We’re back with a special music-related
analytics episode! Following Next Big Sound’s acquisition by Pandora, Julien Benatar moved from engineering into product management and is now responsible for the company’s analytics applications in the Creator Tools division. He and his team of engineers, data scientists and designers provide insights on how artists are performing on Pandora and how they can effectively grow their audience. This was a particularly fun interview for me since I have music playing on Pandora and occasionally use Next Big Sound’s analytics myself. Julien and I discussed:
"I really hope we get to a point where people don’t need to be data analysts to look at data."
"People don’t just want to look at numbers anymore, they want to be able to use numbers to make decisions."
"One of our goals was to basically check every artist in the world and give them access to these tools and by checking millions of artists, it allows us to do some very good and very specific benchmarks"
“The way it works is you can thumb up or thumb down songs. If you thumb up a song, you’re giving us a signal that this is something that you like and something you want to listen to more. That’s data that we give back to artists.”
“I think the great thing today is that, compared to when Next Big Sound started in 2009, we don’t need to make a point for people to care about data. Everyone cares about data today.”
Brian: I’m really excited today for this episode. We have Julien Benatar on the show and he’s from a company that I’m sure a lot of people here know. You probably have had headphones on at your desk, at home, or wherever you are listening to Pandora for music. Julien , correct me if I’m wrong, you were the product manager for artist tools and insights at Next Big Sound, which is a type of data product that provides information on music listening stats to, I assume, artists’ labels as well to help them understand where their fans are and social media engagement.
I love this topic. I’m also a musician, I have a profile on Next Big Sound and I feel music’s a fun way to talk about analytics and design as well because everybody can relate to the content and the domain. Welcome to the show. Did I get all that correct?
Julien: Yeah, it was perfect.
Brian: Cool. Tell us a little about your background. You’re from France originally?
Julien: Yes, exactly. I grew up next to Paris, in Versailles more specifically, and moved to New York in 2014 to join Next Big Sound.
Brian: Cool, nice. You’ve been there for about four years, something like that. You have a software engineering background and then now you’re on the product side, is that right?
Julien: Exactly yes. I joined the company back when we were a startup. Software engineering was perfect, there was so much to do. To our move to Pandora, I moved to a product manager role around a year ago.
Brian: Next Big Sound was independent and then they were acquired by Pandora. I assume there is good stuff about your data. Why did Pandora acquire you and how did they see you guys improving their service?
Julien: We got acquired in 2015. The thing is, Next Big Sound was already really involved in the music industry. We already had clients like the three major labels and a lot of artists were using us to get access to their social data. I think it was a very natural move for Pandora as they wanted to get closer to creators and provide better analytics tools.
Brian: For people that aren’t on the service, I always like to know who are the actual end users, the people logging in, not necessarily the management, but who sits down and what are some of the things that they would do? Who would log in to Next Big Sound and why?
Julien: Honestly, it’s really anyone having any involvement into the music industry, so that can be an artist, obviously, try looking to try their socials and their audience on Pandora. But you can also be a booker trying to book artists in their town. We have a product that can really be used by many different user personas. But our core right now is really artists and labels, having contents on Pandora and trying to tell them the most compelling story about what they’re doing on the platform.
Brian: When you think about designs, it’s hard to design and we talk about this on the mailing list sometimes but it’s really hard to design one great thing that’s perfect for everybody so usually you have to make some choices. Do you guys favor the artist, or the label, or as you call them,the bookers or whom I know as presenters,in the performing arts industry? Do you have a sweet spot, like you favor one of those in terms of experience?
Julien: I think it’s something we’re moving towards, but it hasn’t always been this way. Like I told you, we used to be a startup or grow us to make a product that could work for as many people as possible.
What is funny is we used to have an entity on Next Big Sound called Next Big Book where we used to provide the same type of service for the book industry. If anything, it’s been great to join Pandora because then we could really refocus on creators and it really allowed us to, I believe, create much better and more targeted analytics tools to really fulfill needs for specific people like artists and labels.
Brian: I would assume individual artists are your biggest audience or is it really heavily used by the labels or who tends to...
Julien: I think it’s pretty much the same honestly. I think the great thing today is that, compared to when Next Big Sound started in 2009, we don’t need to make a point for people to care about data. Everyone cares about data today. I think that everyone has reasons to look at their dashboards and especially for a platform like Pandora with millions of users every month. Our goal is really just telling them a story about what does it mean to be spinning on the platform and the opportunities it opens.
Brian: You talked about opportunities, do you have any stories about a particular artist or a label that may have learned something from your data and maybe they wrote to you or you found out like in an interview how they reacted like, “Hey, we changed our tool routing,” or, “Hey, we decided to focus on this area instead of that area.” Do you know anything about how it’s been put into use in the wild?
Julien: Yeah, it’s used for so many different reasons. For the people who don’t use Pandora, something I really like about the platform is it’s really about quality. As you use Pandora, you have the opportunity to thumb up or thumb down songs and as you do, you’re going to get recommended more songs like the ones you like. It’s really about making sure that you get the best songs at all times.
The reality then is that for artists, their top songs on Pandora can be pretty different than their top songs on other platforms because sometimes their friends are going to be just reacting more to some part of their catalog than another one. I’ve heard many times of artists changing their playlists in looking at which songs where their fans thumbing up the most on Pandora.
Brian: Could you go through that again? How would they adjust their playlist?
Julien: Usually, people use Pandora as a radio service. While we already have internet today, most people are listening to the radio because they’re usually are very targeted and it just works really well. The way it works is you can thumb up or thumb down songs. If you thumb up a song, you’re giving us a signal that this is something that you like and something you want to listen to more. That’s data that we give back to artists. We tell them, “This are your most thumbed songs on Pandora. These are the songs that people engage with the most on the platform.” Looking at this data, you can actually inform them songs that they believe they should be playing more on the store.
Brian: I see. A lot of it has to do with the favoriting aspect to give them idea what’s resonating with their audiences.
Julien: Qualitative feedback, yes.
Brian: Got it. Actually, it’s funny you mentioned the qualitative feedback. In preparation for this, I was reading an article that you guys put out back in March about a new feature called weekly performance insights, which is really cool and this actually reminds me of something that I talked about in the Designing for Analytics mailing list, which is the act of providing qualitative guides with your analytics. A lot of times they analyze for turnout quantitative data and whenever there’s an opportunity to put stuff into context or provide qualifiers, I think that’s a really good thing and you guys look like you’ve have done some really nice things here.
I’ll paraphrase it and then you can jump in and maybe give us some backstory on it. One of the things that I think is really cool is there're concepts of normalcy in here so that, if I’m an artist and I look at my numbers, I have an idea. For your Twitter mentions, for example, you say, “For artists with 26,000 followers, we expect you to get around 44 mentions.” When you show me that I have 146 mentions, I can tell that I’m substantially higher than what my social group would be.
I think that’s a really fantastic concept that people not in music could try to apply as well which is, are there normalcy bans where you’d want to sit? Is there some other type of group, maybe, an industry, or apparent group, or another business unit, whatever it may be to provide some context for what these out of the blue numbers mean that don’t have any context?
How did you guys come up with that and can you tell us a bit about the design process of going from maybe just showing, “You’re at 826 apples,” as compared to what? How did you move from just a number into this these kind of logical groupings where you provide the comparisons?
Julien: I think what’s really fascinating is, we really live in an age of data. As an artist, you need to be on social media for the most part. There still a lot of artists I listen to but just decide not to. It’s part of things but at the same time, real big success in the music industry didn’t change. It’s still being on the Billboard chart, getting a Grammy and all these things. But as we see this, we have millions of artists looking at their data every day and just are not able to understand, like is it good or is it not good. Everyone starts at zero.
We have a strong belief that data can only be useful when put in context. Looking at the number on its own can give you a sense of how things are doing but that can also be dismissive. An example is, a very common way to look at data is to look at a number and look at the percent changing comparison to the previous week. You’ve got a bunch of tables and you look at, am I growing or am I not growing.
The reality is it’s actually impossible to always have a positive percent change. There’s no artist in the world that always does better week by week. Even Beyonce, I can assure you that the week she released Lemonade, she had more engagement on Twitter than the week after. With that in mind, we really try to give a way for artists to understand how are they doing for who they are and where they are currently in their career.
Next Big Sound started in 2009. One of our goals was to basically check every artist in the world and give them access to these tools and by checking millions of artists, it allows us to do some very good and very specific benchmarks. For an artist, like the example you said, for instance an artist with a thousand Twitter mentions in a week, is it good or bad in comparison to their audience size? This feature comes because that’s just the question we’re asked. Artists want to know is it any good? What does this number actually mean for me? That’s why we really wanted to, in some ways, get out of being a content aggregator platform and really be a data analytics platform. How can we actually give information that can help artist make better decisions?
Brian: I remember the first time I got what I would call an anomaly detection email from your service and it was about some spike in YouTube views or something like that. I thought it’s fantastic in two reasons. First of all, you identify an anomalous change and I think in this case it’s a positive anomalous change. That tells me that I should log in the tool. Secondly, you proactively delivered that to me. On the Designing for Analytics mailing list, we talk about is that user experience does not necessarily live inside your web browser interface or your hard client or whatever you’re using to show your analytics. Email and notifications are a big part of that. Can you tell me about how you guys also arrived at when you pushed these things out and maybe talk about this little anomaly detection service that you have?
Julien: It all started when we got acquired by Pandora. We decided to just invite a bunch of users and just talk to them, understand how to use our product and what did they think about it. We had artists, managers, and label people come over and we just talk to them and basically they all said, “We love it.” But then, by looking at their actual usage, they don’t use it that much. I guess one of their questions was when should I be looking at my data? Everyone is very busy. As you’re an artist, you need to perform, you need to write music, you need to engage with your fans and same goes with everyone.
When should I look at data? The reality is by being a data company, we do get all the data, we have all the numbers. We have ways to know when things are supposed to be known, when artists should be acting on something. We just turn this into this email notifications. Anytime we notice that an artist is doing better than expected, we just let them know right away.
Brian: That’s great. Do you do it on the opposite end too? If there’s an unexpected drop or maybe like, “Oh, you put a new track out and your socials dropped,” or something like that, do you look at the negative side too or do you tend to only promote the positive changes?
Julien: As far as pushes, we decided to only do push for positive. But as you mentioned weekly performance, weekly performance can give you some negative insights, like, “You’re not doing as well as artists with the same size of audience as yours.” The reason we didn’t do it for our notification is, anomalies are really hard to completely control. A reason, for instance, is Twitter removing bots. Basically, every single artist would have had an email telling them, “You lost Twitter followers this week.”
It was a lot of work to really tune our anomaly factor to actually only send emails when something legitimate happens. That’s the reason we only decided so far to do it for positive but we actually have been thinking about doing the same for negative but that’s another type of work.
Brian: Yeah, you’re right. You have to mature these things over time. You don’t want to be a noise generator.
Julien: Exactly.
Brian: Too many, then people start to ignore you. I’ve seen that with other data products I’ve worked on which just have really dumb alerting mechanisms that are very binary or they’re set at a hard threshold and just shootout noise and people just tune it out.
Julien: I’m glad you mentioned this because this feature was in beta for a year for that specific reason.
Brian: Got it.
Julien: We had to learn the hard way. We had like a hundred beta users. We’ve got way too many emails because anytime there were an anomaly anywhere, they would just get an email. For the most part, it was things that were supposed to help them. If a notification becomes noise, then that’s absolutely against its purpose.
Brian: I don’t know if everybody knows how the music business works, at least from the popular music side, but just to summarize. You have individual artists that are actually performers. They may or may not have an artist manager which takes care of their business affairs, represents them like negotiations with people that book shows. Then you have labels which are sort of like an artist manager except they’re really focused on the recording assets that the artist makes and they actually tend to own the recordings outright at the beginning and then over time, the artist may recoup through sales they make it the ownership act and the sound recordings they make. Of those kinds of three major groups, is there a one that’s particularly hungry or you’re the squeaky wheel that is most interested in what you’re doing?
Julien: I really think that into these three groups, we have a subset of users that are really into the data and into the actionability of it. I don’t think it’s one specific group of user. It could be all around the industry like we have the data-savvy, they really want to know. We have some users that actually would rather get more notifications even if they need to on their end to figure what is right from what is wrong. But since we have such a wide user base of different type of people, we decided to go on the conservative side and make sure to only share things that we thoroughly validated through all of our filters.
Brian: I assume that your group reports into some division of Pandora, I’m not sure of that. Are you reporting into a technology, like an IT, or a business unit, or marketing? Where do you guys fit in the Pandora world?
Julien: We’re part of the creator’s tools. I don’t really have a perfect answer to this.
Brian: Okay. I guess my main question being, because when we talk about designing services, we talk about both user experience, which is the end user thing and about business success or organization success. I’m curious, how does Pandora measure that Next Big Sound as delivering value? I can understand, I’m sure our artist can understand how the artists value it through understanding how is my music moving my audiences, et cetera. Is there a way that Pandora looks at it? Are they interested in just time spent? The analytics on the analytics, so to speak, is what I’m asking about. How do you guys look at it like, “Hey, this is really doing a good job,” or whatever? Do you know how that’s looked at?
Julien: To be honest, I think you said it right. Our goal is to help artists make their decisions through data and having artists use the platform is currently the way Pandora sees us doing a good job. Actually, it hasn’t changed that much since our acquisition.
One of our main KPI for the past and couple of years is something I would call insights consumes. Just making sure that our users, artists, anyone using Next Big Sound are consuming data. That can be them logging into the website or that can be them opening one of our notifications. But so far that was our main KPI. We’re trying to work on some more targeted KPI, potentially like actions taken, that would be the North Star, but we're still working on how to do that right.
Brian: Do you guys facilitate actions, so to speak, directly in the tool or are there things people can do with those actions really take place outside of the context of Next Big Sound?
Julien: There are actions that artists can take to the other creator’s tools provided by Pandora. For instance, artists have the ability to send audio messages to anyone listening to them. If they go on tour into the US, they can have targeted messages in every single song they’re going to play. If anyone listens to them there, they can just click and buy a ticket.
We’re working to make sure that artists are aware of these tools because they are free and they’re generally helping them grow at their careers. But regarding external actions, so far we don’t have any one-click way to tweet at the right time to the right people or with the right content or anything like this.
Brian: Sure and that’s understood. Not every analytics product is going to have a direct actionable insight that comes right out of it. You guys may be feeling a longer term picture about trending and maybe for a certain artist to get an idea if they’re releasing music fairly frequently, what stuff is working and resonating, and what stuff is not. I can understand that. There may not be a button to click as a result immediately.
Julien: That’s the goal though. Everything we do right now is going towards this objective. Maybe I can tell you a little about the way we think about data and that can give more sense to it.
In order to work on any new feature, we follow this concept called the data pyramid. It’s something that you can Google. There’s a Wikipedia page for it. Let me explain to you how it works. The data pyramid, it’s a pyramid formed of four layers. It could be upon each other and each representing an exquisitely useful application of data. At the bottom of the pyramid we have the data layer. Any sort of data that we may have. For our case, Android data, Twitter, Facebook just getting the numbers, getting the raw data.
On top of it, we have the information layer. The information layer is going to be ways you have to visualize this data. I guess it’s like the very broad sense of analytics. We’re going to give you tables, graphs, pie charts, you name it. We’re giving you ways to craft stories about this data but it’s on you to figure it out.
Then on top of it we have what we call the knowledge layer. That’s where things start to get interesting. The knowledge layer is the contextual part of it. It’s like, “What do this number actually mean?” It has industry expertise. For instance, the way we’re going to work about it for musicians and their true data may be different than any other industry. The knowledge layer goes like a weekly performance. It’s a perfect answer to it. It’s what does it mean for me as a musician with a hundred fans to get two mentions this week. Same for notifications. It’s telling you that you should be looking at your data right now because something is happening.
That’s how we get to the North Star and the last part of the data pyramid which is intelligence. The goal of intelligence is actionability. Now that I get to understand what does this number mean to the specific context, what should I be doing?
Following your question, everything we’re trying to do here is to get to a point where we can just send an email to an artist and tell them, “Hey, you should be doing this right now because, with all the data that we have, we believe that this is going to have the highest impact for you.”
Brian: It‘s really fascinating that you just outlined this data pyramid. I actually haven’t heard of this before. It made me think of one of the kind of, it’s not a joke but in the music community, I’m also a composer and when we write stuff, the kind of running joke is like nothing is new. Your ideas for this new song or this new melody I’m composing, it probably came before you. You heard it there before.
I wrote a post on my list that was pretty much exactly the same thing except the knowledge layer. I was calling that insight. Data have been this raw format and information being the first human-readable format that’s like say going from raw data to a chart, a histogram. Now I have a line on a chart and then the insight layer being, I have a line on the chart and another line comparing it to like you said, average, or my social group, or a parent group, or some taxonomy, or an index. Then the action or the prescription for what to do or the prediction those that kind of lead you in about action which would be that fourth state. You’re like, “Oh, is this really a new concept?” It’s like, “Nope. Someone else already thought of that.” I totally want to go read about this data pyramid.
Julien: That’s amazing.
Brian: I’ll find that link to the data pyramid and I’ll put that in the show notes for sure. I thought that was really funny.
Julien: It’s funny that you called it insight because that’s the way we call a lot of our features are working out. The way we define insight is bite-size, noteworthy, sharable content. How can we get into the noise of all of the data that only gives you exactly what you should be looking at. That’s how we got into notification and weekly performances. This is the one thing you should be looking at.
Brian: I understand what you’re getting at there. The insights are, like you said, bite-size chunks of interesting stats that someone can put some kind of context around. That’s great and it’s good. One of the things I liked, too, that you talked about was you said, “Oh we got like a hundred users, like a beta group and that kind of inspired some of this.” Your product response to how do we help people know when to come and look at our service. I think this is really good because one of the problems that I see with clients and people on the list, I think is low engagement. This is especially true for internal analytics companies. Low engagement can be a symptom of a difficult product, it doesn’t provide the right information at the right time, it may not have a lot of utility, or it’s a resistance to change. People have done something the old way and they don't want to do it the new way.
One of the recipes you can follow if you’re trying to do a redesign or increase engagement is to involve the people that are going to use the service in the design process, both the stakeholders as well as the end customers. This is especially true again for the internal analytics people. Your customers or other employees and your colleagues. By engaging them in the design process, they’re much more likely to want to change whatever they’re doing now.
I loved how you guys did some research. Now I want to ask, do you frequently do either usability testing or interviews? Is that an ongoing thing at your company or is it really just in front of a big feature release or something like that? How do you guys do this research? Can you tell me about that?
Julien: Of course. It’s consent. We haven’t released any major feature without doing some heavy user testing. I’m very lucky to be working with two designers, Justin and Anabelle who are very user-focused. Honestly, if you come to our office, at least every week we’re going to have some user interview and just talking to them, showing them prototypes, and just see how do they play with it.
Brian: So you’re doing a lot of testing it sounds like. That’s fantastic.
Julien: At the same time it’s always to find the right balance because you could be overtesting things too. We really are focusing on user testing for new things and make sure that the future that we are working on actually answers their user story that we intended.
Brian: I don’t know how involved you get participating in these, but do you have any interesting stories or anecdotes that you got from one of those that you could share?
Julien: Let me think. I do participate into a lot of them but I’m not sure I have an example right now.
Brian: Are most of the people you interview, are they current users of Next Big Sound or do you tend to focus on maybe artists that haven’t experienced the service yet or you mix it up?
Julien: We mix it up. We mostly engage with users that we already have but then we can decide to go with users that haven’t used the platform for a while, or more active users if you want to understand how we’re useful into their day to day. What I would say is that, surprisingly, it’s very easy to get users to chat about their experience with the product. I didn’t assume that we would get so many responses when we tried to have people come over or just hop on the zoom to check a new feature.
Brian: I’m glad you actually mentioned that because I think in some places, recruiting is perceived to be difficult and it probably isn’t. Maybe you haven’t done it before but as I tell a lot of my clients, a lot of people love to have someone listen to them talk, tell them all about their life and what’s wrong with it, and how it could be better with their tools. They love having someone listen to them and especially if they know that their feedback is going to influence a tool or a service that they’re using. They tend to be pretty engaged with it. I find it’s really rare that I do an interview with a client’s customer and they don’t want to be included in the future round like, “Hey, when we redesign the service, can we come back to you and show you what we’ve done?” “Oh, I love to do that!” Everybody wants to get engaged with it.
There are places where recruiting can be difficult when it’s hard to access the users, some of the enterprise software space that can be an issue sometimes. But generally, if you can get access to them, they tend to be pretty willing to participate. I’m glad you mentioned that.
Julien: I think the great part about testing with current users on the platform is to actually show them prototypes with real data, not just show them an abstract idea that we want to work on. As soon as they can see what we’re working on apply to their own career as musicians, for instance, that can lead to fascinating discussions.
Brian: You made a really good point on the real data thing. I remember as far back as 10 years ago or whenever, I use to work at Fidelity Investments, we would see this issue when we’re working on the retail site for investors. When you show a portfolio that, for example, has Apple stock trading at $22 in it, you’re not really there to test what is the price of Apple stock but you might be testing something entirely different and the customer cannot bear what is going on? They’re so stuck on this thing. It’s all fake seed data in the prototype.
The story here being if you’re a listener, when you test it’s important to have at least realistic data. You don’t want to have noise in the test or whatever your studying or else you can end up on this tangent. Try to make the numbers looks somewhat realistic if you’re using quantitative data.
In some cases, people can be taught to roleplay. Pretend you’re Drake or pretend you’re some big artist and then they can get their head around why they have billions of streams instead of thousands which they’re used to.
Julien: Absolutely. That also helps us just build better products because the reality is we have a lot of artists with maybe 10 plays in a month. As we build visualizations like something that we built a line of looking at Drake’s data, it’s not going to work as intended for a smaller artist sometimes. Having real data involved as soon as possible into the design process has been such a game changer for us. We really have a multidisciplinary team involved into the research and design of everything we do. I’m working with a data scientist, data engineer, a web engineer, and designer on a daily basis.
Obviously, we all have our things to do. But as we get into creating something new, we just make sure to have someone helping us get the real data, interview the right user, and just create prototypes as soon as possible. Working with prototypes is essential into building useful data analytics tools.
Brian: Yes, you do learn a lot more with a working prototype. It’s not to say you can’t test with lower fidelity goods, especially early on but for a service like yours when the range of possible use both the personas and also you’ve got the Drakes of the world, big major label artist and then down to really small independents, it’s really important to have an idea how your charts are going to scale, and what’s going to happen with data. Even just small stuff like how many decimal points should you be showing on a mobile device, some of the numbers might cram up.
Julien: Exactly.
Brian: All this stuff that you never think, if you only look at one version of everything, you can end up with a mess. I’m glad that you brought that up.
Julien: I couldn’t say better. The decimal is actually something that we’ve had to discover through real data.
Brian: To all of you in the technical people out there, I will say this. If I’ve seen one trend with engineers, is they love precision and there’s a lot of times when there’s very unnecessary precision being added to numbers. Such as charts and histograms. Histograms are usually about the trend, they’re not about identifying what was the precise value on this date at this time. It’s about the change over time. Showing what’s my portfolio worth down to three digits of micro-cents or something like that is just unnecessary detail. You can probably just round up to the dollar or even hundreds of dollars or even thousands of dollars in some cases.
It actually is worse. The reason it’s worse is that adds unnecessary noise to the interface, you’re providing all these inks that someone has to mentally process, and it’s actually not really meaningful ink because the change is what’s important. Think about precision when you’re printing values.
Julien: This concept of noise is so essential today for any data analytics tools. There is so much data today. There is data for everything. I think it’s our responsibility as a data analytics company to make sure what are we actually trying to help our user with this data set is not just about adding new metrics. Adding new metrics usually is just going to add noise and not be helpful in comparison to fairing what do they need to make the right decision.
Brian: Right. Complexity obviously goes up. The single verb, ‘add,’ as soon as you do that, you’re generally adding complexity. One of the design tools that is not used a lot, and this is something I try to help clients with is, what can we take away? If we're not going to cut it out entirely, can we move this feature, maybe this comparison to a different level of detail? Maybe it’s hidden behind a button click, or it’s not the default. But removing some stuff is a way to obviously simplify as well, especially if you do need to add new things. Your only weapon is not the pencil, you’ve got the eraser as well in the battle so to speak.
Julien: I couldn’t agree more. On Next Big Sound we have this concept of artist stages. It’s a way for us to put artist into buckets and by looking at their social instrument data. It goes from undiscovered to epic. We do that by looking at all of the data we have and looking at it in context.
I don’t have the numbers right now because they update on a daily basis but every artist starts undiscovered. For instance, as they get 1000 Facebook likes, maybe they’re going to get to a promising stage. We have all of these thresholds moving everyday looking at trends among social services. But what is interesting is that for instance, for a booker, a booker doesn’t need to look at the exact number of Twitter followers for an artist. He needs to know that he’s booking for a midsized venue in the city he’s in and he’s probably going to be looking for promising to established artists and not looking for the mainstream to epic artists. It’s always about figuring a way to use the numbers to tell the story.
Brian: I’m totally selfishly asking for myself here, but I was immediately curious. I live in Cambridge which is in the Boston area, and I am curious who are the big artists in our area and what is the concentration? I’m in a niche. I’m more in the performing arts market, in the jazz, in world music, and classical music but I’m just curious. Is there a way to look at it by the city and know what your artist community looks like? You guys do anything like that?
Julien: We don’t currently. But I think YouTube has actually a C-level chart available. It’s not part of something we do because I think the users it would benefit are not the users we specifically try to work on new features. It’s more something for bookers than artists ,specifically ,but it’s exactly the type of thing that we need to think about when we prioritize new features.
Brian: I’m curious just because the topic’s fairly hot. Everybody is trying to do machine learning projects these days. I don’t like the term AI because it tends to be a little bit overloaded but are you guys using machine learning to accomplish any particular problems or add any new value to your service right now? Is that on your horizon?
Julien: How do you think about machine learning?
Brian: A lot of times I associate it with predictive analytics or understanding where you might be running instead of just using statistics. I don’t know what kind of data you might have for your learning that you can feed in but maybe there’s aspects about artists that can predict. Especially, I would think like in the pop music world where there tends to be more commercialization of the music, I would say, where it’s like we need a two-minute dance track at this tempo specifically because DJs are going to play it. It’s a very commercial thing. It’s very different than what I’m used to.
So I’m curious if there’s a way to predict out how an artist may do or what kinds of tracks are performing well. Like these tempo songs, we predict over the next six months that tech house music at 160 beats for a minute is going to do really well based on the trending. I don’t know. I’m throwing stuff out there. The goal, obviously, is not to try to use like, “Oh Home Depot has this new hammer, let’s run out and get it. We don’t even know what it’s for but everyone else is buying it.” That’s how I joke about machine learning. It’s like you need to have a problem that necessitates that particular tool. I don’t ask such that, “Oh there should be some.” I’m more curious as to whether or not it’s a tool that you guys are leveraging at this time.
Julien: The Next Big Sound team doesn’t worked on features following the musical aspects of things. We really are focused on the user data.
Brian: Engagement and social.
Julien: Engagement data mostly, yes. But at the same time, I’m sure teams have worked on this because of the way that genome works. We have a lot of data about the way songs are made. Regarding machine learning, on the Next Big Song team, we actually have something that is called the prediction chart. You said predictions. We have this chart that is available every week.
Basically, it really goes back to having data for a long time. The fact that we’ve had data since 2009, we’ve been able to see artists actually get from starting to charting on the Billboard 200. By having all of these data, we’ve been able to see some trends, some things that usually happen for artists at specific times in their career up until they get into the Billboard 200.
We actually do have some algorithms that allow us to apply this learning to all of the artists on Next Big Sound right now and have a list every week of artist that we believe are most likely to appear on the Billboard 200 chart next year.
Brian: I see. Got it. Do you track your accuracy rate on that internally and change it over time? Do you adjust the model?
Julien: Yeah, we do.
Brian: Cool That’s really neat. Tell me, this chat has been super fun. I’ve selfishly got a little indulgent because being a musician, it’s fun to talk about these two worlds that I’m really passionate about so I could go on forever with you about this. But I’m curious. Do you have any advice for other product managers or analytics practitioners about how to design good data products and services? How to make either your own organization happy or your customers happy? Do you have any advice to them?
Julien: Yeah, of course. I guess it’s all about asking questions, honestly. What is very good with working at Next Big Sound is that it all started in 2009. Maybe actually I can go back and tell you the story about how it started and why it’s so different today.
It started in 2009. It was actually a project, a university project by the three co-founders. Basically, they were wondering about one thing. How many plays does a major artist get on the biggest music platform in the world? At that time, it was MySpace. The artist they picked was Akon. Basically, they just built a crawler, went to bed, woke up, and discovered that an artist like Akon was getting 500,000 plays on MySpace in one night in 2009.
The challenge in 2009 was to get the data. That’s why for the most part in Next Big Sound as it started was, I really think a data aggregation tool. Our goal was to get as many sources as possible and just make them easily accessible into the same place. We really are much into the information layer here. We’re giving you all the numbers and you can compare Tumblr to Vimeo, to YouTube, to Twitter, to Facebook, to Vine, to you name it into a table or a graph that you want to.
The reality is, today things change. We don't need to fight to get data anymore. We don’t need to hike our way into getting the numbers. Now, data is accessible to everyone in a very easy way. It’s kind of a contract. You, by being an artist, you know you’re going to get access to your Spotify, YouTube, Pandora, Apple Music or any other platform data very easily just by signing up and authenticating as an artist. That’s where our goal changes. Thankfully, we don’t need to convince people to care about data, we know they do already. But now the challenge is different. Now, the challenge is to make them understand what does their data mean and how can they turn it into getting even more data, getting into having even more engagement, and having even more plays.
I think that’s something that is very interesting because it really resonates into the question we’ve been asked in the past few years like, “What does my data mean and when should I be looking at my data?” If anything, these two things correlated pretty well. People don’t just want to look at numbers anymore, they want to be able to use numbers to make decisions. That’s the core of what we’re trying to achieve today. We couldn’t be there if we didn’t have users that ask us the right questions.
Brian: Cool that’s really insightful. Just to maybe tie it off at the end and maybe you can’t share this but what’s your home run? What is your holy grail look like? Is there a place you guys know you want to get? Maybe it’s the lack of data or you don’t have access to the data in order to provide that service. Do you guys have kind of a picture of where it is you want to take the service?
Julien: What is very noble about our goal at Next Big Sound specifically is we’re here to help artists. The North Star would be to make sure that any artist at any time in their career is doing everything they can do to play more shows, to reach to more people, and to make sure their music is heard.
Brian: Nice. I guess it’s like you’re already there, just maybe the level of quality and improving that experience over time, that’s your goal. It’s not so much that there’s so much unobtainable thing at this moment. Is that kind of how you see it?
Julien: I think the more we don’t feel just a data analytics tool, the more we’re getting to that goal. I really hope we get to a point where people don’t need to be data analysts to look at data. We’re always going to provide a very customizable tool for the data-savvy because they know what they need more than we can ever do it for them. We want to make sure that for everyone else, we can just make it very easy and as simple as a click for them to do something that’s going to impact them positively.
Brian: Cool, man. This has been really exciting to have you on the show. Julien, can you tell the listeners where can they find you on the interwebs? Are you on Twitter or LinkedIn? How do they find you?
Julien: For sure. @julienbenatar on Twitter, nextbigsound.com is free for everyone. Actually, we made our data public recently, so if you ever want to learn more about what we do, please check it out. We try to post on our blog about what we learn through data science, through design, and share more about why we build what we build. I recommend to just check blog and do some commitment to learn more about what we do.
Brian: I definitely recommend people check out the site. The fun thing is again, as you said, it’s public. If there’s a band you like or whatever, you can type in any group that you like to listen to and you can get access to those insights. Just kind of get a flavor of what the service does. I’ll put those links in the show notes as well as the data pyramid.
Julien, cool. Thanks for coming on. Is there anything else do you like to add before we wrap it up?
Julien: No, thank you so much. I love reading your newsletters and I’m very happy to be here.
Brian: Cool. Thank you so much. Let’s do it again.
Julien: Cool.
Brian: Cool. Thank you.
We hope you enjoyed this episode of Experiencing Data with Brian O’Neill. If you did enjoy it, please consider sharing it with #experiencingdata. To get future podcast updates or to subscribe to Brian’s mailing list where he shares his insights on designing valuable enterprise data products and applications, visit designingforanalytics.com/podcast.
Never forget to look up the online HTML CheatSheet when you forget how to write an image, a table or an iframe or any other tag in HTML!
Simon Buckingham Shum is Professor of Learning Informatics at Australia’s University of Technology Sydney (UTS) and Director of the Connected Intelligence Centre (CIC)—an innovation...
Dr. Bob Hayes, will be the first to tell you that he’s a dataphile. Ever since he took a stats course in college in...
Ganes Kesari is the co-founder and head of analytics and AI labs at Gramener, a software company that helps organizations tell more effective stories...
