Too much content, too hard to find

No one will disagree this year has been surreal. Due to the Covid lockdowns in most parts of the world, consumers have spent more time this year consuming content across an almost “unlimited” catalogue of content across SVOD and pay-TV.

With the proliferation of new SVOD players – beyond the NADs (Netflix, Amazon Prime, Disney+) – and a constantly increasing catalogue of content, rich metadata has become pivotal to content discovery and therefore keeping consumers glued to the TV platforms.

In addition, recommendation engines have also become a huge part of content discovery to help personalise the consumer experience. Despite the powerful search and recommendation algorithms used by some of the top SVOD or OTT players in the market, it still takes between 9-12 minutes for consumers to discover something they like (see figure 1) given the amount of choice they are faced with.

One of the reasons is that recommendation engines can be sometimes too simplistic relying purely on broad metadata categories such as genre, cast and previously watched content. Even with this relatively basic methodology, every platform has nuances in how they define genre (e.g. does everyone have a common understanding of the difference between ‘RomCom’ and Comedy as genre categories?).

This rapidly evolving landscape is pushing content owners and distributors to understand and properly define their metadata in order to apply it at a much deeper and granular level. With the use of enhanced metadata, these terms are automatically tagged and along with other pertinent descriptors, service providers can quickly and efficiently create connections to serve subscribers better. Most platforms are still struggling to achieve this which is evidenced by viewers’ perception that personalised recommendations are not accurate in c. 80% of the cases (see figure 2).

This is especially an issue for owners with a large amount of content and a long-tail catalogue, which may not be updated and could be missing key metadata such as synopsis and images. Without the support of smart search capabilities to address the long-tail catalogues, content owners will ultimately experience a reduction in customer engagement over the long term.

To stand out, content metadata needs to be vividly descriptive. This goes much deeper than the structured metadata that is currently widely used such as programme hierarchy (programme, series, episode), synopsis, cast, crew, length, release year, genre, ratings, etc. To engage viewers, content providers need to entice them with rich imagery and an understanding of the thematic elements in the video, which are relevant to improve content discovery for their subscribers. 

Having said this, this is a challenging task to achieve without AI/machine learning playing an important role. AI can be used today to automatically create descriptive metadata by reviewing content catalogues at scale by including and making searchable unstructured metadata at a scene by scene level – for example length, dialogue (including markers for profanity), tags for unpleasant content (violence, horror, sex) and colour tone.

By moving towards building super rich but unstructured metadata within their respective services, TV operators and streamers can ensure that an informed selection of content is presented to consumers resulting in happier users who dwell longer within a platform.