A Chinese AI firm which has had over $25 million of investment has come up with a radical TV system that uses image recognition to drive product placement ad revenues, threatening to revolutionize advertising on the TV.
The company is called Yi+ and at the recent “Global Family Internet Conference in Shanghai its founder, Ms Zhang Mo, showed off its AI assistant and what it calls an intelligent advertising platform. This is not a platform which would gain traction in its current form anywhere else in the world as it opens up the way to censorship checking as well as to advanced advertising.
But if you simply switched off the censorship options it may well find an audience in a number of APAC countries for OTT video.
The Yi+ AI system claims to have already served hundreds of millions of impressions with what it calls “scenario-based” marketing.
The Yi+ AI system analyzes the content in a broadcast TV stream in real time, and tries to identify anything in the content that can be made interactive. For instance advertisers can push viewers directly to where they can buy any product identified, in fact we get the impression the system can offer to sell you anything seen on any TV program.
If you roll back to 2005 when Japan introduced ISDB-T 1 Seg mobile TV, a Broadcast Markup Language was created by the consortium leader NHK, the Japanese public broadcaster, and it envisaged links from the phone screen that would lead to e-commerce pages and direct purchases. This looks like the same idea (and may suffer the same fate, which was to be largely ignored) but in this case the opportunities are handled automatically by AI. The Japanese effort required individual workers to mark up each TV screen.
So after an image or video has been uploaded onto the platform, the system will analyze and identify the celebrities, the objects and scenarios in the scenes and look within its inventory for an appropriate advert, which it will somehow insert into the screen (we think it is in the next available ad-break but it might be in real time). These advertising labels are being generated all the time from ingest (can also be VoD ingest) and we understand that advertisers can then bid for the ad labels to conduct targeted marketing.
Yi+ says it can drive this from investor Alibaba’s cloud which is where its AI system sits, which has been trained to target multiple 100s of product categories.
Yi+ says it is already working with prominent ad agencies and retailers and this is driving even more data at its AI to tune the process further. It boasts that a video can show less adverts, but drive higher direct sales. We assume that each ad slot identified by the image recognition system could be auctioned in a conventional real time bidding system.
The system has previously come out with a Content Recommendation service based on the same image recognition capabilities and prior to this we had not heard of a native recommendation system from China.
Yi+ gives the example that when the user sees his favorite celebrity on TV, he could press any key on the remote control to get more information on that celebrity and use a voice search service, to show only the scenes with that celebrity in.
Another example was the for travel programs, a side-screen display will indicate relevant information about air tickets’ prices, available hotels, entertainment choices and food recommendation using creative AR and gifs.
Of course any system where all of these choices have to have human involvement in the first place to prepare this data, is less likely to have a major impact, but we can see that the system is only really partially developed and now has to be taken to market by any advertiser which is enthusiastic enough to do the work and take the risk.
The censorship system could also be used for parental control, highlighting AI developed tags such as contains “sex” or “violence” before the video plays and these labels could be banned for particular viewers.
But the system definitely crosses some lines, using a camera in the smart TV to identify children among viewers, so they cannot watch certain scenes. This sounds alright in principle until you ask the question what happens to the pictures of the children which the system takes, and work out that such a system could be used to advertise sweets to children, who have less defenses against advertising, a continuing problem in western TV systems.
The Yi+ visual recognition engine can detect thirty thousand objects and scenes in videos and identify them in real-time including types of scenes and even particular faces.
Yi+ has become a strategic partner with CITIC Guoan Media Group in a deal that offers real-time visual recognition technology in live-streaming videos for on-demand cable TV.
We suspect that US and European AI systems have to navigate a far more regulated environment, but even this system could influence some local Asia Pac markets and may give US and European companies some ideas on novel ad approaches.