Focusing on actionable data will allow a label to make more targeted business decisions across all verticals while driving revenue. According to Goldman Sachs, streaming data will drive over 100 billion dollars worth of revenue in the music business by 2030. That is a whole lot of streams, a whole lot of royalty payments, and a tremendous amount of data.
The motivations of the person holding the filter affect how the filter is shaped and how the filter is shaped affects what type of gold you get. For example, the primary goals for a streaming platform like Apple, Spotify, Google Play or Deezer is to turn non-paying subscribers into paying subscribers. A major label’s goal is to create and market hit songs while turning passive fans into superfans, similar to the rabid Beliebers or Drake aficionados of the world. These motivations have a direct impact on how the gold is filtered.
Industry players who have access to the gold are now competing with the help of their filtration ability. How can we design our sieve to get the gold we need, when we need it, to drive a higher return than my competitor.
But music is not the only industry working to create the perfect filter. In fact,one should pay attention to the advancements being made in other industries because of the parallel applicability to the music industry.
Automated Marketing Tools
In 2013, Amy Webb went on The TED stage and spoke about breaking the online dating code. She amassed 72 data points of her perfect man, everything from jewish, to athletic, wants 2 children, is an adventurer, to even his appreciation of things. It was extremely important for Webb that her perfect man had an appreciation of a good spreadsheet. She then prioritized each data point, breaking them down into two tiers and gave each characteristic a score between 1-100. Amy then built a scoring system, if her perfect man scored 700 points she would send him an email. If he scored 900 points she would have a phone call and if her potential perfect man scored 1500 points, that meant there was long-term relationship potential and they could go on a date.
Amy started getting all of these amazing matches, except there was one problem. These matches weren’t liking her back. Amy had forgot to analyze the competition. She scraped the top profiles on the dating site, in music this could be compared to analyzing the social or streaming profiles of similar artists. She analyzed her competition’s photo, humor, tone, voice, communication style, average length of description, and time between posts. Amy’s profile ended up becoming the highest ranking profile on the dating site. Soon after, a man scored 850 points, which she hadn’t seen before. Three weeks later they went on a date. A year and half later they got engaged and two years later they had their first child. Now, If an algorithm can be used to narrow down your choices for a life long partner, then it is reasonable to believe that an algorithm can be used to find a fan that is guaranteed to spend $100 on your artist per year.
Similar to how Webb broke down her perfect man into 72 data points, so can an artist break down the characteristics of their potential super fan. For example, Beyonce’s base of super fans could be broken down as female, ages 27-34, with a common purchasing pattern of buying Huggies Diapers because they want to show that they are good mothers. If you rank these points, give them a score and run them through a scoring system. It is then possible to target the fans that are most likely to engage with your artist. For example, if your fan scores 700 points, you send a targeted facebook ad, if your fan scores 900 points you send them an email and if your fan scores 1500 points you send them a personalized email with a free concert ticket.
The ability to find your true fan suggests that automatic marketing capabilities are not only possible but in our near future. However the types of marketing actions that a label would engage in would differ based on the stage of the artist and genre. This is based on the assumption that fans types differ per genre and engagement tactics are different depending on the stage of the artist. However, online dating algorithms are not the only industry that provide interesting parallels to music.
Recommendations & Predictions
Assume that you have access to granular level engagement data from streaming platforms like rate of collections, rate of replays per user, all by a zip code level granularity. How could you use this information to not only target market but predict the likelihood of a potential superfan. One of the best industry parallels to consider is YouTube’s Recommendation algorithm.
Youtube, fueled by their parent companies artificial intelligence division, called Google Brain, has successfully accelerated their recommendation capabilities through a series of micro-improvements.
For example, roughly four years ago, YouTube made its first major improvement to their recommendation algorithm when they decided to value the amount of time users spent watching a video higher than the number of people who had clicked on a video. In one fell swoop, creators who had profited from misleading headline and thumbnails experienced their view counts decline. All of a sudden, higher quality videos which were directly correlated with long watch times came to the forefront. As a result, over the next three years, watch time on YouTube grew 50 percent year over year.
Google Brain learns independently by picking up on less obvious patterns at an accelerated rate. This technique is called unsupervised learning.
Another micro-change caused by Google Brain, was the choice by YouTube to recommend shorter videos for users on mobile apps and longer videos on YouTube’s TV app. Google brain picked up on the notion that varying video length by platform would result in higher watch times. In music, this could be compared to varying advertising length based on the platform assuming shorting ads for mobile and longer ads for desktop.
In 2016, Youtube launched 190 changes and is said to be on pace to release 300 more micro changes by 2017. The implementation of Google Brain has increased the time people spend watching videos on YouTube by 70%. In the present, the ability to parallel this type of rapid growth, would most likely only be possible in companies like Sony, Universal and Warner. Simply due to the increased access to granular level data due to the size and scale of repertoire and shareholder status in Apple Music and Spotify.
When Joe Lacob became owner of the NBA’s Golden State Warriors, he adopted a data driven strategy. Lacob and his team analyzed player behavior across the NBA and identified the number of three point shots taken as being “Market Inefficient.” They concluded that roughly the same amount of shots were being taken from just inside the 3-point line as shots taken outside it. Therefore they built their strategy around the notion, that if their players, particularly Stephen Curry moved back a few inches from the three-point line before shooting, it would improve their point scoring average by 43%.
In this case they were focusing on overarching trends and patterns similar to consumptions trends across streaming platforms. Here are a few examples of music related trends worth considering: what types of playlists are people listening to, at what period of time do people listen to your artist, and what holidays or political events are most likely to affect music or playlist consumption and when.
With a data driven strategy, Jacob Lacob took the Golden State Warriors, a team that hadn’t won an NBA Championship since 1975, to winning against the Cleveland Cavaliers in the 2015 Championships.
Insightful Data vs. Actionable Data
Buzzfeed is a social news and entertainment company. Buzzfeed invented an internal proprietary metric that curates articles based on reader preference. They do this by measuring the rate of shares over time, within the first two weeks of an article’s release. Buzzfeed decided that a reader sharing an article was more valuable than a click. Have you ever clicked on an article or played a song and walked away from your computer? What the act of sharing an article or saving a song to your spotify collections shows is a greater level of engagement in comparison to a stream or a click. Arguably, this is helping to ensure a return on your investment.
Taking it one step further, a share is more valuable than a click and adding a song to your library is more valuable than a stream. By calculating the rate of shares over time or the rate of collections over time, you’re not only making sure that the consumers you are targeting are engaged but that they are growing significantly over time.
Buzzfeed is now estimated at approximately 1.7 Billion dollars, and has 7 billion monthly content views.
With the sheer volume of streaming data increasing year over year, the ability to enhance and fine tune marketing capabilities in the music industry is endless. It comes down to access to the data that you need and the software capabilities to intelligently process and act on the basis of that information. The moral of the story is - the difference between a successful and not successful label moving forward into 2030 will come down to each company's ability to pair their their human capital with intelligent software.