FuturePulse just launched a music forecasting platform – but needs the industry’s help

Daniel Johansson wants your assistance. And your experiences. And your expertise. And your ideas. But Johansson doesn’t want something for nothing. This is a trade-off. He has something for you.

He is a researcher and analyst at Sweden’s Linnaeus University as well as at the Spotify-backed company, Soundtrack Your Brand, which he represents as one of the leads in the EU-backed, €3m-funded research project, FuturePulse.

Other music and tech stakeholders in FuturePulse include IRCAM, Musimap, Sonar, CERTH, Athens Technology Center, Playground Music, and BMAT.

Their collective aim is to create a technology platform that accurately predicts trends on social media and streaming platforms – and, as a result, the commercial performance/potential of artists, tracks, albums or genres – up to 12 weeks in advance.

Johansson stresses that this is a public platform; all research and findings to date are available via published scientific papers (which can be seen via the FuturePulse website).

More significantly for music industry types, an alpha version of FuturePulse’s technology is available to peruse and use online today through this link.

This is a raw set of tech tools, the real purpose and value of which will be shaped by being used by music industry experts, in the field, testing its limits and sharing their experiences.

However, Johansson claims that the tech can already “predict popularity – and other things – for genres, artists and tracks, based on both open and closed data”.

Here, Johansson explains to MBW why you should be taking a closer look at FuturePulse and how it can help provide perhaps a sliver of certainty in uncertain times…


What was the headline goal behind starting FuturePulse? What did you feel was missing from the world of data – and how it is used – in the modern music industry?

In 2015, Playground Music Scandinavia, BMAT and CERTH approached me with an idea of creating a new kind of statistics platform, based on predictive analytics, that would combine open and proprietary data (from labels, aggregators, streaming services, social media etc).

The problem as we saw it was, although we have so much data, it is still hard to extract true insights and knowledge from that data, insights and forecasts that can support stakeholders in the music industry when making important decisions.

“the holy grail of the music industry today, recorded or live, is to be able to detect trends in social media and on streaming services very early, and use those signals, as well as historical data, to predict what is going to happen with certain genres, artists and songs on specific markets.”

Daniel Johansson, FuturePulse

It is one thing to see all these graphs on different analytics platforms, and it is another to have a combined overview of both open and confidential data, to be able to create insights and knowledge through the data we are seeing; we need more meta-level analysis of the data.

Also, the holy grail of the music industry today, recorded or live, is to be able to detect trends in social media and on streaming services very early, and use those signals, as well as historical data, to predict what is going to happen with certain genres, artists and songs on specific markets. The headline goal from the beginning was simply to do the research and innovation necessary for such a platform to come alive.

We managed to get approval from the EU program Horizon 2020, for a €3 million research project, where 70% was financed by the EU, and 30% by the music industry. The project has been running since fall 2017 and we now have the first public alpha demonstrator of a platform that indeed can predict popularity – and other things – for genres, artists and tracks, based on both open and closed data.


What have been your most important findings in terms of the key factors in predicting the popularity of a track or artist?

That it is sometimes much easier than one thinks, but that sometimes it is almost impossible. It is one thing to identify a trend that has already started, or is steady; it is a totally different thing to predict a ‘future pulse’ that has not yet started to happen.

To make it more clear, identifying artists and tracks that are trending on streaming services and in social media is quite straightforward today. Since we have so much open data today, compared to just a few years ago, it is just a matter of tracking playlists, charts, social media, etc. That, of course, is a technical challenge, but from an innovation point of view it is just a matter of processing power, handling API limitations, and storage capacity.

“It is one thing to identify a trend that has already started, or is steady; it is a totally different thing to predict a ‘future pulse’ that has not yet started to happen.”

The difficult part is when you have the ‘cold start’ problem. For example, a totally new artist that is going to release a debut single or album. Based on earlier experiences, the music industry has always tried to forecast this, but it has mainly been an activity based on ‘gut feeling’ rather than data.

Now, a system that gradually learns from past success stories, in different genres, in different countries, on different platforms, based on data, could work as a complement to the human experience and knowledge that still are so important.


Can you give a brief outline of your methodology?

The FuturePulse platform automatically collects both open and closed data from Facebook, Twitter, Spotify, Apple Music, Deezer, YouTube, SoundCloud and others, and uses this data to provide predictions on the number of streams, mentions, likes, views, etc. for up to 12 weeks.

Data is often updated on a daily basis, but for different functions there are different time frames. Areas covered include:

  • Genre Popularity: Spotify Top 200 charts are collected weekly every Wednesday, and we calculate genre popularities and predictions at the end of that day, which means that updated genre popularities are available every Thursday;
  • Artist Popularity: Data is collected from all sources weekly. Each artist is tracked on a different time point of the week to ensure that we distribute the requests throughout the whole week period. Artist popularity calculation starts on Saturday and runs until Sunday so we can present the updated artist popularity score, and predictions, every Monday;
  • Track Popularity: We gather and update track popularity scores and predictions on a daily basis;
  • Label tracks (from distributor): We insert new releases every Saturday (as new products are usually released on Fridays);
  • Audio Descriptors: Audio analysis of new tracks takes place on Sundays;
  • Playlist data: Data is collected from all sources daily, weekly, or every two weeks based on the number of followers of the playlist. In that way, popular playlists are tracked more often than less important playlists. Each playlist is tracked on a different time point of the day or week to ensure that we distribute the requests throughout the whole period.

How accurate have your test cases been and what sort of thing can you now predict, and across what sort of time range?

The more data you have, the better the forecasts. We have a public version and a full-featured version of the platform, where we have loaded the first catalogue from our partner Playground Music where we use restricted data (streams) and audios for audio analysis.

That has allowed us to use advanced aggregated data, which makes it easier to create predictions for back catalogue. But, the only way to be able to somewhat forecast the popularity of new releases is to combine manual work, knowledge and experience, with data and forecasts from other similar releases.

“the only way to be able to somewhat forecast the popularity of new releases is to combine manual work, knowledge and experience, with data and forecasts from other similar releases.”

That is also one of the reasons why we have opened up the demonstrator in a public alpha version, with limited data, to the music industry, to get as much feedback as possible about the algorithms and functions. It is like we have created Frankenstein’s monster, now we need help to see what it can do.

There are indeed some very interesting predictions that we have seen that the platform can do, e.g. predict how social media will behave when artists are posting news, predict track popularity during the corona pandemic, as well as forecast the overall interest in a larger catalogue.


Is the benefit of an accurate prediction model in giving companies a level at which they can benchmark and hopefully improve performance, or is it to help them be ready for the level of success they will pretty much inevitably achieve?

There are several benefits that the FuturePulse technology can provide: combined statistics for open and closed data in one place, predictions of genre trends on different markets, forecasts of popularity of tracks and artists that in the end could be combined to understand catalogues, recognition levels of songs etc.

But, to be honest, we feel that we have created so much interesting stuff behind the scenes, and that the best people to really understand how this technology can be used in the best way are the music professionals.

“We have created so much interesting stuff behind the scenes, and the best people to really understand how this technology can be used are the music professionals.”

So, in a way, we have developed a lot of new potential technology, which we are demonstrating in this open, limited alpha version, but the full application of these models and technology I would say is still in the future.

Right now we are conducting a large scale pilot testing of our demonstrator platform, which anyone is invited to test at http://app.futurepulse.eu.


You are focused mainly on independent labels and artists, why is that?

When writing the proposal to the EU some years ago, we discussed how this project best could support the European music industry.

The largest music industry companies already have resources and competence in-house to create this kind of deeper data analysis based on machine learning. In a sense we want to level the playing field by making the same kind of advanced analytics available also for small companies, labels, small venues, festivals and even DIY artists.

Of course, the results from this project can be used by both majors and indies, but our hope is that FuturePulse will make it possible for smaller actors to create insights and knowledge based on big data.



Can you talk a bit about how the platform can be used to shape live strategies?

Among other things we have a specific model called Venue Rank that can be used to identify which venues that are the best for an artist when it comes to social media and communication. The model has been published at the ISMIR conference, and we really have just started to scratch the surface on how this model could be used for the live industry.

Sadly, because of the situation in 2020, we have not had the possibility to test this model fully together with live actors, but hopefully this will happen in the future.

Also, being able to analyse and predict genre trends is quite important for live actors, as well as being able to forecast the popularity of an artist in the coming months.


Finally, what is the plan to disseminate this technology throughout the industry and help labels, managers etc. learn how to use it and make the best of it?

On October 9th we held an open webinar. We will also hold a webinar at c/o pop later in October, as well as an innovation challenge, kind of a hackathon, around predictive analytics together with Sonar in November.

All the information is on our web page, and we welcome anyone to try out our public demonstrator.


Further Information on partners in the FuturePulse consortium:

BMAT: is a music innovation company with a mission to index all music usage and ownership data. They monitor and report music globally across TVs, radios, venues and digital to help artists get paid for their plays. In FuturePulse they provide daily airplay data and integrate the back-end services on a platform with visualization tools.

CERTH: is one of the leading research centres in Greece and listed among the TOP-20 E.U. research institutions, with the Information Technologies Institute (ITI) focused on Informatics, Telematics and Telecommunication Technologies involved in the consortium. In the project it is developing the tech for artist and genre popularity, as well as predictions.

Musimap: Musimap is an Emotional Artificial Intelligence (A.E.I.) company that provides automated engines to businesses in the fields of music, video, media, advertising, dating and e-commerce. In FuturePulse it has developed the technologies for audio analysis and track prediction.

Athens Technology Center: ATC is a research and development company from Greece that provides customized business IT solutions to different industries across the globe. It has built the backend of FuturePulse.

IRCAM: the Institute for Research and Coordination in Acoustics/Music, is a French institute dedicated to the research of music and sound. It is one of the world’s largest public research centers for musical expression and scientific research, and has been focused on audio descriptors.

Playground Music Scandinavia: is the largest independent label in the Scandinavian countries. It is one of the three user cases in FuturePulse.

Sonar: one of the largest music festivals in Europe, as well as the arranger of the Sonar+D music industry event. It represents live music in the project.

Soundtrack Your Brand: a streaming platform for background music, backed by Spotify and Rokk3r. It is the third user case in the project, representing the streaming sector and B2B music.Music Business Worldwide