Your browser is not supported. Please update it.

22 September 2022

Tug-of-war over AI vs. human metadata evident at IBC 2022

Metadata is essential for powering all user-facing aspects of the video industry, meaning that the vendors generating the stuff are having to adapt to its commodity status. The clear answer is increased automation and scale, but this can lead certain vendors to overstate the impact that AI and ML is having on their service.

There were great expectations going into our conversation with Danish metadata vendor Simply.TV at IBC 2022. COO Daniel Ruhmann told Faultline that the company was founded in 2019 to bring automation to the metadata game and embrace its commoditization with a “fully automated, end-to-end metadata offering.”

We understand that most metadata vendors employ a mix of human and artificial intelligence, but where the differences really start to show is on the exact division of labor.

A company like Gracenote proudly states that humans are solely capable of handling any subjective, emotional, or thematic metadata. Meanwhile, companies like Simply.TV and VionLabs claim to be as close to artificially intelligent as possible.

“Every vendor has manual bits in their system, but we are only using humans to check that the AI and ML algorithms are working correctly,” Ruhmann told us.

Where this is certainly true is in the ingest and mapping of practical metadata. Here, Simply.TV can operate at scale, taking direct feeds from over 10,000 large broadcasters across the world and processing data to suit its customers’ needs. Ruhmann, a former Gracenote employee, points out that his rival has a team of nine full-time employees to do the same task.

But with a bit more digging, it transpires that Simply.TV seems to be following a more similar workflow to Gracenote than it would like to admit. “We have an editorial team of 30 that go through and select all metadata to do with genre, mood and feeling. If you leave that to ML, it will see an elevator in the latest Aquaman movie and tag it as a film about elevators,” Ruhmann admitted.

If anything, Gracenote might be ahead of Simply.TV in some fields of automation. “You can automate mood, but AI and ML struggles with places and themes,” said Gracenote’s SVP of Product Management, Trent Wheeler, also speaking with Faultline at IBC 2022.

Our chat at the Gracenote stand was a more placid affair. The giant of the industry has less reason to trash talk the competition than the hungry start-ups circling its walls, and instead has time to curate neat features that make the application of metadata to the end-user experience far more accessible.

For instance, the new Gracenote ID Distribution System (GIDS) allows Gracenote customers to make a private database to repackage the metadata that wraps around their content library. With the advent of FAST, content owners want dynamic content tags to constantly repackage their titles into new pop-up channels.

Other cuddlier offerings include the Advanced Discovery suite, which allows thumbnail personalization according to mood and wider social themes – celebrating women actors, for example. Gracenote found that dynamically altering the thumbnail for a top five US streamer created 11% more engagement with titles and increased watch times by 6.7%.

As for AI and ML, Wheeler revealed that Gracenote uses the technology for mapping and deduplication purposes. “We also use AI to make sure we are tagging with finesse. If one tag is used too much, it needs to be broken up into more specific descriptors,” he explained.

Nonetheless, Simply.TV’s self-identification as a lean, mean, automated machine, as well as its comparatively less cuddly UX compared to the likes of Gracenote, would lead us to believe that Simply.TV is massively undercutting the competition.

However, Ruhmann says its prices are in line with its competitors, as customers save big further down their workflow. “We tailor our data to fit seamlessly into the customer’s workflow so that they have near zero implementation costs,” he shared. Rather than a cuddly, Gracenote-esque online dashboard, Simply.TV’s customers get a raw data feed to ingest directly into their own systems.

The pitch seems to be working as Simply.TV is fast moving up the value chain of operators, since beginning with tier 3s in 2019. “We are now looking at the likes of Deustche Telekom, Liberty Global. Giving them modern metadata to run an interface like Netflix, but more affordable,” Ruhmann said.

Aside from operators like Telia and DNA, the company also sells to OTT platforms, smart TV OEMs like Phillips and Vestel and royalty collection societies. Notably, advertisers are not yet seen as a worthy pursuit.

The company has just acquired EPG specialist Gvidi to become the company’s sole supplier of metadata. “Gvidi was already profitable. Now it will just have richer metadata,” Ruhmann explained.

Metadata is also on the mind of recommendations market leader ThinkAnalytics, which is consistently looking to expand and diversify its personalization empire. “We’re now looking beyond video into ultra-aggregation,” said the company’s CEO of Europe and APAC, Samuel Sweet, uttering an unwelcome buzzword evolution of super aggregation. Sweet shared that the company has recently undertaken a project with an unnamed North American media powerhouse to bring media, news and gaming into one experience.

This is the kind of service that only applies to service providers that have far reaching relationships with content owners across multiple verticals – direct-to-consumer giants like Apple and Amazon, and tier 1 operators.

While recommending someone an album based on their film choice sounds rather abstract, Sweet feels that ThinkAnalytics’ scale and in-house metadata capabilities make it best suited for spinning multiple plates and joining the dots. “People talk about breadth and depth, but metadata vendors are too often fragmented and local in their focus – it is often the Achilles heel of the industry. That is why we enrich stuff,” he told us.