We aren't giving out data continuum of the user: Amagi's Baskar Subramanian

In a Q&A, the CEO of the newly-minted unicorn dwells on the company's role in the adtech supply chain, its technology pipeline and data privacy

Amagi, a newly-minted unicorn, is a mediatech platform that connects ad networks, online streaming platforms and content producers. It currently caters to a viewership of 180 million hours a month and 30 billion annualised advertising impressions. Co-founder and CEO Baskar Subramanian spoke with Deepsekhar Choudhury about the company’s role in the adtech supply chain, its technology pipeline and data privacy. Edited excerpts:

The advertising supply chain is quite complicated. Where does Amagi fit into the equation?

Amagi is not in the business of selling ads. We are essentially the bridge between the ad exchanges and streaming platforms or content owners. For example, when an ad break comes while we are watching live sports on a streaming platform, we are the ones enabling operators to trigger an ad break as we run the live production.

Once the trigger comes in, we call the ad networks, saying X, Y, Z are watching, and what’s the right ad to place for each one of them? They decide which ad to show, to whom, and Amagi’s role is to insert those different ads for millions of users at the same time -- and then give analytical insights on them to our customers.

So, is it not ultimately your algorithm that decides what ad is shown to a viewer?

It is actually a very complex thing and there are multiple layers to it. The programmatic ad exchanges are almost like stock exchanges. Let’s say you are watching something and I get a two-minute ad for you. That information is first sent to multiple ad exchanges. Each of those ad exchanges are working with different demand side platforms and everyone is bidding for the same opportunity. Eventually, all those bids come to us and we select the one with the highest bid.

Amagi has said that its customers see better ad economics, and impressions rise by five to 10 times. What are the tech chops that enable you to deliver such outcomes?

The first thing is that the basic service of ensuring consistency of streaming delivery is itself a non-trivial technology challenge. For example, nobody likes a black screen and so you need very high availability. When it is a live event like sports, you also have to take care that there is very low latency. These demands themselves involve such a deep level of tech that very few outside the FAANG companies have an in-house capability for such things.

Next, our scale of 180 million viewership hours a month gives us an enormous treasure of data. With our machine learning models, we can analyse viewing habits to predict whether a household has a girl child or a boy child, if there is a pet in the household, does the viewer have a right wing sensibility or left wing, or even are they vegan or not. The way we see it, all these capabilities will not only help show viewers more personalised ads, but also empower content platforms to serve more personalised content.

Data privacy concerns are being raised the world over on such personalisation. Tech giants like Apple and Google are being forced to cut back on data tracking. Do you have any plans to recalibrate your tech, given this reality?

As a company, we truly believe that privacy and targeting need not be orthogonal to each other. The technology we are evolving seeks to take out the viewer identity from the equation and just rely on the viewing behaviour. We don’t want to say X is watching a particular show and solicit ads for him. The approach is to tell the advertiser that there is a household in Mumbai with a pet and the viewer is a news junkie.

We want to lead the trend of blocking any sort of cookies that give away the identity of the viewer. Although we have access to the identity in terms of IP address, we do not want that information to go out.

But studies have shown that you can take a parameter X from one anonymised data set, a parameter Y from another anonymised data set and zero in on the identity of the user.

In a cookie-less world, people are trying to build a universal identifier based on the actions of users. But, the information is so generic that it is not possible to build an identifier as, for instance, hundreds of people in Mumbai might be watching the same channel and have a pet and a child at home. There are no other signals coming out.

The situation where anonymity can be broken is when I say that an ID called 1234 is watching a film in Mumbai, and say the next time that 1234 is watching a soccer match. Now, two different ad networks may have this individual identified separately as 1234 and 2345. That’s when somebody can try to match the two and find out who the person is.

So, we are not giving a continuum of the user itself. You cannot match it as there is no ID and it is per instance. We are just giving the interest levels and the content they are watching.

Are there any moonshot projects you are working on, such as virtual reality or the metaverse?

We are working on how to deliver more personalised content to the user and that itself is a big problem to solve. Next, we are working on autonomous production. For instance, in camera-based operations like news media and sports, there are a lot of operations. We are working to see if those things can be automated just like self-driving cars.

There is also work happening on building virtual reality engines. We think that the difference between a movie and a game will go -- it is just about interactivity coming in to push the story forward in a better way. I don’t know what you would call a game or what you call a movie. It might be the case that you can literally play in the movie.