As explored in Man Vs. Machine - Does AI pose a risk to A&R Managers?, professionals working within the music industry are becoming more inclined to use analytics to assist with their decision-making. However, some of the most common measures of analytics and testing are limited in their predictive abilities. If the music industry is to fully draw upon the potential of big data, it must go beyond a reliance on retrospective statistics and look more towards predictive analysis.

We explore the limits of current data analytics used within the music industry, and present NoticeSound, a forward-thinking software solution that is breaking ground in predictive analysis.

Sales as a predictor

As record sales continue to decline, first week numbers are now more important than ever. These first week sales figures are able to provide a concentrated platform upon which artists can then build their entire sales and marketing, and form the backbone of their campaigns. In fact, for many artists, first week sales can equate to as much as 30% of total sales. Mark Mulligan, music industry expert and author of Music Industry Blog, believes that this upward trend is set to continue, with first week sales becoming an even larger focus for labels, in the coming years [1].

As first week sales are more important than ever, lead singles become equally so. For the majority of record labels, this allows only a small time frame for success. It follows that exploration of single data should be a useful measure for the future success of artists.

However, there are clear differences between consumer listening habits and buying habits. For example, in a single analysis by Coleman Insights, it was found that while both radio and on-demand streaming builds over time, downloads show a steady decrease after the first 1-4 weeks. The research used data from both the US Billboard’s Top 10 Digital Songs chart and their Top 10 songs for Radio exposure and On-Demand streaming. [2]

Considering the above, any data analysis conducted on downloads in the first month, would not be useful in determining song success over a longer period. The reason for this is simple - a reduction in purchases after the first month does not mean that consumers are no longer listening to the music. Instead it means only that new consumer numbers are reducing. Thus, while first week sales may hold increased importance for record labels, to recoup their outlaid finance, it is perhaps not a valid measure of longevity, or future success.

Herein lies a flaw in much of current music analytics. The discrepancy between consumer listening habits and buying habits, means predictive decisions based upon initial sales data, is risky.

Indeed, it fails to place sufficient value on listeners who are currently engaging with music at any given time, which in the streaming age is more important than ever before. There was a 257% increase in streams between 2013-2015 and this accompanied a 23% decrease in downloads. Evidently, streams are not an additional option to downloads, but the replacement [3]. Of course, as sales continue to decline, the ability to predict based on this data only weakens further.

In a second study by Coleman Insights, it was found that even hitting the Top 10 on the Billboard Digital Songs chart, is not a good measure of success [4]. They found only 1 in 6 songs within the top 10, later become big radio hits. Further, songs that actually became hits are the ones that remained within the Top 10 for at least five weeks.

These findings speak clearly. Current sales data, especially over the first few weeks, is a poor measure of continued success. This is despite their increased importance to record labels and this period being the focal point for sales and marketing campaigns. What is worrisome is that sales data still holds strong as a key source of insight for the music industry, even by individuals who are reluctant to engage with newer forms of data analysis.

Viral songs as a predictor

If sales are non-valid predictors of success, perhaps a better metric is viral material? By definition viral music is popular, so on the surface, this would appear to make sense. However, this does not hold true.

74% of songs in the top ten streaming charts stay there for less than a month [5]. Moreover, songs that prove to have the most success long-term are those that typically take a longer time to rank highly in streaming charts.

Viral status may come with success, but only for a very limited time period. This can be at least partially explained by the research showing that social influence drives likeability [6].

User engagement as a predictor

Evidently, broad statistics such as single sales, first week album sales, and viral popularity, are not accurate indicators of success. For music in the modern age, there are new, more complex forms of analysis available.

Spotify

These newer forms draw upon big data and have some supporting evidence for their predictive prowess, such as the ability of Spotify’s algorithms to correctly predict Grammy winners [7]. These predictions were made on the basis of users’ streaming activity, which considering their user base of 140 million, is an impressive feat [8].

However, it is important to remember that while Spotify was able to predict award winners, this was from a sample base of music already successful in the market. While it shows that Spotify has a measure of reliability to its data capture, it speaks little to its ability of predictive analysis for music without existing exposure.

Shazam

Shazam is also commonly used as a tool to understand how listeners are engaging with music. Shazam is a popular application which, based on audio, can tell its users the title and artist from a huge quantity of songs. Much has been said about its capabilities in predicting hits and the software has gained notoriety for its comprehensive data capture.

Despite this, even Shazam has clear limitations in its predictive benefit, as Matt Bailey has reported [9]. Firstly, Shazam is only used when its users are unfamiliar with songs. Just because a song is falling on the Shazam charts, this doesn’t necessarily equate to diminishing popularity. A song could very well be enjoying increased success, but simultaneously become ‘shazamed’ less because it is more well known. Similarly, while users are unaware of the songs when they ‘shazam’ them, these songs are already in the market and likely to be familiar across a wider population.

Together, these two factors make it difficult to know if Shazam is at all capable of predicting the success of music. While songs may test high in Shazam, this could (due to the brand’s popularity), serve to give further exposure to these songs. It then becomes impossible to determine what percentage of a song’s success is attributable to its organic likeability, and how much is instead from the increased exposure resulting from the app.

Ultimately, Shazam does have use in exploring which songs are testing well and are likely to increase in popularity over the coming weeks. However, as users need pre-exposure to a song in order to input it to the Shazam framework, it is not useful in its predictive analysis.

Analytics for the future

The above has demonstrated that music testing and analytics has a valuable role to play in decision-making for record companies. That said, when it comes to predictive analysis, there are clear flaws within the existing methods. In order to increase the chance of success for music, it is important to use a tool that can remove social bias. This tool must also be capable of creating data for music that is not already exposed to a public audience. NoticeSound is one example of technology that has been developed, specifically for this purpose.

NoticeSound

NoticeSound allows its users to make informed, data-driven decisions, and fully benefit from the capture of big data. It provides a track management system for users to screen and discover artists, and the testing capabilities to control for the influence of social bias.

As a result, users are presented with objective consumer feedback, which is refined to specific target markets. It was created specifically for use within the music industry and designed to discover music with the highest rates of success. Moreover, the software allows users to test music amongst specifically designed groups. This allows effective replication of market conditions, and reliable predictions for consumer uptake.

Essentially, NoticeSound is a platform for users to validate their creative decisions with data, which is designed to help record companies in their acquisition decisions, as well as with the discovery of trends and potential hits.

Concluding remarks

While record companies are slowly warming up to the idea of digital data informing their decisions, streaming platforms were designed for that purpose. These platforms were born within the climate of big data and were created to harness its power. However, while many forms of data analytics exist, there is a gap in the market for a service that is able to discover popular music. NoticeSound represents a novel solution and opens the door for record companies to start controlling the ever-evolving environment that they exist within.

Sources