iVS taps into AI-powered hyper-personalised video content serving

iVS (formerly known as iVideoSmart) leveraged BasisAI’s bespoke machine learning solution to serve more relevant and contextual video content to unique viewers. The solution augmented iVS’s proprietary video delivery and monetisation platform with AI-powered hyper-personalisation, in order to drive exponentially greater video engagement and maximise revenue.

  • From conception to execution 8 weeks only

  • Up to 32% uplift in video engagement

  • A global first AI-augmented video-in-a-box solution for media publishers

iVS is one of the largest independent premium video delivery platforms in South East Asia, serving over 120 million unique users across more than 300 million videos on over 1.2 billion pages each month. The company provides a brand safe video network that provides superior-quality video delivery for major media publishers, and therefore inventory that can be monetised through in-stream advertising. Through its innovative Natural Language Processing (NLP) and Artificial intelligence (AI) driven recommendation engine, iVS has a proven track record of converting existing page view traffic into new video views. 

High-quality, immersive and engaging videos drive user engagement. To push the boundaries on this, iVS wanted to offer an even more relevant video content selection process to their publisher partners, with videos that are better contextualised to user preferences in a variety of language and cultural environments. Drawing from decades of experience in solving data science problems, BasisAI helped to develop a machine learning strategy which defined the key drivers of impact and offered a game-changing AI solution for iVS. 

“Always showing the right video, at the right time, in the right place is no easy task. As our NLP, key-word-tagging-based recommendations had hit a plateau, we were looking to find an automated way to further improve the video selection process and maximise page view to video view conversions. This is why we turned to BasisAI for a machine learning model to support this.”   

- Milan Reinartz, CEO, iVS

Bespoke hyper-personalisation drives a competitive edge

A personalised experience communicates customer understanding and increases brand loyalty, and consumers have come to expect tailored product and service recommendations. AI is now raising the game for personalisation marketing entirely. 'Hyper-personalisation' is achieved by leveraging machine learning and real-time data to deliver instantaneous and highly relevant content or product recommendations to users. For example, as an individual reads an article online, they are simultaneously recommended similar products based on the customer’s profile and search history. The result is better click-through rates, greater engagement and a direct result on a business's top line. In an increasingly competitive market, companies are starting to unlock the full potential of hyper-personalisation in order to gain a competitive edge.

In order to develop the hyper-personalised machine learning models, the BasisAI team worked closely with iVS’s product and engineering team to integrate massive data streams of video content, analytics and metadata. This was implemented through a dual-prong approach of pulling applicable user data, and synthesising the relevant attributes to match with iVS’ video library. Extensive use of natural language processing was used to process this data which was in different South-East Asian languages, such as Bahasa Indonesia and Filipino (Tagalog). 

By building and training personalised algorithms on their own data, iVS was able to sustain its competitive advantage and achieve maximum lift, which generic models would not be able to achieve. The reason for this is that generic solutions are pre-trained on alternative data-sets as opposed to an organisation’s own data, and therefore do not leverage historic individual user data to drive future recommendations. For example, even if a reader who previously has been fan of sports articles is reading an article about food,  sports videos can also be recommended, and still be highly relevant. Furthermore, regional behaviours for different media publishers in Indonesia and the Philippines were detected by the algorithms and enabled the trending videos specific to each region to be incorporated as drivers in the predictions.

Machine learning deployed through Bedrock, on AWS

Within 8 weeks, a real-time personalisation engine for serving video content based on users’ content preferences was developed and then deployed using Bedrock, BasisAI’s proprietary end-to-end machine learning platform. Bedrock provides a fully managed and automated infrastructure and acts as the foundational stack to a customer’s cloud infrastructure. With this solution, iVS is able to retain control of their data and code as Bedrock deployments do not require data to ever leave a customer’s cloud environment. This means that iVS is able to continue to benefit from the world-class security provisions that Amazon Web Services (AWS) provides. Maintaining the hyper-personalisation models is also automated through the Bedrock stack, and operated securely on AWS.

“The BasisAI personalisation and machine learning algorithm implemented within our existing infrastructure enabled significant growth in inventory generation for iVS. In some cases the video view conversion increased by up to a third. It also helped us learn about differentiating factors between publisher partners that will influence some qualitative decision-making.”

- Milan Reinartz, CEO, iVS

With BasisAI’s bespoke machine learning offering, iVS now provides a video-in-a-box solution to enable better video recommendation, delivery and monetisation in 5 pan-Asian markets including Indonesia, Philippines, Malaysia, Singapore and Taiwan.

By leveraging NLP and AI to profile both the text and video content, iVS is able to recommend and deliver highly relevant recommendations to users at scale. This has set iVS apart from its competitors in its ability to drive relevance for end users in different cultural contexts, unlocking additional quantifiable audience reach for the media industry. The result generated significantly increased monetisation opportunities through improved user engagement. Handling thousands of queries a second, the solution was able to generate in some cases between 20% to 30% greater engagement, which has the potential to result in double-digit net revenue growth for the company.