Partner Article

How SoundOut predicts the hits with 80% accuracy

SoundOut fuses crowdsourcing and artificial intelligence to predict the hits

Did ‘Sorry’ by Justin Bieber reach number 1 in the charts because the song was exceptional, or was it because the most famous male artist in the world was the singer?

Where would ‘I took a pill in Ibiza’ have peaked in the charts if it was sung by your mother?

Until now the relationship between artist, song appeal, record label and commercial success has remained a mystery that generations of music executives have sought to unpick (and their hit rate remains patchy at best). But a 12month artificial intelligence project by SoundOut has resulted in a prediction model for the US music industry that can predict commercial success of any given track in any of nine different music formats with over 80% accuracy.

SoundOut currently provides research around the hit potential of new songs to many of the major US labels (and to thousands of independent artists) but bringing data science and machine learning to the party has transformed the value proposition of their service.

To achieve this feat, SoundOut collected information about 1,300 songs, in nine different radio formats, that had been released in the US over the past three years. The information collected included everything from the song writers, the producer, artist social media popularity, the performance of all their past singles, the label they released under and many other factors.

These datasets were then supplemented with the detailed predictive insights gathered on each song prior to it becoming well known – this proprietary data set was derived from SoundOut’s 2.5 million strong panel (reviewing on www.slicethepie.com) including market potential, passion, reviewer profiling, detailed semantic analysis and genre analysis.

Finally, SoundOut data scientists added details of the actual commercial success of a subset of the tracks into their machine learning algorithms to train sophisticated prediction models for commercial success in each format. These prediction models were then tested on the remainder of the tracks to see how well they would have predicted actual commercial success.

The results were compelling, and demonstrated that when tested with SoundOut’s 2.5 million strong panel, these models can now predict a top 10 hit in almost any format with over 80% accuracy.

This is not an academic exercise – SoundOut is in the business of predicting bestsellers for all the major record labels and, increasingly, major high street retailers as well.

David Courtier-Dutton, SoundOut CEO said “We’ve always known that the intrinsic appeal of a track was only part of the story, but until recently did not have the tools to map the complex relationships between all the elements that conspire to create a hit. The fusion of over a million crowdsourced reviews and ratings with social media metrics and machine learning has finally made this possible – we look forward to making this unique insight available to the music industry in early 2017”

The perfect hit ‘recipe’

What was particularly revealing was that the process also revealed the key factors that combined to produce a hit in any given format. The machine learning process makes it possible to identify which are most (and least) important factors in creating a hit single in different genres [illustrative charts available for pop and rock]

The ability to individuality separate out every element of the recipe for any given genre offers some tantalizing options for artist managers and record labels – they could test a demo of a new track with SoundOut and then run it through the model multiple times to decide which artist should record it.

It could also be used to match the perfect track to a new upcoming artist or old artist seeking a comeback to optimize the chance of their success.

It could even be used to analyze the relative strength of the current top 10 line up to decide the optimum time to launch an attack on the charts.

SoundOut is currently crowdfunding, aiming to raise £600,000 in order to capitalise on recent successes in predicting best sellers in fashion retail and to drive international expansion – for more information see: www.seedrs.com/soundout.

This was posted in Bdaily's Members' News section by Kenny Mac .