ITV’s Head of Data Capability Hardeep Dhaliwal explains how media companies can ensure their AI solutions get the data required to produce optimum results.
Data is the lifeblood of AI. As such, broadcast and media entertainment companies need to think very carefully about how they obtain and utilise data to extract the best results from AI for critical tasks such as streaming site search, discoverability, marketing, and commercial advertising.
Hardeep Dhaliwal, Head of Data Capability for ITV, will dissect this very topic as part of the ‘AI in Action’ case study series, taking place during the IBC Conference on Sunday 15th September.
“My presentation will be centred around why solid data foundations are critical to the success of AI models, ensuring they produce trustworthy, unbiased and accurate results,” says Dhaliwal. “In particular, I’ll talk about ITV’s data products approach, which follows a data mesh methodology [which introduces the notion of a self-serve data platform conceived to avoid duplication] and I’ll share some insights into why this works effectively, as well as some of the challenges and learnings from my experiences.”
He selects an example of film tagging to illustrate the importance of high-quality content data: “Using machine learning, you might want to tag streaming content with moods. For example, with Toy Story you could say it’s a ‘funny family comedy with a heartwarming tone’ and it’s similar to Despicable Me, another animated movie. You could then take it further with generative AI and automatically generate [extensive] content descriptions and more tags. At the outcome of that, all of those tags and the richer description could be used to enhance recommendations – in our case, on the ITVX platform – as well as improve search and discoverability of the content on the platform, and possibly even form a specific editorial rail on the ITVX homepage.”
PERILS OF A POOR EXPERIENCE
“The quality of what you put in impacts the quality of what you get out” Hardeep Dhaliwal, Head of Data Capability for ITV
Contrast all of that with the potential difficulties caused by inaccurate or incomplete tagging. “It’s an extreme example, but can you imagine if the same family comedy was incorrectly tagged as a horror film? That would lead to a poor experience, loss of trust, and lower levels of user engagement as a result,” says Dhaliwal.
Watch now AI - Episode 2: Importance of personalisation in media
To avoid this danger, he perceives the entire content cycle as one big data supply chain: “The quality of what you put in impacts the quality of what you get out. And I would say there are essentially four steps in the chain. The first step involves the raw content data, and as you can imagine we have our own in-house content as well as content purchased from other providers and suppliers. Inevitably, you get varying bits of data with different levels of quality.
“The second step is to consider how you want to standardise that data; in some cases you might want to enrich it by involving other data sets. The third step involves feeding that data into the [MI and AI models] correctly. Then the fourth step is hopefully achieving the outcome you’re hoping for, which is to improve the viewer experience.”
BUILDING THE PROPOSITION
It’s a complex task to which Dhaliwal brings a wealth of data consulting experience in different industries, including finance (Barclays), mobile telcos (O2) and food delivery (Just Eat). But it was likely his involvement in building a new data platform for Sony PlayStation in 2008 that gave him “the bug to work in media entertainment”. He became head of data strategy marketing at ITV in April 2022, then Head of Data Capability in June 2023.
“It was a really interesting time to join, ahead of the launch of ITVX in December 2022,” recalls Dhaliwal, alluding to what was at the time the UK’s first integrated advertising and subscription (AVOD/SVOD) platform. “We had this new proposition and then the question became how are we going to grow it? And when you talk about growing a [content proposition], you are thinking about how to drive the acquisition of new viewers, how to retain existing viewers, and how to encourage people to view more.”
Dhaliwal oversaw a team that built a suite of more optimised ML propensity models: “One of those models would look at viewers, their behaviour and characteristics, and then see which of those viewers had a propensity to view more content as well as those which might be interested in upselling to ITVX Premium,” which of course is the subscription tier of ITVX that allows viewers to watch most shows ad-free and download them to their mobile devices.
The latest viewing figures indicate that all the hard work is paying off. June 2024 saw ITVX enjoy its best month on record, with 375.8 million streams across the month. In addition, June 30 was ITV’s biggest-ever day in streaming, with a powerful combination of two Euro 2024 matches (England’s win against Slovakia and Spain’s against Georgia) and Love Island contributing to 30.3 million streams across the day.
Despite the success, Dhaliwal is keen to emphasise that the work of optimising ML/AI models is always ongoing. Every time you have a use case for a model, he explains, “you go and design it, build and implement the model, and then test it. The measurement piece is really important; you don’t just put the model out there [and leave it]. Instead, there is always an element of tweaking the models to further improve the performance and ensure you are driving the objectives you want.”
STRIKING A BALANCE
“Everybody’s got a lot to learn, which possibly makes an event like IBC even more valuable today” Hardeep Dhaliwal, ITV
Whilst acknowledging his “fascination” with the future potential of generative AI, Dhaliwal is also mindful of the need for caution and, increasingly, a firm regulatory framework. “My day job is focused on the data, so I’m not so close to the legal side of it, but broadly, I think that a clear strategy from regulators in consultation with the industry to assess the risks and then design and implement safeguarding measures seems increasingly important. In the media and entertainment space alone, there are so many concerns – ranging from deep fakes to copyright and the data that the large language models (LLMs) are training on – and all of that needs to be addressed.”
Read more Training models responsibly: AI and creator rights
He sounds optimistic that, with careful thought and “effective collaboration”, a judicious balance can be achieved. “Whether you’re an industry [organisation], a regulator or even an academic, collectively we need to strike the right balance between AI for good and managing against the risks as well. And it’s even more essential in these relatively early days to make sure that we all set off on the right path.”
Meanwhile, Dhaliwal is enthusiastic about attending what will be his first-ever IBC Show, accompanied by colleagues including ITV Chief Architect Clive Santamaria, who has a long history of supporting IBC Accelerator projects including this year’s Generative AI in Action project working alongside RAI and EBU. “The more I find out about IBC and the collaborative work across the media and entertainment and broadcasting landscape, the more involved I want to get involved – especially at this exciting time, which is set against a backdrop of innovation and change. So I am looking forward to being there for the whole event, sharing some of my knowledge and experiences, but also finding out more about the industry’s use of new technologies and the thoughts of some of my peers as well.
“It’s a really fascinating time and everybody’s got a lot to learn, which possibly makes an event like IBC even more valuable today than it has been in the past.”
Discover more from the AI in Action series and Hardeep Dhaliwal’s presentation: ‘Building a solid foundation to enable AI’, by attending the IBC Conference on Sunday 15 September, beginning at 11.15.
For the full IBC Conference programme, click here.
Read more IBC2024 unveils conference keynotes and sessions tackling critical media challenges
No comments yet