In this episode of The Analytics Edge (sponsored by NetSpring), we’re joined by Sanjeevan Bala, Group Chief Data & AI Officer at British media giant ITV, for a discussion on Data and AI in the Media & Entertainment industry. Sanjeevan shares how being data-driven allows ITV to drive business outcomes across every function: doubling MAU (marketing), doubling viewing hours (product), and doubling digital revenue (commercial). To achieve this, he discusses the need for both business and data literacy, delivering self-service or “silver service” to the business, last mile challenges like shared segmentation or unified metrics across functions, and how the cloud data warehouse helps address many of those challenges. Tune in as Sanjeevan shares his team’s biggest wins, including the impact of AI on contextual advertising, and how AI will continue to disrupt ITV and the Media & Entertainment industry.
In the Season 1 finale of The Analytics Edge (sponsored by NetSpring), we are delighted to have Sanjeevan Bala, Group Chief Data & AI Officer from the British media giant ITV, on the show to discuss the role and impact of Data and AI in the Media & Entertainment industry. Sanjeevan goes in-depth on how ITV's commitment to being data-driven has propelled the business forward. He shares ITV’s targets for doubling monthly active users (MAU) through marketing activities, doubling viewing hours within the product, and doubling digital revenue on the commercial front.
In order to achieve these objectives, while we often talk about the data literacy of business teams, Sanjeevan exposes the need for business literacy on his data teams. This helps him determine the level of self-service vs. offering the “silver services” of an embedded team of analysts. Either approach presents their own set of last mile challenges, and he explains how a cloud data warehouse and having a single source of truth, can help align business teams around a shared view of customer segmentation and unified definitions of cross-functional business metrics.
Join us as Sanjeevan shares the triumphs and challenges for Data and AI at ITV, and leaves us with his expert predictions on how AI will continue to disrupt the Media & Entertainment industry!
Sanjeevan Bala is a leader in the field of data and artificial intelligence, currently serving as the Group Chief Data & AI Officer at ITV. Recognized as the Most Influential Person in Data on the DataIQ 100 list for 2023, Sanjeevan has over a decade of expertise in the Media & Entertainment industry, having contributed significantly at ITV and previously at Channel 4. With a bachelor’s degree in management & computing from King's College London, his influence extends beyond his current role, as he holds multiple board and advisory positions, including engagements with Bakkavor, Evanta, and DataIQ.
“Across Product, Marketing, and Commercial you sometimes get very verticalized KPIs. For example, Marketing often will look at cost per acquired user. Historically, the conversation would be we’ve acquired them – they don’t go and watch something, that’s not our issue! But increasingly what’s happening is Marketing will look at cost per acquired hour of a viewer. We start to join up parts of the organization with these unified metrics, driving the right kind of behavior thinking about the next step in that journey.”
- Sanjeevan Bala
(Segment 1) (:56) Challenges
(1:29) Career journey in data leadership
(2:56) Integrated publisher/broadcaster model
(4:20) ITV’s unique data requirements
(9:38) ITV's modern data stack architecture
(12:02) Multi-cloud strategy for different workloads
(13:05) Challenges to making data-driven decisions
(15:25) How data warehouses address last mile challenges
(Segment 2) (17:07) Solutions
(17:30) Organization of data team to support business units
(20:44) Standard data models in Media & Entertainment
(23:26) Behavioral analytics and segmentation
(27:10) Data gaps in product analytics tools
(31:42) Impact of the data team at ITV
(Segment 3) (34:50) Business Opportunities
(34:52) The future of AI at ITV and for Media & Entertainment
(Segment 4) (39:53) Takeaways
Announcer: [00:00:00] Hello, and welcome to the Analytics Edge, sponsored by NetSpring.
Tom: The Analytics Edge is a podcast about real-world stories of innovation. We're here to explore how data-driven insights can help you make better business decisions. I'm your host, Thomas Dong, VP of Marketing at NetSpring. And for today's episode, my co-host is Vijay Ganesan, co-founder and CEO at NetSpring.
Thank you for joining me on the show today, Vijay.
Vijay: Great to be here, Tom. Super excited to talk to Sanjeevan you know, a renowned leader in the data space, and very excited to have this episode.
Tom: Today's topic is data and AI in media and entertainment, and we're very lucky to have one of the industry's foremost experts, Sanjeevan Bala, Group Chief Data and AI Officer at ITV, on the show with us.
For those of you outside the UK, ITV is a British media giant consisting of ITV studios. which creates and owns and distributes content in the UK and globally, and their media and entertainment division, which distributes content across their traditional ad-funded broadcast TV [00:01:00] networks and ITVX, their free ad-supported streaming platform.
So I grew up with black and white TV, so it's not hard to recall a time when media companies had to read reviews and interview customers to gather customer data and gain very limited insights. Now today, across all devices and media types, companies can track clicks, views, engagement, and more, thus creating massive amounts of customer data.
Sanjeevan, you are recently recognized as the most influential person in data. Can you please share with us your career journey as a data leader, how that has led you to your current role as Chief Data and AI Officer at ITV?
Sanjeevan: Absolutely, it would be my pleasure, Thomas. So I guess broadly, I described my career track and journey so far in four broad stages.
I very much started off in, in consulting, management consultancy and you really learn around things like stakeholder management, client management skills, and, and the ability to kind of consume information really quickly and traverse different sectors and be very much credible with clients. I then moved into startups and scale-ups and this allows [00:02:00] you to sort of put a lot of the theory into practice because as a consultant.
You're not always there to see it all the way through, whereas when you go to the client side, you really do think about some of these scale-up challenges. And the thing with the startups is you learn very quickly how you might take a five-year plan and execute that within five months. The whole hyperscale challenge becomes quite a different way of thinking.
From there, I then moved into client-side roles, really into large corporations. And that was very much around how you drive value using data and drive the innovation agenda. And the latter part of my career very much moved into non-exec director roles and advisory roles. And that's when you're looking at driving value end-to-end in businesses.
And that's very much the path I've taken through in terms of my, my career journey to date. Great.
Tom: So, I guess ITV would be an example, one of these large corporations. You're clearly dealing with massive amounts of data. Coming into your organization. Could you please describe ITV a little bit more for us and your company's business model?
Sanjeevan: Absolutely, Tom. So, ITV is an integrated publisher [00:03:00] broadcaster. And so effectively in our value chain we produce content. Promote it, distribute it, and then monetize it. And there's really two parts of our business, if you will deeply integrated. So, for some of the international audience you might be aware of some of our content like Snowpiercer on Netflix, or Love Island USA, for example Gordon Kitchen Nightmares, again, another one of the ITV Studios shows.
So, effectively, we produce content in our studio's business, and studios will then sell that content into international markets, sometimes to streamers. Like Apple TV, Netflix, Amazon Prime, for example, and other times to other broadcasters in other markets. And then on the UK side, we then have a network, a TV station, and an ad funded business model, where we have ITBX, which is our streaming service, and also a subscription based model as well.
In some markets, you'll also be aware of Britbox, which is a joint venture where ITB makes our content available in international markets, certainly in the US, and that's the best of [00:04:00] British available for a streaming subscription service. And that's very much the business model around ITV in terms of an integrated publisher broadcaster.
Vijay: So, ITV, I understand, has over 90, 000 hours of content and about 12. 5 million monthly active users. That's a lot of viewers, a lot of content. As a publisher broadcaster and streamer, what are ITV's unique data requirements?
Sanjeevan: Yeah, Vijay, the way we looked at this very much is we linked value and value outcomes in the organization and use that to then drive our data requirements.
Now what I mean by that is earlier on I talked a little bit around our value chain. So we promote, produce, promote, distribute, and monetize sort of content, if you will. And within that and a layer underneath that is we then set about quite ambitious goals. And we talk about the double, double, double KPIs.
So effectively what we're saying is we're using data in our strategy to effectively double the number of monthly active users. So that's all around [00:05:00] how do we acquire viewers onto the platform. And the second one is doubling viewing hours. So once we've got them on platform, how do we ensure the content they're watching is relevant to them?
And how do we ensure they watch more than the show that they just came in to watch? And the last one is doubling digital revenues. So with all that incremental viewing. How might we use data to then drive innovation in terms of advertising, in terms of ad products we might deliver. And those three things then lead you to then drive into our data requirements.
So I'll give you some examples, Vijay. On that first one it's very much talking to the marketing part of our organization. So how might we use data in marketing to promote our shows off platform? So for example, on Instagram, TikTok, or other social media platforms. So we can tell viewers about our content.
on platforms that they're currently engaging in and using. So that's an example where marketing would enable via a customer data platform, they could build audience segments, and then promote our shows off platform. And that allows them to attract viewers back onto [00:06:00] ITVX to then view content that we're promoting.
Another example where we're using data would be in our commercial side. So, talking a little bit about our data requirements, and some of the things we're doing in, in the more advanced side of our business is how we're using artificial intelligence to drive commercial innovation. So, an example here is let's say you're watching Love Island, and coffee is spoken about very, very positively in, you know, in the first part of the procedure episode.
What we've got is we have an engine that's watching the content, listening to the subtitles, And reading the audio file, and it can understand the fact that coffee is spoken about very, very positively, which then allows us in the first ad break to put a coffee ad in. The reason that's really important is it allows advertisers to become much more relevant to the content.
That means the, the, in terms of brand metrics and spontaneous brand recall metrics, they outperform given sort of a data driven method of targeting. So these are some jobs of innovations that we're using. It's driven the data requirements, but ultimately it's [00:07:00] about driving those business outcomes around the double, double, double KPIs that I talked about from the value based perspective
Tom:.So you guys are both a broadcaster and a streamer, when it comes to collecting data, is there a difference in the type of data you can capture in both channels, or are they pretty much consistent? You're capturing the same level of information about
Sanjeevan: viewership. So there are differences, Tom. So, for example, in our, in our direct to consumer streaming model you know, there's a registration mechanism, so viewers will sign up to the service, they'll provide information like an email address, so you can capture more about the viewer, if you will, the actual viewer that's consuming content.
On the network side, because there's still a lot of the content that's consumed through broadcast media, it's not as easy to capture information, right? Because there is no mechanism necessarily in traditional broadcast mechanisms to capture some of that information. It is getting better with the run of smart TVs and smart Services, for example, increasingly, there's an [00:08:00] element of registration so you can capture information about the viewer.
And that allows you to then bring together who the viewer is, with what they're watching, and why they're watching through our research panel. And we bring those three things together to then drive a lot of the insight and the outcomes we're looking to effect.
Tom: And are you able to attribute A single user across both, i.e. can you connect the two?
Sanjeevan: In some instances, it's possible, but not across the entire estate. So it very much depends on the environment in the home, and how smart that environment is, and whether there's a feedback loop back to you. So technology service and therefore a feedback loop that allows you to match those two things together.
Vijay: Great, I love the coffee example and ads based on, you know, parsing the content of the show. Reminds me of what Amazon Prime did last Thanksgiving at a football game where the ads were Personalize based on your prime mem, you know, purchase history. So I think it's the innovative way of personalizing [00:09:00] not just ads, but also eventually even content with generative AI that, you know, everybody gets a version of the show that's tailored for their, for their taste.
So, that's brilliant. And I love the double, double, double methodology. Just it is so catchy and it's so relevant and it keeps the company focused on on the, on the key metrics. So I love that. What does your stack look like at ITV? You know, you've got some fairly complex data requirements, a complex business.
How have you architected ITV's data stack?
Sanjeevan: Yeah, and I think I think the way we approached it is we, we first looked at Sort of the prevailing organizational culture, if you will, across the organization. And I think, like many organizations, what we found was, you know, we had microcultures across the organization, like, yes, there's an overarching culture in ITB.
When you compare and contrast the commercial team versus the commissioning and scheduling team they're quite different culturally. The second big consideration was the fact that in those sort of different [00:10:00] areas We very much work on a model where we are empowering the business to become very much self sufficient and can sort of run very, very quickly and go end to end.
So what that effectively meant was we implemented a data mesh architecture, which effectively meant a lot of the technology, the way the data was structured, the way the data was governed would be decentralized at the outset. And we use cloud environments, we use AWS for all of our data and processing cloud.
So then we have other components of the stack. So we use Databricks for a lot of provisioning and usage in terms of building data pipelines. And we then have a combination and a variety of last mile technologies. And then allow the business to access that information and access that information. So for example, you know, we have things like Tableau, ThoughtSpot, and Looker that allows a classical kind of BI and MI to be generated for the business.
And then within the product environments you know, they've used ContentWise for recommendation and Amplitude for product multivariate testing and, and, and sort of experimentation. Over in Data Governance, when you think about how you govern your data, we use a product called [00:11:00] Alession for how we want to govern data, make sure it's documented, it's accessible, and stuff discoverable.
So there's some of the component parts that pull together to create the environment that allowed us to kind of run in a very decentralized sort of operating model.
Vijay: One of the things we hear a lot is Companies your size and scale are invariably multi cloud. I mean, there's nobody ties themselves to one particular cloud.
You know, either by design because they don't want lock in or because, simply because of acquisitions and so on, you pretty much, everybody has every cloud now, right? Or at least the three clouds. And you said something interesting here about how you're multi cloud, but you use AWS for data. So, is that the model where for certain workloads, you use certain clouds if it's, you know, data, it's, you centralize on a single cloud.
Is that the approach?
Sanjeevan: No, I think we're no different to some of the clouds you talked about, Vijay, right? So, so we do run with all three cloud platforms. And I think it is sort of use case dependent is, is kind of where Vinesh is at. We took a very conscious [00:12:00] decision, you know, that that would be in AWS.
Other things like video processing, you know, it's held elsewhere. So there's different component parts where we've made very conscious decisions. And I think what we're realizing as there's greater commoditization happening with all the cloud providers is you can now take sort of display parts where they've sort of had You'd have slightly more advanced pages, technically.
And you can still run those across multiple cloud environments. And that's kind of the model we're moving much more towards.
Tom: So you talked a little bit about some of those last mile tooling that you have in your stack. Reporting and BI, you've got like things like Tableau and ThoughtSpot and Looker. And you have a number of end user tools.
So I'm curious what you have seen as the primary challenges or obstacles for your company to be making more data driven business decisions.
Sanjeevan: One of the first things we sort of spent time on is You know, what do we mean by being sort of data driven, right? Because effectively we're, we're a very creative organization.
So the most extreme, you know, we're not describing, we're going to be [00:13:00] doing kind of color by numbers, right? Because there's the whole creative organization that's really been set up and that does that very, very well. The second aspect is the cultural change that's needed, you know, to become much more data enabled.
And that's, that comes from a couple of areas, right? There's some parts of our organization like digital ad sales or digital marketing that are incredibly comfortable with digital and using data to optimize. There's other parts of our organization that this is going to be quite new for. So for us, the way we looked at this is, is really thinking about What's the right level of maturity across the organization?
I talked a little bit around different microcultures across the organization and what does that mean? And when we set up things at the last mile, it was really thinking through what parts of our organization are ready for self service and what parts of our organization might need a silver service model.
And what I mean by silver service is where perhaps they are earlier on in that journey and they probably need a little bit more hand holding. from analysts and data sort of data leadership teams to kind of help [00:14:00] them sort of work through and understand the potential of data and how that can then drive sort of decision making in the organization.
Other aspects that have been quite challenging, I think, have been around, you know, where do you contain a single source of truth? So when you think about things like segmentations, for example, right, there's obviously the need to have sort of enterprise wide segments and standard definitions and names.
But increasingly, as you get into the organization, marketing versus commercial will have a different need to have different set of segments and sub segments. So the way in which you need to kind of orchestrate a lot of these things becomes quite material and that can become quite challenging around how do you make sure there's a common language that the organization uses, but it allows a unification at an enterprise level, but it also allows bespoke sort of functionally specific models and understanding to happen that will then empower a particular functional area in the organization.
Vijay: How do you see the cloud data warehouses and what's been described as a modern data stack? Addressing some of the last minute [00:15:00] challenges, you mentioned self service, you know, consistency of results, common definitions of things, even real time insights.
Sanjeevan: There's a few themes that play out here that become sort of increasingly complex to manage, right?
So there's the whole sort of access to data, source of truth, you know, where should it sit? Is it source aligned or do you put it into the lake or the warehouse? And there's elements around security and governance, right? So how do you make sure that access controls are in place? How do you make sure it's governed?
Who has responsibility for how data is governed? And think about business ownership and do business leaders understand their role in owning and governing the quality of the data? And what I think the big shift is going to be in terms of how these things are manifesting themselves is fundamentally, I think the model historically has been around, you know, moving a lot of the data.
To where the modeling happens. So effectively you're moving everything to the center to, to sort of run a, a quite advanced compute computational capabilities. But increasingly what you're seeing is some of these last mile tools, [00:16:00] like a customer data platform or multi barrier testing tools is increasingly there's, there's the need to kind of move the model to where the datasets so much more into the edge environments.
And I think what you're starting to see is the emergence of companies sort of almost acquiring AI by stealth where, for example, the CRM tool that they bought. And it has a set of AI capabilities that can create segments, that can do content creation in terms of, you know, text creation for your eCRM emails, for example.
I think that's where the model is shifting, moving much more away from sort of everything being completely done in the center to a more federated model where things are being enabled in the last mile. Some of the cloud capabilities need to be able to enable that and then feed that into the last mile solutions that you can then buy off the shelf.
Tom: Let's switch gears here a little bit and want to really deep dive into kind of how you've set up your organization. At ITV. Curious to know how you set up your data team and how are you collaborating with the various business units you mentioned? I think you had marketing you had your commercial team[00:17:00] for example.
So how do you collaborate with them on, let's see, the art of the possible, everything they could be doing with data and AI today?
Sanjeevan: Yeah, that's a really good point, Tom. So I think earlier on we talked a little bit around those sort of the microculture and a decentralized federated model. And so the way we set everything up is one of the principles we felt was really important is that we do want to build an ivory tower in the center, right?
Of all these specialist data skills that are slightly business. So one of the key principles was that the team had to kind of go native. So they need to feel part of marketing, or product, or commercial. They, they, they wouldn't, they shouldn't be part of the data unit per se. What that then meant was, we had to enable these sort of squads, cross functional squads to be able to go fully end to end.
So that meant everything from identifying kind of a use case or a value case, if that's the answer possible, what could data do? Getting that sort of buying by a business stakeholder around here's how the outcome will be achieved and how's the outcome drive one of those double, double, double[00:18:00] KPIs.
It was always linked back to one of the KPIs. But effectively, we put things like data engineers, data analysts, data scientists in those business units because it's really important that you go end to end. The beauty of that model then means everything gets aligned very nicely because the team are then working on things the business units really want and care about.
It's a commitment from the business around the last mile adoption, be that business process changes or technology that needs to be put in, in order for value to be fully realized. And it also means that when one particular area is having some challenges, let's say if, you know, ad sales market is changing and it becomes structurally challenging.
The team also feel that pain, right? So there's a complete alignment in terms of the culture, how that team are operating, why they're there, their purpose, their remit. And that creates a great environment in which those teams can then co collaborate with the business stakeholders and they effectively become part of the business units.
Tom: Yeah, that's fantastic. So it sounds like your data analysts are actually subject matter experts. They actually really understand the data of the [00:19:00] business and how to answer the questions that the business teams will actually have.
Sanjeevan: And I think there's a couple of points to build on that, Tom, actually.
So I think there's an element of, and you often hear in the, in the industry, you know, the need for sort of, you know, data literacy programs, right? How do you, how do you train the business to kind of better understand data? The flip side is also true, right? Because it's important to run business literacy such that when you do put analysts into marketing, They're already sort of conversant with the language of marketing, right?
They understand ROMI or ROAS and some of these metrics that, you know, a marketeer would look at. So things like cost per acquired viewer or cost per acquired viewer hour, for example. So it's important that you sort of do all that, some softer training before you release these teams into the business unit.
Otherwise, I think you get a bit of a chalk and cheese kind of experience. It doesn't quite work. So you have to invest on both sides, I think, to get that product to work well.
Tom: Yeah, exactly. So there's very well defined business metrics that the teams are going to be thinking about and how they model the logic of their business.
So I would imagine that this is quite important when [00:20:00] you're dealing with this last mile self service tooling, package tooling that may be available. So just curious, is there perhaps a standard data model in the media industry today for all of your end user application vendors that they're adopting for your business?
Sanjeevan: There's not one that's merged. Because I think there's, there's some nuance and differences between different media and entertainment organizations. So, for example you know, not all broadcasters have a studio's or own studio's IP and content, for example. So I think because of that, I think there's some nuances, which means there hasn't been a standardized model.
If you step back a little bit, Tom, and you think about the ecosystem that we all play in, you know, we all have viewers. And therefore there's certain requirements around what information about a viewer do we want to capture, be that name, email address, you know, location, some of that information. Depending if you're ad funded or not, you then have the notion of an advertiser, right?
So we capture information around what advertisers and what kind of advertising is being sort of sold and therefore what advertiser we're matching up against a [00:21:00] viewer's viewing kind of behavior. And then you have producers or the labels that we work with. So I think what you do see in the market is, I think those ecosystems are very, very similar.
I think increasingly, everybody in the media and entertainment world are looking to kind of identify who the viewer is, matched with what they're watching, matched with why they're watching with the research panels, and they're increasingly looking to integrate those three things. Some of the trends you do see happening, though, is, you know, increasingly, while everyone has got an understanding around who the viewer is and creates an identifier for the viewer, I think organizations and media entertainment companies are trying to better understand, at a more nuanced level, more about the content.
So what I mean by that is, historically, broadcasters would describe content in genres, so factual, entertainment, genres like this. Increasingly, the viewer doesn't always think about content. through a genre lens. So there's a lot of tools that are available that allow you to kind of really create nuanced understanding about the content.
So it could be [00:22:00] around the mood, the show, the narrative arc, and it's much more nuanced understanding through content metadata that we're starting to understand content. That's definitely happening across the industry where all broadcasters are now trying to much deeply understand some of the signals and some of the profiles around the content that they're broadcasting, they're using.
Vijay: Yeah, that's, that's very interesting Sanjeevan about you know, how nuanced a content is. It's not just about, okay, a genre or You know, there's you know, age segmentation. It's really about things like mood. That's fascinating. It's the importance of of things outside of just product interactions that are important in in, in behavior analytics and segmentation.
How do your product teams do behavior analytics and segmentation today? How much of it is self service? How much of it is Silver service, as you call
Sanjeevan: it. I think we've been very fortunate at ITV, so we have a very product oriented team that sort of run ITVX. And so what you tend to find in teams that are very sort of [00:23:00] product oriented is they really are very digital and experimental by nature because, you know, they understand, you know, product development, and they're constantly like, looking to make sort of, I guess, fact based decisions.
Around what features going to work, what isn't working, what isn't working for a particular segment of our viewers. So what we find in certainly within ITV is product team are very, very self sufficient, right? So they will use tools like Google Analytics to understand how the viewer flow. They'll use tools like Amplitude, for example.
So these are product managers that will be actually using these tools. To better understand, you know, user journeys and user flows. Where it gets really interesting is product often, and product management teams will often create their own sort of sub segments. So, an example is, you know, at an enterprise level, we have a segment called domain streamers, which is a group of viewers that have a familiar relationship with ITB content, and they tend to come quite often.
And that's quite a large homogenous group. What the product team will look at is, so it's [00:24:00] great that we can see them coming in, but actually they want to understand of those viewers how many are an ad funded viewer versus a subscribing viewer. They want to go down to another level further from that.
What sort of content are they watching and can they create content segments and subscription based segments to better understand the nuance of their particular journey? Because what you find is that then helps shape and inform product features and innovations that the product team are looking to do.
So in ITV, we tend to find the product team are probably a lot more self service, at the self service end of the spectrum. And so all the tooling and all the enabling we have put in place needs to make sure it's very much driven from a product outcome perspective. And therefore, a lot of the tools that you're putting in place, allowing that commoditized access to that data and information, allows the team to manipulate themselves, and critically, allows the team to then activate against themselves.
So, an example again might be, let's say a tool like Amplitude, where they're seeing certain segments coming through, and perhaps dropping out of the flow. The team would then like to create that [00:25:00] sub segment that drop out as a segment and then promote them on a marketing channel and perhaps tell them about some of the new features that are coming up or new content that might be coming out when they're dropped out of a particular flow.
Those are examples where you're joining up a segmentation or a micro segmentation and being able to use that across multiple rath mile applications and tools.
Vijay: So there is a centralized Definition of certain segments like mainstreamers and then individual teams can then create micro segments after that.
Sanjeevan: Absolutely right, Vijay, because I think what you find is there'll be a few macro ones where there's a common language, the whole business kind of gets behind it. It typically is used to define sort of product market fit around what, what segments we're going after. But what you then very quickly find is each of the functional areas of our business, be it marketing.
Commercial or product have a different way in which they interpret the viewer, right? And they're trying to achieve a slightly different outcome against those double, double, double KPIs. And therefore, you need to provide that flexibility in the business so they are empowered to then do that level of more detailed [00:26:00] experimentation that still hangs off the main, main groups, but it's important they're able to kind of go down a level and much more in a much more data driven way, shift the needle.
Tom: So I would imagine that as you have different teams, you know, building their own definitions sometimes they're going to be requesting data that doesn't exist. How do you, how are you handling that challenge? I would imagine that there's certain data that may not be in a product analytics tool that may sit in.
Maybe some subscription data that might be over in a finance tool. How does your team support that through your silver
Sanjeevan: service model? Yeah, absolutely, Tom. So I think because we went down a data mesh kind of architecture, one of the other principles we adopted very quickly was data as a product. So what the teams do in when they're sitting with business users and business leaders, is they're building data products that are fundamentally then feeding a lot of the value we then create.
So an example you gave, What actually happened is where a particular product or a particular data that's not available is needed, the [00:27:00] product team will either use their existing data engineers, data analysts to go off and build that data product by accessing some financial information, or they'll put a request into the finance team, who may already have a data team that's working within them, and then commission the work to be done to then build that particular data product.
For example, a subscription data product. Once that product is built, that then interconnects with the rest of our ecosystem and then becomes available in terms of the output that the teams can then build other models or higher order products against. So because we've gone down that data as a product model it means everything gets atomized at a much more lower level and products are then built, maintained and managed very, very actively to the link through to the value and the outcomes that we can affect.
Vijay: Do you have any examples, Sanjeevan, of insights that your product teams have been able to unlock using data?
Sanjeevan: Yeah, actually, so, some of the, some of the outcomes that developed, for example, is, you know, when we think about recommendations and personalization, that's a really good one, actually. So in the past, [00:28:00] what would sometimes happen is we might recommend a show where we you know, have invested quite a lot.
You know, time, effort to kind of, you know, commission a particular show. And we still, we still do elements of that, but actually what the teams are finding through the recommendation work is actually, for some segments, where they already are very familiar with the ITV content, it's about increasing the breadth of their relationship with us.
So what I mean by that is, if you think about our complete content portfolio, there's different genres, but within the genre, there's a lot of depth. Some of the experiments that the product team are looking at is, are there segments they can increase? the number of genres a particular viewer watches in.
And the reason that's important is the more genres you tend to have a relationship with us with, the stickier you are in terms of your relationship with us. So, for example, where one of the objectives might be for certain sub segments, how do we increase the average number of genres that you're watching and you have a familiarity with ITV?
In other examples, there are some viewers that come in and watch one particular type of content, so some of [00:29:00] our long running shows like Corrie, for example. So for those viewers, it's a strategy to get them to watch different content, because we know they come in and they religiously watch Corey, or it's actually a strategy to superserve their current need and therefore not distract them with other kind of shows and other kind of content.
So this is an example where you have a very different strategy based on each of the different segments you're working with, but it's all through the way in which product understand the viewer, the content, and the product experience, and that then allows to try these differentiated strategies.
Vijay: I'm curious how how granular these micro segments are.
Just some rough scale, are we talking about thousands of users in a particular micro segment? How granular is this?
Sanjeevan: I mean, it could go all the way down to that level. I think the trade off often, so with the capability exists, shall I say, in terms of the customer data platform, the way the teams are using Amplitude to go to that level of granularity.
But I think one of the things they always [00:30:00] look at is effectively when they're running an experiment, they need to scale up the finding to then really ship the needle, right? So there's no value in going all the way down to kind of complete micro forensic segments, right? Because the application, when you scale that up, you're going to lose a lot of that sort of benefit and that value.
So they tend to have quite large groups they still work with. The capability does exist. They can go all the way down and become quite granular. But I think it's that case of how do you ensure that the testing experience you run are sizable and scalable enough that you can actually affect one of those KPIs we talked about earlier on.
Tom: Yeah, that's really interesting. The the level of insights that your product teams have been able to uncover through the data. Stepping back on, you know, your data team obviously supports the entire business. What are you most proud of that your team has delivered over the last couple of years that you've been with ITB?
Sanjeevan: Yeah, I think some of the So surprising benefits have been around sort of, I guess, organization, agility, and what we describe as learning cycles. So,[00:31:00] one particular example that I think, certainly we're very proud of is the whole idea around historically for marketing. It used to take about three months to sort of pull together a segment, identify what segment needs were, and then be able to promote our shows off platform, so on, on social media, for example.
Because a lot of the data products that were developed, a lot of the last mile tooling we put in place, we've now taken that down from three months down to three minutes. If you think about that, that order of quantum change, what that then means is We can now start learning about how accepting some of our marketing promotions are in three minute cycles, right?
So as we're promoting a particular show, if it's not working particularly well on, let's say, Facebook, we can start to optimize and change that. We can make differences in terms of the creative execution, the copy we might use, the targeting we might use. That ability to kind of start to rapidly go around in those three minute learning cycles has been a tremendous benefit in terms of how we [00:32:00] continue to kind of test and learn on that in that part of the sphere.
Another example is how we've moved from historically what you sometimes see is when you're working across product marketing and, and commercial is you can sometimes get sort of very verticalized KPIs. So, for example, marketing often will look at cost per acquired user. And, historically, the conversation will be, well, we've acquired them, then they go and watch something.
That's not, that's not our issue. But increasingly, what's happening here is marketing will look at cost per acquired hour of a viewer. Not only will it drive acquisition, they really worry about and think about, have I acquired, and they're going on to watch. And that's another example where we start to join up parts of the organization.
With these unified metrics, but it's driving the right kind of behavior. It's not just worried about acquiring and reducing the cost per acquisition. You're also thinking about the next step in that journey. So do they actually acquire and do they go on and watch? That's not a great example. And the last one I'll touch on this is just [00:33:00] around diversity, right?
When you're, when you're looking at these business problems. One of the key strengths a data unit can have is having diversity of thoughts, right? So that's how we're recruited and we're incredibly proud that we have about 51 percent female roles across all the levels across the data unit, which means we've got a really nicely diverse team across other diversity markers.
Which means as we're solving these problems, we're bringing in a lot of outside in thinking. So a lot of our teams in the business have come from outside the media and entertainment sector. So we're getting, you know, leaders from retail environment, from financial service environment. And that creates an incredible melting pot of ideation that allows us to unleash that onto the business.
Tom: Yeah, that makes a lot of sense to bring in experts across different industries, obviously, different modeling, different data requirements, especially in this world of AI, we're continuing to see an evolution of the technologies there. I thought we end this show here with you know, your, your view of the future and how you're seeing AI within ITV.
[00:34:00] Impacting and changing the landscape, not only maybe for you, but maybe for the industry as a whole as we move forward
Sanjeevan:. Yes, I think, I think there's a theme that's coming through at the moment. I think what we've seen so far is the prevalence of these large language models, as an example, and whether there's a lot of focus on sort of generative and all of those areas.
I think what we're, what we're seeing increasingly, though, when you look at traditional businesses, is there's tremendous opportunity in any activity that's around forecasting. So that can be where you're trying to forecast how successful a show might be through to forecast the likely demand for a particular show through to forecast a particular demand from an advertiser.
Let's say the second big process that happens in a lot of media and entertainment organization is how do you optimize that so effectively? You know, you've gotta a set of content where you predict audiences are gonna be watching games. You've got some demand from advertisers, they're looking to advertise against particular content, and you're optimizing what ad goes against what show that's hitting what particular audience.
And that's the big [00:35:00] optimization challenge, right? So these big macro processes around forecasting audience performance and content performance And optimizing are absolutely right for, you know, AI to be applied across them. And we have a number of projects underway that's enabling AI across these things.
And that'll create a huge amount of value because you effectively optimizing, you know, the value of your, your, your, your IP effectively. I think as I touched on earlier, I think a lot of organizations will see AI by stealth because I think as you look at last mile. Adoption of a lot of tools that are being embedded in the business.
Increasing those vendors will start to apply and broaden more horizontally and include AI capabilities within that. That means organizations can benefit from that. And I think the other part, I think increasingly organizations will find is as more of the cost to produce goes down, the real challenge will shift from not so much around production per se, but around search and discovery of content.
I think we'll see a model where there'll be more content [00:36:00] available. Then there is sort of, you know, supply or demand as, as it were, and how do you best match and how do you solve the problem where viewers are trying to search for content and, and find content when there's so much content available. I think when there's a surplus of supply, the shift and the challenge I think municipal will face is how do you then solve the search and discovery problem?
Vijay: Can you share some examples of you know, the impact AI has had at ITV especially in terms of monetization and revenue?
Sanjeevan: Yeah, absolutely, Vijay. So I think I talked earlier about the contextual advertising example, you know, the coffee ad at the start. And now what that has done is historically, a lot of advertisers would target on using data.
So it could be demographics, it could be age, gender, very classical, traditional kind of mechanisms. What context advertising allows them to do is it allows them to be much more relevant. In terms of the type of show, and that's meant there's new types of advertisers now wanting to advertise by TV. And there's advertisers that think about how they might advertise very, very differently because the ad response is [00:37:00] significantly better when they've gone down a contextual kind of form of advertising.
Another example I'll give you, given the increased kind of regulation around data generally across Europe is There's been a real shift towards brands that have their own first party data, right? So moving away from reliance on cookies and things like IDFAs or, you know, internet sort of tags and identifiers what that's then meant is a lot of brands that have a lot of rich first party data, and they're keen to activate that first party data in a very GDPR compliant way with media data.
So what we have in another product we built here is by using InfoSum, which allows you to kind of very safely match data. We allow browsers to bring their own first pass data. And then target their own customers on ITVX, but also non customers on ITVX. And again, that's all done in a very GDPR privacy compliant manner using InfoSum, which kind of does a double blind match effectively.
And that's another great example [00:38:00] where, you know, we now have hundreds of advertisers that have now signed up to that particular product. Because it effectively allows browsers to use their own first party data, do look alike modeling, and then do a lot of targeting in very rich, long form sort of advertising.
And so that's another example that's really an innovation. And there's a huge demand in terms of the signups we've seen in terms of browsers and advertisers that are keen to then use their own data to activate their own data in premium environments.
Tom: Great Sanjeevan. Oh, it's, it's awesome to see how your team is moving the needle over there with data and AI and you certainly establish yourself.
As a leader as a data leader in the space, really appreciate your thoughts and insights and thank you so much for sharing them and joining us today
Sanjeevan: on the show. Thank you, Thomas. It's been an absolute pleasure.
Tom: That was a really fascinating conversation we had with a foremost leader in the space.
I've learned a lot about the media and entertainment business and certainly have a greater appreciation for all the subscriptions I have for Netflix and [00:39:00] Disney Plus. A lot to take away from that conversation. Let me start with you, Vijay. What were some of the key themes that came across from you from that conversation?
Vijay: Yeah you know, great, great discussion. I enjoyed it and learned a lot. You know, the few things that, that I got out of it, one, you know, this, this double, double, double business KPI that he was talking about, it's always interesting that, you know, great organizations often have these very crisp definitions of business KPIs that everybody rallies around, you know, they're catchy, they're easy to understand, and, and they're very effective in, in, in making a, you know, impact on business.
So I thought that was very nicely articulated. You know, the two other things that caught my attention. One is you know, there's always this struggle between centralization and decentralization, right? You know, centralization often slows things down but it has advantages of consistency and single source of truth and so on.
Decentralization gives you agility, but [00:40:00] it results in multiple sources of truth. And I felt like they've struck a good balance. You know, the example he quoted of a, you know, you have centralized definitions of segments but each group may have micro segments that they. That they might create for very specialized analysis and and then there's sort of these cross functional teams that work off this governed centralized data that have the freedom to fork off from that centralized system.
And, and so that that seemed like a good, good strategy. And you know, in a very large organization, complete centralization is often very, very hard to do. And and so these guys seem to have struck a good balance. The other interesting thing that he said, which Which I thought, I'd never thought about.
You know, we always hear about data literacy programs, right? You know, we, we, we need to educate people to be more data savvy, especially business folks to be data savvy and do data driven decisioning and so on. But he said the reverse is also true. You know, you need [00:41:00] programs for business literacy.
And I thought that was very interesting and you know, people never thought about it. Talk about it, and I've never given it too much thought. You know, you need to educate everybody in the organization, particularly your data folks, about some of these business metrics and the business concepts, because end of the day, there has to be a communication between the data teams and business teams, and just data literacy for business teams.
Is just as important as business literacy for data teams. And I thought that was great perspective.
Tom: Yeah, exactly. They've certainly struck a, an optimal balance for the business teams to be working with the data teams in this decentralized model. And for me, it really cemented how well they've done that when he was talking about this cost per acquired user or cost per acquired hour.
user, right? Because that's a business metric that cuts across two different teams and two different parts of the customer journey, right? You don't care just about the acquisition [00:42:00] cost, but you actually want to make sure you're acquiring a good user who's spending the time on the platform and viewing it.
And even in this decentralized world, they're bringing together the inputs of the business for both the marketing team on the acquisition side and the product team who's trying to drive. User engagement.
Vijay: Yeah, that's a great example. You know, this is what we've been talking about. This idea of a broader customer analytics umbrella, right?
You know, it's no longer the siloed product analytics or marketing analytics or digital experience. It's really bringing together all of these. A marketer should care about, you know, hours watched, you know, not just about how many users they acquired. So that sort of validates some of the things we've been talking about of this convergence of, And certainly marketing analytics and product analytics and the broader convergence into this customer analytics category.
Tom: Yeah, and how all this comes together in this last mile, I thought was a very interesting concept. I had not heard this term silver service before, but as a [00:43:00] contrast to self service. That's certainly a model that needs to be supported where you do have end user applications that are very narrow and siloed, and you do need the data teams to offer through a silver service program the ability to build what he called data product, right?
So You're oftentimes moving and creating data that exists in other systems. And so the model they based on the tooling and the stack they have, just how critical that is to be able to answer that next question. It sounds like they've been able to build an organization to support all of this ad hoc analytics that needs to be done outside of the the tooling that they've built for their, or they've acquired for their end users.
Vijay: You know, on the SilverService one, one thing I thought about was, um, so I think the SilverService model is important and is, is, is needed for large organizations with very complex businesses and in a complex data and so on. And you cannot expect [00:44:00] your end business user to self serve for everything. They do need experts that they go to the for, for harder questions.
But then the, the folks providing that silver service, they should be able to self serve. They shouldn't in turn be have to call their data teams and have to wait for two weeks, right? So, so I think some of the platforms like ours you know, we talk about self service for business users, but you also need self service for these you know, for these analysts power users who are providing the silver service to the end users.
Tom: All right, that concludes today's show. Thank you for joining us and feel free to reach out to Vijay or I on LinkedIn or Twitter with any questions. Until next season, goodbye.