Skip to main content
Search form
  • 00:00


  • 00:13

    SANJEEVAN BALA: My name is Sanjeevan Bala.I'm the head of data science at Channel 4.So my background has been a consultancy,where I spent a lot of time helping brandsunderstand how to put the customer at the heartof the organization.Then into dunnhumby, where we spent lot of timewith the Tesco program, Clubcard program.My role here at Channel 4 overall is very much

  • 00:35

    SANJEEVAN BALA [continued]: thinking about how can we as a broadcaster, firstly,understand a little bit more about our viewers,and secondly, use that understandingto drive both commercial innovation and alsocreative innovation.A lot of my time is spent meetingwith other parts of the organizationand working with them collaborativelyto understand how could data create valuefor those parts of our business and then

  • 00:57

    SANJEEVAN BALA [continued]: working with our data science teaminternally to then build a lot of the algorithmsand the models that the business can then use to realize value.Globally, we're a very unique broadcasterin that we're owned by the government,but we don't receive any financing or fundingfrom the government, in that we arecompletely self-sufficient and self-financed

  • 01:17

    SANJEEVAN BALA [continued]: through our advertising revenues.So what that means in terms of datais that the more we can apply datato drive commercial innovation, itallows us to be far more self-sufficient in that we canthen generate greater revenue, which we can then put backinto our programming.That ensures that that value that we createand the remit obligations that we have that are set by Ofcom

  • 01:37

    SANJEEVAN BALA [continued]: can then be met.And we're then a far more sustainable business.

  • 01:40

    SPEAKER 1: Now, for the first timeever, Channel 4 is offering individuallypersonalized video ad experiences to our audiences.

  • 01:47

    SANJEEVAN BALA: The data science capability at Channel 4is still relatively new.We started about six years ago.And at the time, we were probablyone of the first broadcasters globallythat started to invest in data and data science capabilities.There was a feeling, something strategicallyfrom our CEO, that data and broadcasting as an industry

  • 02:07

    SANJEEVAN BALA [continued]: were very late to embrace the value of data.And when you look at every other sector,it's very apparent how they're using data.And I think there was a real feeling that broadcastingneeds to get on board.

  • 02:19

    AMY ROBERTS: At Channel 4, I work as a graduate datascientist.Now I'm working full time in the team.Data science, really, a lot of itis not just sitting down and coding, but actuallyinterfacing with the business.So planning projects, trying to develop other projectsin the business, and recognizing areas in the businessthat we could create for each project,and then kind of feeding back.

  • 02:40

    AMY ROBERTS [continued]: So constantly having this feedback loopin terms of what we've developed,what we need to develop, and how we could improve.

  • 02:45

    SANJEEVAN BALA: So in terms of the datathat we capture about a viewer, it's the bare minimum.It's the first name, last name, an email address,date of birth, and then an optional postal code.And that's all the information that we directlycollect about our viewers.And once the viewer volunteers that information whenthey register with us, we then startto understand, whenever they're logged in and watching

  • 03:08

    SANJEEVAN BALA [continued]: any of our content, what sort of content they're watching,when they're watching that content,on what sort of device they watch that content.And with all that information, we'rethen able to understand, firstly, your taste preference.So for example, do you tend to watch more comedy, or factent,for example.And based on that viewing history,we then build a richer picture about our viewers,

  • 03:29

    SANJEEVAN BALA [continued]: which then allows us to then tailor the experience on All 4and give you, the viewer, a far more tailored and personalizedexperience.

  • 03:37

    AMY ROBERTS: Data scientists ourselves, we'renot gathering any data.Our department just gets it kindly handedto us from data engineers who've like packaged it and cleaned itand made it all nice.And we go, thank you.It takes partly intuition but also just playingwith the data.So the first thing everyone will tellyou to do that most people don't do, and to be honest,

  • 03:59

    AMY ROBERTS [continued]: I'm so really bad for not doing, is visualize the data.So first things first, just plot it.Plot it and see what you get.And that will actually give you a huge amountof information just about what variables are correlated,like how the data looks.So is it that you get lots of distinct clumps so youknow that there's three main groups, or is it

  • 04:21

    AMY ROBERTS [continued]: all just spread all over the pageand you know that there's no real strong correlation?It's a lot of little moving pieces.And often, what you'll do is you'll apply several modelsand see which performs best.

  • 04:33

    SANJEEVAN BALA: Our team will builda set of models that will create a set of recommendations.The editorial team will then look at those recommendationsand just get a feel for, is it too narrow,and is it introducing our viewers to a widercross-section of our concerns?And it's the editorial team at the endthat make that final decision.And that's how we ensure that we don't inadvertently fall down

  • 04:54

    SANJEEVAN BALA [continued]: this trap of getting ever narrower in terms of contentwe might be surfacing to our viewers.

  • 04:60

    AMY ROBERTS: We've all been on Netflixand we've seen that thing where we have all the different barsof shows recommended for us.Obviously, Channel 4 is like a smaller scaleand we have fewer programs to choose from, so really, whatwe're trying to do is create a hybridbetween editorial knowledge and what we can do in data science.

  • 05:18

    SANJEEVAN BALA: The vision for All 4that we'll start to deliver a formal personalized experience.But we're forever mindful of what's commonly calledthe echo chamber problem.And one of things we spend a lot of time onis working out and understanding,how do we ensure that once you start watching,let's say, comedy, you don't justget more and more recommendations for comedy?And some of the challenges we're faced with

  • 05:39

    SANJEEVAN BALA [continued]: is how you get somebody that's watching,let's say, Made in Chelsea, to start thinkingabout watching the news.Because that's a big shift.And I think part of the work that we do within datascience is trying to understand, firstly,our public service remit, but howdo we make sure that that gets delivered in the workthat we do with things like recommendationsand personalization?

  • 05:59

    SANJEEVAN BALA [continued]: We use, almost on a daily basis, machine learning techniques.That's everything from clusteringto regression to support vector machines,for example, all the way through to increasingly lookingat applications of artificial intelligence,where we're looking to train models and datasets with labeldata.And then use that information to then enable artificial agents.

  • 06:23

    AMY ROBERTS: In the case of supervised learning,it would be, for example, for us trying to predict interest.So are you interested in cars?Are you interested in cats?Are you interested in beauty products?And they're all things that we'd usedfor addressable advertising.In the case of unsupervised learning, a lot of it'sto do with trying to find behaviorpatterns between our users.So based on people's viewing patterns,

  • 06:45

    AMY ROBERTS [continued]: we can cluster them together and thenwe can say, OK, we believe these people to be in segment one.We believe these people to be in segment two.And within those segments, we can thenlook at the viewing patterns that makepeople common in that segment.So are they always going on the websiteand watching Made in Chelsea?Or are they always watching weird documentaries?

  • 07:04

    SANJEEVAN BALA: And what we're finding increasinglyin the data is that viewers almost create their ownmicro-genres, where they're almost pulling togetherprograms across a number of different genres and they tendto watch them--quite a few viewers will watch the same kind of shows thatspan the same sorts of genres.So that idea got us thinking more about, well,how do we start to bring together some learnings

  • 07:27

    SANJEEVAN BALA [continued]: across these different genres?And what does that actually mean for the way in which we presentcontent to our viewers, and also,how we think about the experience of our viewers?Some of the ways in which we've usedthe data to drive commercial innovationis around personalized advertising.That's when a user will hear or see their name in the ad

  • 07:47

    SANJEEVAN BALA [continued]: creative itself.

  • 07:49

    SPEAKER 1: George here is watching the latest episodeof Gogglebox.Good, isn't it, George?Then in the break, up pops an ad for our humble milk carton.But what's this?Is that your name on the side, George?Brilliant.Go share it on Facebook.You're famous.Meanwhile, Sophie is out and aboutwatching Made in Chelsea on her mobile.Again, our favorite milk carton makes an appearance.

  • 08:09

    SPEAKER 1 [continued]: But this time, Sophie's shown whereshe can buy it from locally.Isn't that handy, Sophie?

  • 08:14

    SANJEEVAN BALA: We're always mindful of notover-personalizing the advertisingso that it loses that relevance and losesthat sense of creativity.

  • 08:22

    SPEAKER 1: We know that being too personal in advertisingcan feel a bit creepy.That's not what we're into, even if you are.Our data promise to viewers will always take priority.We'll never release their data or use itwhen they don't want us to.

  • 08:35

    SANJEEVAN BALA: Very, very recently across Europe,there's been the whole GDPR regulationthat's been rolled out, which, in essence,aims to give consumers far greater controlof their own data and encourages brandsto be far more transparent and open about whatthey're doing with their data.And we take privacy and consumer privacy very, very seriously.And we built a strategy around two key principles.

  • 08:57

    SANJEEVAN BALA [continued]: First is around transparency, in thatwe'll be really open and clear, and in plain English,explain to viewers what we're doing with their data.And the second is around control,in that if there's anything we're doing the viewer is notcomfortable with, they have full controland they can opt out of anything that we're doing.The data we receive from our viewerswhen they first register, that goes

  • 09:17

    SANJEEVAN BALA [continued]: through a process of, firstly, encryption,so that no one can get access to it very easily.And secondly, anonymization, in thatwe convert that information into a randomized number.Because we've been so successful commercially, we'renow starting to evolve our thinking around,how do we start to affect the organizationand how do we start to support some of our creative

  • 09:38

    SANJEEVAN BALA [continued]: and commissioning teams?And very much working hand in hand with those creative teams,how do we get the best of both worlds in terms of the datascience and the analytics, but alsothe creativity that we have in abundance at the channel?


Sanjeevan Bala, head of data science, and Amy Roberts, data scientist at BBC Channel 4 in the UK, discuss how Channel 4 is using machine learning and big data to bring a personalized experience to their viewers.

Looks like you do not have access to this content.

Embracing the Value of Data in Broadcasting: Channel 4

Sanjeevan Bala, head of data science, and Amy Roberts, data scientist at BBC Channel 4 in the UK, discuss how Channel 4 is using machine learning and big data to bring a personalized experience to their viewers.

Copy and paste the following HTML into your website