David Boyle interview for Big Data Summit Aug '15

EVP Insight at BBC Worldwide. Hi, David.

Hi.

How's it going? Yeah, if you could just give us a brief rundown of your role at BBC Worldwide and how you tell us how you got into sort of data visualization and analysis.

Sure. Well, BBC Worldwide is the commercial and global wing of the BBC, so we help to monetize BBC content all around the world. And my job is insight, which to me means getting data and helping people to use it to make decisions, helping people to learn things from that data and helping people to make smarter or better or quicker decisions using that data. So mainly around TV shows, but that means it's around how people consume TV and technology and tastes and all sorts of different fun things. How did I get into it? Well, I love data. I just really love data. I did a maths degree and I just really liked in maths, how you can take a really big complex problem, break it down into small pieces, chip away at the problem, and ultimately end up solving this big, big, big complex problem. I just felt like we could do the same thing in business, you can take big, big business problems, just chip away at them with data piece by piece, and eventually you can solve the whole thing. And I love doing that. It's a puzzle to me. And I love helping people to use data to make decisions.

How have you kind of done it at the BBC? Have you seen it change the BBC's culture? Do you see that in other companies cultures?

Yeah. Well, I think the first thing you need to do is to get the right data together, which is easier said than done because the temptation is to work with the data that you've got or work with the data that's available to you, which often isn't the right data in the first place. So half the problem is getting the right data. And then you need to give it to people in a way that really works for them. And that's where visualization comes in, in a big way. And just give people a spreadsheet, give them something that's analyzed or interpreted or presented or visualized in a way that they can actually make sense of. So get the right data, give it to people in a way that works for them, which is largely, I think, data visualization. And then if you do that to people, then people use it, which is amazing. So they had problems that they've been facing for a long, long time. Maybe they didn't have the right data to help with those problems. Maybe they had the right data, but it was buried in the boring spreadsheet. But the thing I really love is when you give people the right data, visualize them the right way, you see them start to interrogate it and embrace it and repeat it back in meetings and make use of it through their daily lives. And you can really see that starting to happen. Around some of our big shows like Doctor who, we have set of visualizations that show the audience opportunities for Doctor who in different countries around the world, for example. And one of the opportunities that it showed was in Korea that there are a lot of people who are really interested in Doctor who and really strong opportunities to more deeply engage them in the show, maybe, or to introduce them to the show, because they looked a lot like people who loved Doctor who. But we weren't seeing the Doctor who in Korea as a big business.

South Korea?

Yeah. South Korea. Yeah. So the data was. The visualizations were showing us this great opportunity. We weren't seeing the business there. And so the business people were able to say, well, how do I go get these people? How do I do things differently? And. And for example, they took the cast and crew to South Korea for promotional activities, and they made 4,000 tickets available for people to come and see the cast and crew and just be at this event. And tens of thousands of people applied. Tickets sold out instantly. So they were timidly saying, well, I see the visualization, I see the opportunity. Let me go try and realize it. And sure enough, there it was hidden from us because we hadn't had the right data to see the opportunity before, or we'd had the right data, but we hadn't visualized it in the right way. And there it was. Even for a big show like Doctor who, new opportunities with the right data visualized in the right way. It's really exciting.

Is that kind of proven not successful? You've already seen success. You've already seen some of success from that. I mean, how long ago was that that you introduced?

Well, we've been working at this for about a year at the BBC, at BBC Worldwide. And yeah, we're taking it step by step. So with certain business problems, like what opportunities are there for my established brands, and we've done a pretty good job so far of getting the right data and visualizing it in the right way and helping people to understand it and to use it. And I think we're seeing bigger and bolder decisions being made, and we're seeing new opportunities being found for that. And then we're working business problem by business problem through all the different things that BBC Worldwide does, to say, well, can we get the right data to people and can we make it available to them in the right way so they can use it as part of their day to day? I'm really encouraged by that. As I said before, when you give people the right data in the right way, people are excited to have help to do what they do alongside their skills and judgment. It's great.

It's not just, I think it's on the BBC website actually there was a thing about data visualization being something just style over substance. You don't think it's just a case of style over substance?

Well, I think it definitely can be and I think it maybe even often is, which is why I say it has to be the right data as you can draw beautiful diagrams or beautiful visualizations of all sorts of data sets. But if it's not the data that's pertinent to the conversation in question, then, you know, who cares? In the music industry, you know, a lot of people used to like to visualize, let's say last FM consumption of songs, or visualize at the time, you know, Spotify consumption of songs. But what, five, six years ago there were very, very niche methods of consuming music. So it's lovely that you could do great visualizations from those data sets, but they're wildly unrepresentative of the wider music consuming audience. And so yeah, it was style over substance, or even worse, it was style on misleading data sets that were going to lead you to draw the wrong conclusions. So I think it definitely, definitely can be the case. But like I say, I think the challenge, half the problem is getting the right data together. The other half is then visualizing it in the right way so that it teases out the right conclusions. We had this debate quite a lot internally. There's one particular visualization that I really like. I want to talk about it at the conference a little bit, which is a scatter diagram, lots of points in two dimensional space. They're color coded, which is a third dimension to the data sets. And there's five of them on the page, five of these charts on the page. It's a pretty complex set of data and I laugh at it because it's really not stylish, but it does allow people to pretty easily draw the right conclusions from the data and it's pretty hard to draw the wrong conclusions from that data. A couple of rules I really like to follow. So with just a couple of pieces of advice for how to read this chart, it's not visually beautiful. With a couple of pieces of advice, it's Fun to watch. I remember the music industry. You know, rock stars and DJs look at this and it's not sexy and appealing, but with a couple of pieces of advice for how to read the chart. You could see them to get really engaged in where the opportunity was for their music around the world. Yeah. And it was pretty hard to draw the wrong conclusions. Now the debate is that people want it to be more visually appealing. So I say, okay, well, show me a way to make it more visually appealing. Show me a way to make it more stylish. But it doesn't compromise on the substance. And we've explored all sorts of alternatives over the years and they all look much more beautiful than mine. They really do. The style is great, but you lose something, you lose a dimension from the data, which is really important, or you lose a benchmarking against competitors, which is really important. And, you know. So, yeah, style over substance is a real challenge, definitely. But if you get it right, you can. Yeah, you can help people.

Like a tight line to. Tight line to walk.

It's tough. It's really tough. Yeah, yeah. And there's lots of things that need to come together for it to really work, you know, So I said the right data, they're analyzed in the right way. But, you know, there's a lot more to it than that as well. There are many, many things that need to come together. Many, many tips and tricks to use for it to work and very, very easy for it to fail. Definitely. Okay.

So, yeah, looking at BBC doing a lot with crowd emotion, analyzing faces.

Yeah.

To kind of pick up on what they're feeling about programs. Tell us a bit about that. Are you using it?

Well, this is a great example of getting the right data together, first of all, which is that for years we've been researching our shows and I worked in book publishing and we researched our books, and I worked in the music industry and we researched our songs. So this is a well established thing to do. And we would show lots of songs or books or TV shows to people, and for each person we'd ask their opinion of it, which they're most excited about, all this great stuff. And it led us to wonderful insightful conclusions that made a big difference to all of our businesses. But I think what Crowd emotion came along and said is that on top of that, on top of what people tell you that they think about each of the shows, which we know to be really helpful, we can look at the facial expressions of somebody as they watch the show and we can tease from that More data out about their engagement. So, for example, at BBC Worldwide, we've got some shows that people in the research did not tell us they were excited about. The kind of shows that we didn't want to say that they loved or maybe slightly embarrassed to say that they love Guilty Pleasures. But where by analyzing their facial expressions as they watch the show, you could just tell that actually it was super engaging for them. So, yeah, Dark horses, or guilty pleasures, we call them. So crowd emotion. Teasing that out of the facial expressions allowed us to add a whole new data set into our mix and into our decision making, which posed a really nice challenge on the visualization front, because we already had quite a nice visualization of the kinds of people who are excited about the shows and the kind of demographics and the nature of their enthusiasm to have another set of data. The emotional engagement of people and to overlay it on top was a real challenge in data visualization. So we've ended up doing a slightly separate visualization for that. We ended up adding a whole new set of charts instead of over complicating the ones that were there. So you need to know when to stop, when to break the problem into chunks and visualize each chunk of the problem instead of trying to cram too much into 1. The one dashboard as it is in this case.

How you worked at publishing music. So also quite artistic forms. How do you see data marrying up with. I guess I don't know if artists kind of sometimes see it as being juxtaposed to what they do. They like it. When you kind of go, people kind of like this. What do you.

Sometimes they're absolutely horrified by it. And I think that there's a couple of reasons for that. I think one is that the data they've been given in the past has often been absolutely terrible. So widely unrepresentative data or the wrong data, you know, analyzed in the wrong way. I think in part because sometimes data people don't fully appreciate the artistic creative process and try to step on that with analysis. So I think you can solve both of those things. And when you solve both of those things, I think you end up in a really nice place. So make sure that the data is the right data. It's fit for purpose. First of all, I sound like a broken record on that point, but I can't emphasize it enough. In the music industry, let's say, you know, we used to analyze. Oh, yeah, they used to look at YouTube charts or Spotify consumption, which was lovely, but it was a very small, unrepresentative part of the audience. When we started to research music amongst representative groups of people, you got much richer, much more representative. Read on. The popularity of different songs and the opportunity for different songs. So step one was just overcoming the barrier of saying, look, actually, you know what? This data is telling us something about the opportunity for this song in the world. And then step two is to use it for the right things and not for the wrong things. So we really tried not to say, this part of the song works, this part doesn't, so you should do more of this and less of this. We really tried not to say, here's the ideal formula for a song. We really tried hard to say, okay, you've recorded this song, great. Let us find you the best possible audience for that song, which might be very niche, right? Or it might be very big. But let us find you an audience for your creative work. Let us not influence your creative work. And I think, you know, we all make mistakes. And there's one mistake along these lines I made that still makes me laugh and taught me a valuable lesson about this. And there was a guy in the music industry and he gave us unreleased music from this band that he'd signed. He was very excited about this new band, but no one had heard this music before. He gave it to us to research. We had five tracks, I think, and he wanted to pick the first single for this new band, the way that they would be introduced to the world. So we researched the five tracks in two or three countries, and I remember going back to him with the results and saying, look, we've got a clear winner. There's one song which should definitely be the next single, really outperformed all the others. People were way more open and enthusiastic about this song than the other songs, and with loads of great data visualizations to prove this right. But then he put this smile, this big smile on his face, and I said, what are you laughing at? And he said, okay, well, that's definitely not the lead single. I said, hang on a minute, what do you mean? He said, well, we want to grow this band into something that's big. We want to start small. And so we're going to pick one of the other songs that was way less appealing in the first instance to launch their career to the right audience. And we're going to grow, grow, grow, and we'll save the song performed really, really well and use it later on. And we want the more mainstream audience. And that, for me, is a great use of data to help to refine the strategy but also led by strategies, led by the creative vision of saying, what do I want this band to be and where do I want them to be at different points in their career and how do I kind of weave that journey? And that for me is the coming together of creativity, vision and data to help to support that. I think that if you combine those things properly, then data is not something to be scared of or worried about and it's not something that will tread on the tides of creative vision. Interesting.

Music, books, tv been heavily impacted by digital and kind of the rise of the Internet. You've kind of been dealing with similar problems in each of those when you've come into them. The data's been looking to kind of impose.

It's been very similar challenges, like you say, the transition to digital and the move to a much more global way of working, global way of consumption, and therefore a global way of working as well. So in businesses that used to get away with each country being its own decision making structure, having to suddenly think and behave and act much more globally, releasing products into the world at the same time around different countries, for example, licensing deals that get content to retailers or digital services in a more global way as well. So there's been some real challenges that are similar across all the businesses. I think the fascinating thing for me is that the answer to those challenges has been quite different for each business. So the licensing model and the global ways of working in music are quite different from book publishing and are quite different from tv for pretty good reasons, actually. I think on the data visualization side there. So the similarity is the need to think and act globally. And the difference is that the data sets are very, very different in each company and they're very, very un global. And so there aren't many. Let's take music or books or tv. There are lots of big digital services. So think Netflix, think Spotify, think itunes or the Amazon Kindle Store. But there aren't many that are truly global and that represent the bulk of the market. The itunes download store is big, but only big in certain countries. In other countries, people consume in very different ways. In other countries, Spotify dominates itunes, for example. And in some countries, itunes never got traction and never will. And on Netflix they're rolling out globally, but they're far from being a truly global player. And there's lots and lots of countries that they don't operate in. So as a business, you're trying to analyze data and visualize data in a way that's global. So we can make globally consistent decisions. But you've got different retail partners giving you different data sets in different countries that all need of aggregating and visualizing in a consistent way. So it's really tough nowadays looking across countries and looking across lots of digital services.

How do you think that can solve that?

Well, I think there's a couple of lessons. I think one is that it comes back to my broken record point about getting the right data. And often if you do primary research amongst consumers, amongst a representative group of consumers, you can create a data set that's much more appropriate for the problems you're trying to solve than digital transaction data, for example. So by asking a few thousand people in each country all around the world what they think about Doctor who or the Beatles, you can get a nice representative data set that tells you about different types of people and different types of behaviors. If you just look at, let's say Netflix data on Doctor who or on itunes data, or Spotify data on the Beatles, then you only get to understand that type of person. You only get to understand it in the countries that they're operating in. So by creating globally consistent data sets, like primary consumer research, for example, you can solve the problems of globalization and you can solve the problems of bias in the data sets as well. So we found for global problems or cross country problems, I should say that primary research is much better data set, much richer, much more appropriate data set to use than lots of the data sets that you naturally have at your desk every day. Then I think it allows you to be much more in control of the type of business problems you apply that data to, because you can shape the data to fit the business problem and allows you to standardize the way you visualize that data to the business so that every country has the same way of looking at, let's say, the Doctor who potential audience. And that way as an organization, you can become familiar with that visualization. You can learn more about how to interpret it and get quicker and easier interpreting it. And over time, as an organization, you learn and get more skilled as opposed to every time the business problem comes up. Looking at a different data set with a different visualization and a different way of thinking about it in a different set of caveats. So I think being much more in control of the data and therefore enabling standardize some of the kind of outputs from that data has been a really helpful trick. It's worked very well.

Great. I'm nearly out of time. That's really interesting data visualization, I think.

So look forward to sharing some more tips tricks and particularly some examples at the conference.

Yes.

Wondering whether I should have a little whiteboard here and draw some examples today, but I decided against that conference.

I would have liked some graphics saving.

The conference next time. Whiteboard.

Yeah. Date of his summit in Boston in September. So what kind of stuff will you be talking about there to go with the graphics?

Well, I think you hit on some pretty good topics. We'll use examples from the music industry, from tv, from. From politics. I worked helping Democrats to win elections. So a range of different subject matters, I think. Yeah. Visualizations that have really worked and some of the lessons that apply to them. Some of the ones that have not worked less well. Some of the lessons we learn there. Yeah. And how I think data really can and should be used to genuinely transform the way a business works. Come and see.

Yeah, that'd be brilliant. Cheers. David, thank you very much for talking to us.

Thank you.

Lovely. Thank you.