Deezer CEO Explains How New UMG-Backed Streaming Royalties Model Works

Last week, French music streaming service Deezer joined with the Universal Music Group to roll out what they called an artist-centric music streaming model, which they said was “designed to better reward the artists and the music that fans value the most.” It’s the result of a six-month partnership announced in March that promised to examine the current “pro-rata” streaming royalties model, in which artists and labels are paid according to their share of streams out of the available pool of revenue generated by streaming services. They aim to identify a new way of paying out that revenue, at a time when streaming service catalogs have exploded to north of 200 million tracks and fraud and streaming manipulation have proliferated on platforms.

The artist-centric model, which Deezer says will begin rolling out Oct. 1 in France for UMG artists with plans to expand it to more content owners and additional territories, relies on a “boost” model that rewards artists who are actively searched for by users, as well as those who maintain a level of 1,000 streams per month from at least 500 unique accounts — what Deezer/UMG are terming “professional artists.” And it has generated plenty of scrutiny from many corners of the industry, despite its initial limited scope.

Here’s how it works: Under the “old” pro-rata model — or the one still in effect at every major streaming service — one stream equals one play, and the total number of plays is divided up by artists and labels according to how many they accrue. Under this “artist-centric” model, if an artist qualifies as a “professional artist,” one stream would get “boosted” to count as two plays; and if a user actively searches for or clicks on an artist’s song, that stream would get “boosted” to count as two plays. If a user actively searches for or clicks on a song by a “professional artist,” that stream counts as four plays when the pool of revenue gets divided up. As part of this, “non-artist noise” content — essentially, things like the sound of rain or a washer/dryer that contains no music — will be removed from eligibility from the royalty pool, and eventually deleted from the service altogether, to be replaced by in-house noise uploaded by Deezer that will not generate revenue.

That’s the headline change, but there are many other elements to this switch as well, some designed to root out streaming fraud or bad actors gaming the system, and others that are designed to promote human artists at the expense of general audio. Deezer also released some statistics to support the changes, including that “non-artist noise” content accounts for 2% of all streams; that in 2022, 7% of all streams on its platform were fraudulent; and that, contributing to the clutter on the platform, 97% of all uploaders to Deezer generated just 2% of total streams. All told, Deezer eventually expects the changes to increase artist royalties by as much as 10%.

Still, there is work to be done for the service to implement this more widely. Deezer CEO Jeronimo Folgueira says the company is actively looking to bring more partners aboard, and expects to have more content providers on the system by the time of the Oct. 1 launch, with a full rollout with all providers across all territories intended by next year. In the meantime, “the royalty structure of labels and artists that are not signed on yet will not be affected during the transition period,” he says. The model will also initially only cover recorded music royalties, though he says “our goal is to include publishing royalties as well and will begin discussions with publishers in the near future.”

Folgueira spoke to Billboard to explain how it all works and break down how the companies created the thresholds and distinctions that underpin the new system.

Billboard: Can you walk me through the last six months of how you guys got to this point?

Jeronimo Folgueira: Deezer has been promoting a change in the model for more than four years, advocating for UCPS [User-Centric Payment System]. UCPS is much better than the old model that we had, but we figured that there’s a better way of implementing this, which is artist-centric. Artist-centric is better than UCPS, which is why we were able to get this one over the finish line, whereas with UCPS there was a lot more resistance.

Basically, given our background, it was obvious that we would engage in reviewing the system. And Universal has, in the last few months — since Lucian Grainge took on this topic personally very strongly — supported changing the model to artist-centric, so we announced a collaboration with them where we looked into the data with a consultant that they hired to see, basically, what would be the right way of moving the model.

It started from different parts. We came from a UCPS base, Universal came from an artist-centric point of view that was different from where we ended up, and we tried to find something that would make sense and would be fair for the whole industry and achieve the benefits of what we wanted while minimizing the negative impact. Because with UCPS, there were some really good artists who got negatively affected. But with the artist-centric model we’ve created now, basically all professional artists creating valuable content will get a benefit. Some get a huge benefit, and some get a small benefit, but creators making high-value content all benefit. With UCPS, there was more shuffling for artists.

That’s why in this first version of artist-centric, we’re focusing mostly on eliminating noise from the royalty pool and giving a boost to professional artists that create valuable content that users love and want. We’ve been working on this for months, working on different versions of the model, running data to make sure that we eliminated the wrong incentive and created the right reward for the right content and behavior. 

What do you expect the effect to be?

Overall, the pool doesn’t really change, it changes the distribution of the pool. But effectively what we’re doing is reducing the economic incentive for fraud and gaming the system. We’re eliminating the payouts to pure noise, and we’re boasting the payouts to real artists. So effectively there will be a shift of money from low-quality content — or not even real music — back to real, professional artists. So what we see is that producers of valuable content will get an uplift, on average, of around 10%.

What does a “boost” mean?

The boost is for a professional artist — and we consider that to be if you have more than 500 listeners a month and more than 1,000 streams. The threshold is very low, and any small, independent artist will reach those levels, so as long as you have a minimum amount of a following and fans, you’ll get to that boost. And if people search for your song, or add it to favorites or have it in a playlist, it gets another boost. So it basically means a stream of a song from one of those artists will count four times for the pool system. So it’s still a pool system, but those streams will count four times. Whereas rain, for example, will count zero, and functional music will count once. So they get boosted 4x for producing content that people actually love.

And where does the extra money come from?

The pool is the same, but the way that pool gets distributed is based on the share of streams. But that’s where the boost comes from. Noise will not get paid at all, so that’s where some of the money comes from; functional music, or music from artists that do not qualify for the threshold, will get paid less; and then artists that create valuable content will get the boost, therefore they’ll get paid more.

How did you come to the “professional artists” distinction?

We looked at different thresholds. We wanted to create a threshold that was transparent and fair, so that a small, up-and-coming artist could get there, because we want to support new up-and-coming artists and independent artists. So it was very important that this was something that was good for all artists, not just artists that were signed to a major record label. With that threshold, even though a lot of the artists on the platform will not qualify to get that boost, the majority of the streams actually do. If an artist doesn’t get to 1,000 streams and 500 listeners a month, they cannot make a living [through streaming] regardless of what the payout of the model is. So you’re not technically a professional. And any up-and-coming artist that is rising up gets to those levels pretty quickly. You don’t need big marketing budgets or promotions behind that. We’re talking about levels that are relatively easy to achieve once you are a professional and do this seriously.

But wouldn’t those smallest artists need that money the most?

Yeah, but we’re talking about people that are making €3 or €5 euros per month; it doesn’t make any real difference. It will not change anything at all. That’s why the threshold is so low — that economically it makes no impact whatsoever.

What effect would this have on playlisting? If you click on an artist’s song, they qualify for the boost — is that just if you’re looking at an artist’s page and seeking out their music? Or if you click on their song that’s first on a playlist?

If a song is on a playlist, it will always get the active boost. You would not get it if it’s algorithmically pushed to you. So if you’re listening to [algorithmic playlist] Flow, for example, and you discover new songs on Flow, you haven’t really chosen them, so those would not get the boost. If you come across a song [on an algorithmic playlist] and favorite it, that would get the boost.

What do you define as “non-artist noise”? Is there a threshold there? 

We wanted to be very fair and transparent and start in a very simple way, which is noise that has no music at all. Right now what we are going to stop paying, and eventually deleting, will be pure white noise — the sound of a washing machine, or rain, but without any music or anything else. That is the first stage, because it’s very easy to detect and very fair.

Then, there are different layers. Once it has music, then obviously it will not have the artist boost, most likely, and will probably not get to the active boost, but it will still be paid and still be there. So it won’t qualify for the boost, but it will still be paid and be available. Later on we’ll look into how that evolves and make sure that people aren’t abusing it, and if it becomes an issue then we will address it. It has to be a model that gets reviewed regularly, the same way that the Google search algorithm gets reviewed regularly to make sure that it’s always giving you the most relevant results, to make sure that there’s no gaming of the system, that it’s actually helping real artists.

What we’re trying to do here is support the creation of high-value content from real artists. And therefore we will continue to monitor it. Initially, it’s a very simple execution: pure noise gets kicked out, but anything with music will stay for the time being.

Where do you draw that line between what is “functional music” and what is artistry?

Right now, we don’t, because it’s a very difficult line to draw. If we find a way to draw that line then we will, but it has to be fair and it has to be very transparent. It cannot be subjective. We haven’t found a rule that is fair and transparent to define what is functional music and what is not, so that’s why we decided not to go there and went for the boost instead. Because what we see is, if it’s functional music, people don’t really add it to a playlist or follow it or search it or put it in favorites. So usually, things that are functional music, by nature, will not qualify for the boost. So the boost is basically a smart way of letting the behavior of the users boost what is real, high-value content, versus what is purely functional music.

Is this also about AI protection? Protecting “real” artists vs. AI artists?

Initially, we’re not taking any steps against AI. The model is not designed against it. However, it is a model that is built in a flexible way that can protect real artists from AI in the future, and what we said is that the real artist boost should be applied to real, human artists, so if it’s a machine it should not qualify for the active boost.

Your press release also mentioned a “stricter provider policy” that you guys are implementing. What does that entail?

Basically right now, like every other DSP, we allow people to upload music through these do-it-yourself [distribution] platforms; there’s plenty of them. And there’s a lot of content being uploaded. What we want to do is make sure that we get content that is valuable. We don’t want more noise getting uploaded to the platform and we want to be very strict with fraud and gaming [the system]. There are certain providers where more than 50% of what they uploaded we had to take down because of fraud. So we’re going to potentially block those providers altogether. We do not want to be used to game the system. Until now we had been allowing everything, and only when something gets detected as fraud did we deal with it. Now we want to be a lot more strict with what we allow to be uploaded.

But as you were saying, so much gets uploaded every day. How do you screen that?

AI. There will be clear rules, and then the machine will be screening all content that gets uploaded, and once you get to certain thresholds where they’re providing too much content that is detected as fraudulent or gaming the system, then we will just block them, the same way that Google will penalize anyone that is gaming their SEO and will remove them from search results for at least six months. There are penalties for bad behavior. Right now in streaming there are no penalties for bad behavior, and we’re trying to introduce them, the same way that Google and many other platforms do.

What other practices are you instituting to combat this fraud?

One really important aspect of eliminating the fraud element is we’re going to put a cap on the impact of a single user on the pool of streams: only 1,000 streams per user per month will count. So if you listen to 2,000 streams, then your streams will count half. That way, you cannot have one account racking up 10,000 streams and stealing money from the pool. A normal human will consume anywhere between 400 and 600 tracks per month, so we’ve set the threshold at 1,000. At 1,000, more than 90% of the behavior is captured and then only the outliers go beyond that. Some of it is not fraudulent — it’s usually young kids listening to K-pop or rock day and night. But the behavior of the fraudulent accounts, or gaming the system, happens by hacking accounts and generating huge amounts of streams to steal money from the pool. So by putting a cap of 1,000 streams per user, we are eliminating the economic incentive. You’d have to fake or hack a lot of accounts to have an economic impact, whereas right now with only a handful of accounts you can have a massive impact on the pool. 

That 400-600 tracks, that was a result of your research?

Yes, our data. We have 10 million monthly subscribers, and over the last 15 years it’s pretty statistically significant that a normal human will listen to something in the range of 500 tracks. It really depends on age; the younger you are, the more tracks you listen to. But generally speaking, in normal human behavior, everything will be captured below 1,000 streams. If you’re above 1,000 streams you’re an outlier, and we don’t want those outliers or gamers of the system to have an impact on the pool.

What other tweaks are possible as you guys start to roll this out?

One thing we left out that we looked at was potentially adding another layer, which was streaming time. So instead of calculating it by stream, calculating it by the time you spend streaming a song. But what we saw is that with the current boost, the impact is already captured. So if you added listening time on top of the current layers that we created, the impact is minimal, because if you love a song, you usually listen to the whole song. We explored it, looked at the data and decided it wasn’t needed, and we wanted to keep it as simple as possible. But we haven’t completely ruled out listening time.

The other thing we haven’t completely ruled out is moving more and more towards a user-centric approach. Right now we cap things at 1,000 streams. But that can come down eventually to make it closer and closer to a UCPS approach. So that’s another variable that we’ll want to keep an eye on. And the other one is the threshold for a “professional artist.” We need to make sure that the 1,000 streams and 500 listeners a month is the right level and that it doesn’t have negative consequences. Because we really care about new, independent up-and-coming artists. We want to support them. So we will be reviewing that and its impact on new artists as well.

What might make you lower that threshold?

We have looked at so much data, which is why I feel like the level is in the right place. But feedback from the community and if there were any unintended consequences that we couldn’t see in the data that we already have.

When you roll this out, does this only apply to UMG artists?

Yes and no. Right now, the agreement is with Universal, however we’re in discussions with all content providers. The majority of content providers are very happy with the artist-centric model, because everyone who produces high-quality content gets a boost, whether you’re a major record label, an independent record label or a small indie artist distributing yourself. As long as you create content that people value, you will benefit from the model. I expect a big chunk, if not more than half, of our content will be on the new model by the time we launch this on the first of October. And our intention is to roll this out to all providers in all countries in 2024.

What would be a mark of success for this program? Six months from now, what would tell you that this is working?

I think it’s if real artists really get the boost, if they see an uplift in royalties, that’s where we would say that this model is working and helping good artists create valuable content. That’s ultimately what we want to do. The pool of money is the pool of money. Obviously we’re working to raise the ARPU [average revenue per user] and grow the pie, but that’s a different discussion. But from the pie that we have, more of the money has to go to artists who create valuable content, to implore them to continue to create valuable content. If those boosts work as intended and the real artists creating valuable content see an uplift in royalties, this model will have succeeded.