StarzPlay: A personalised recommendation

StarzPlay has made a mark in the Middle East and North Africa with a market share of 29%, slightly ahead of global giant Netflix and Shahid Plus, both at around 24%, according to a 2019 IHS Markit Report. With more than 10,000 hours of content refreshed regularly, partnerships with most regional telcos and flexible payment models, the streaming service has marched ahead of its competitors by a significant margin of subscribers.

The streaming service, however, has not just developed commercial strategies to stay ahead of the game. A technical team distributed between its offices in Dubai, Lahore and Madrid is responsible for the continuously evolving StarzPlay Connect platform, which is the company’s 350-plus micro services embedded in a technology stack developed over the last five years. The most recent feather in StarzPlay’s cap is enhancements to its recommendation engine, which promises a greater personalised service with improved discoverability and user engagement, hopefully resulting in less churn. The StarzPlay Connect platform has evolved to such a level that it has now become attractive to other OTT service providers like Lionsgate, which has licensed the platform in India in order to launch its own streaming service.

“What we have is a continuously evolving in-house technology platform that is being developed to keep pace with new technologies and user engagement,” explains Faraz Arshad, VP of Technology & Infrastructure at StarzPlay. “This newly improved recommendation engine offers greater agility and scalability, with myriad algorithms embedded to help us personalise the dynamic carousels for each user, i.e. the personalised user experience and the way the content is being recommended. Essentially, every user is being recommended something based on their viewing habits, preferences and search keywords.”

A number of factors lead to a better recommendation. Besides the normal methods, scoring content and referencing a recommendation based on third-party data, even if it is not present in the database, is now becoming key.

Cagdas Direk, StarzPlay Director of Software Engineering and the man in charge of the platform’s user interface, elaborates: “There are two phases of recommendations. One is based on user consumption, and the second is content to content. This is based on third-hand metadata that is available on open source platforms, and we use some of those services to match the content to content. For instance, if you have watched Spiderman, it’s likely you will want to watch Spiderman 2 and perhaps other superhero movies as well. Irrespective of whether that content is on our service or not, the engine then starts to deduce what kind of content you are likely to watch.

“That’s the kind of algorithms we are working on to create scores based on third-party data, internal matches and user consumption. Scoring is not just based on what content you consume but what you search for as well, even if it is not available on our platform. That search is scored and enables us to make sharper recommendations. Why? Because we know your likes through the different content choices you make, not just one. Therefore, third-party data is becoming increasingly important to enrich and address the customer’s demand.”

Arshad says discoverability “is another key factor that drives more interactivity … based on what you watch, the platform keeps making recommendations, ensuring that our customers are consistently consuming the service, which leads to more viewership and reducing churn. That is the whole essence of an OTT service.”

Direk adds that many of these heavy-duty solutions are achieved through artificial intelligence (AI) and machine learning, alongside the use of graph databases, which gives the platform the flexibility in real time to score and make recommendations. “We are training machines to get a better score calculation. This is where we think our technology is sharper and more accurate than other technologies.”

Providing a seamless service, however, requires strategic planning on multiple layers within an ecosystem. It is critical to ensure that the customer can watch content without any interruptions caused by network latencies. To make this possible, StarzPlay has worked on multiple layers to address the challenges of regional infrastructure and last mile issues, by offering offline downloads and improved video compression codecs such as HEVC.

“You can screen very good quality in almost 60% less bitrate than you’re getting today. So a lot of work has already been put into place,” explains Arshad.

The team is also experimenting with new codecs such as AV-1 and CMAF, though their adoption would be dependent on the global device adoption roadmap as well.

On an infrastructure level, StarzPlay has partnered with several regional telcos and ISPs to reach customers directly through an internet exchange (IX) served from StarzPlay Connect’s points-of-presence (POPs).

“Since we are mostly connected through internet exchanges, it gives us the flexibility to serve our customers from the closest edge location. This gives an opportunity to local ISPs to deliver the best quality video streaming experience to their OnNet customers while saving their international IP transit carrier costs, and helps us to optimise and maintain the overall user experience,” explains Arshad.

Furthermore, StarzPlay Connect’s advanced technology stack allows the streaming service to add or alter functionality and capability without relying on any third-party development timelines, which means the team is not constrained by external factors.

“The in-house implementations are fully equipped to adapt and scale while considering the time-to-market factor,” says Arshad.

Direk explains further: “Whenever we come up with new ideas and different ways of achieving them, we embed them into our existing technology stack. We have not built them on proprietary algorithms to ensure greater integration with global technologies, but the platform itself is StarzPlay’s intellectual property (IP). When we want to expand our services across any region, we just have to deploy this same footprint into different regions.”

He goes on to add that the way the team has built the system enables greater compatibility with other platforms and allows communication between different recommendations, thereby meeting the bigger objective of personalising the user’s journey as much as possible on the platform.

“All the services we develop are based on API-driven technologies. They are not super customised but have been designed to be modular. As a young company, we try to be lean and agile, and modularity helps us to maintain that. So whenever there is a new technology, which is quite often the case in a young sector like OTT where everything changes so fast, we have the ability to change aspects to the stack quickly and adopt new ones.”
Alongside modularity and agility, the executives also hail the importance of data integrity.

“Well-defined customer events such as browsing and watching are recorded by the platform, so we have to ensure that this data is accurately reflected in our recommendations. The appropriateness and accuracy of those recommendations will depend on the integrity of the data we collect. That’s why we have 350-plus micro services to ensure that the level of integrity is maintained when a user starts browsing.”

Relying on open-source technologies also enables the recommendation engine to recommend content to users based on the activity of their friends. The team typically uses a graph database, which allows them to start recommending from the attributes of friends to friends. Asked if it is the same technology that Netflix uses, Direk says StarzPlay’s objective and strategy are both considerably different.

“For us, it is important that customers are consuming all the content that they would like to see, irrespective of the region they go to,” says Arshad, clarifying that StarzPlay viewers can access the content on its platform in all regions where the service is available, unlike other OTT service providers, where access to content changes according to region.

The biggest lessons at StarzPlay have come from the better understanding of subscribers’ varying tastes across the many countries and regions it covers, says Arshad.

“Since no two viewers are the same, it is imperative that our platform serves content closer to their own preference, especially when new content is constantly being added. This is where our in-house developed recommendation engine comes into play, with better analytics collected from each user, allowing us to serve custom content covering multiple genres, languages, ratings and any other preferences, allowing the user to discover more content suited specifically for him or her,” he says.

Perhaps due to being a young company, StarzPlay has also been quick to address piracy.

“With regard to content piracy and security, our DRM detects all the devices that are modified for a particular box. If the devices are not certified by firmware/operating system vendors, our content will not play on those devices. Additionally, we work closely with the device providers and keep updating our DRM policies with the latest, to be sure there is no possible piracy risk,” says Arshad.

At the moment, the SVOD service is available on ten different devices and the streaming service’s product teams are looking to expand its footprint. “We are in talks with major TV brands or set-top boxes to achieve this,” confides Arshad.

In the meantime, BroadcastPro ME learned at the time of going to press that StarzPlay has partnered with Image Nation Abu Dhabi to create original content – a first for the regional platform, but also a sign that the streaming service will do everything it takes to keep its place right at the front.