Thought Leaders in Big Data: PubMatic CTO Anand Das and VP Engineering Sri Gopalsamy (Part 5)

Sramana Mitra: Go to the 30,000-foot level in your industry and point to some open problems that you see ought to be solved that you see are not being solved in consideration of new startups.

Anand Das: Let me just give you a brief overview of the kind of problems that are in the industry. When you talk about ad industry, there are multiple players out there from the advertisers and training desks to data providers. There are a lot of players in between the advertiser and the publisher. The advertiser cannot work with everyone individually. Neither can a publisher work with every AdWord individually. That creates a lot of interesting problems. Everybody has a secret sauce. Nobody wants to share their piece of information. Getting data to actually identify what is happening at what stage is very difficult in this industry because everybody is being protective about their data.

Let’s take the challenge from a different perspective, say an advertiser. An advertiser wants to spend a certain amount of dollars to get a certain ROI. They get the ROI and they’re happy with it. In the next year, they’ll say, “Let’s increase the ROI by X%.” From an advertiser’s perspective, >>>

Thought Leaders in Big Data: PubMatic CTO Anand Das and VP Engineering Sri Gopalsamy (Part 4)

Sramana Mitra: I want to ask you a specific question about the crossover between media and e-commerce. Let’s take the example of Mail Online for instance. This is a very big lifestyle publisher. Are you familiar with them?

Sri Gopalsamy: Yes.

Sramana Mitra: This is a big lifestyle site. They have hundreds of millions of users. I would say that the biggest category where they would be able to monetize is in lifestyle. I would say Vogue would fall into this category. In general, lifestyle media has a lot of impressions out there. There’s a huge inventory of impressions that is lifestyle content. Do you do anything to tie lifestyle content directly into commerce?

Anand Das: It depends upon the publishers and their business policies. We do help them do that. We don’t do it on their behalf, but we do integrate into their systems that they use to actually give them insights that will allow them to bucket users into interesting categories. The other thing is whenever you deal with e-commerce, the challenge for any company which does sales and ads is >>>

Thought Leaders in Big Data: PubMatic CTO Anand Das and VP Engineering Sri Gopalsamy (Part 3)

Sri Gopalsamy: We have done a really good job over the years to build this platform where we respond to bids in real time within few milliseconds. We provide insights to our publishers within a minute. For all the impressions that are going through our platform, how much of them actually get converted? What’s their average pay per thousand clicks? We expose all of the data in real time so that they can better manage their inventory, deals, and relationships with advertisers.

Sramana Mitra: When you’re calculating in real time what ads to show, across how many parameters is this optimization happening?

Sri Gopalsamy: Right now, it’s well over 350 parameters in real time. >>>

Thought Leaders in Big Data: PubMatic CTO Anand Das and VP Engineering Sri Gopalsamy (Part 2)

Sramana Mitra: Let’s double-click down on the PubMatic context. Let’s talk about what you’re trying to deliver for your customers and what does Big Data bring to that process. Let’s get pretty granular and technical. Our audience is very sophisticated so you can get as interesting as possible in a nerdy way.

Sri Gopalsamy: I’ll take a shot at it. As Andy said, PubMatic is a marketing automation software company. In fact, PubMatic is the only independent company that created these automation platforms specifically for publishers. We enable publishers to automate their sales environment. Essentially, that’s what we’re trying to do. Our solution normally helps unify their platform to monetize their sites. Now, there’s a trend going on in the digital media industry where there is more advertising on multiple screens. Many of our publishers’ advertising revolves around desktop-based to mobile and tablet. Our platform helps identify and fully leverage the revenue of our publishers. >>>

Thought Leaders in Big Data: PubMatic CTO Anand Das and VP Engineering Sri Gopalsamy (Part 1)

Publishers are using Big Data and machine learning to optimize what ads to put in front of their audience. This discussion delves into the depths of that process, and also explores open problems in that world.

Sramana Mitra: Let’s start by introducing you to the audience. Tell us about yourself as well as about PubMatic.

Sri Gopalsamy: I joined recently as the VP Engineering heading the Big Data and machine learning area. I joined the company about two months ago. Andy is one of our co-founders. He has been with the company since the beginning so he has a lot more context around company introduction. My role here is to drive the entire Big Data and real-time data analytics at PubMatic.

I manage a team of Big Data analytics engineers who are building the platform and innovating in the core area of data analytics, natural language processing, real-time analytics, and machine learning. At PubMatic, we view Big Data processing as the key differentiator in helping us grow. I have over 20 years of industry experience. I spend many years in the leadership level with Walmart Global E-commerce. I was heading all of >>>

Thought Leaders in Big Data: Dave Copps, CEO of Brainspace (Part 5)

Sramana Mitra: Give me a good use case of deep learning that you would apply that technology to and which is an interesting problem to solve. In other words, share with me a pain point that deep learning would solve elegantly.

Dave Copps: I think that the use case I find most fascinating is visual. We’re getting into machine vision where computers can actually see something and recognize the object. We’re getting to the point where you can pull out your phone, just face it towards the world, and have it come back and tell you exactly where you are and things that are around you. Once computers can go beyond words where they can start to understand images and start to relate those images to knowledge, it gets really interesting. Our phones can become instantly connected with anything just by pointing our phone at something. It’s not something we’ll ever do. Deep learning is certainly a space I would encourage people to go into. >>>

Thought Leaders in Big Data: Dave Copps, CEO of Brainspace (Part 4)

Dave Copps: Customers are primarily using Discovery Five for e-discovery and litigation support. When companies are taken to court and there are tens of millions of case documents, they use Brainspace Discovery Five to build a brain on those case documents. That product is interesting. I’d like to call it augmented intelligence. We built it in a way that Brainspace can do a lot of the heavy lifting that people can’t do. A person can’t possibly read 10 million documents and remember how every word is related to every other word. Brainspace can do that.

We give people the ability to use Brainspace, interact with Brainspace and see the concepts and ideas that Brainspace is relating to your query. The other product is Brainspace Enterprise. That product is launching in three weeks. We have two very large Fortune 10 beta customers. I can’t tell you who they are. The idea behind Brainspace Enterprise is we want to turn the social curation of documents inside of a company into a brain. As people are curating, collecting, and sharing documents, Brainspace is learning from that curation, and forming those documents into an asset that literally every person in the company can now engage to connect with other people and documents. >>>

Thought Leaders in Big Data: Dave Copps, CEO of Brainspace (Part 3)

Dave Copps: I heard a term the other day called feral knowledge. How do we start to tame the wild? How can we find answers in that large environment? It’s a lot different from what you’re talking about for sure. If you have a simple environment, that’s a very good environment to use a keyword search. The minute you go beyond that, it doesn’t work anymore. We’ve taken a very different approach. Brainspace formed a collective intelligence from an enterprise’s documents.

We’ll take the tens of millions of documents that a company has. We read them and connect the concepts inside those documents into a multi-dimensional space. Before anyone ever pushes a search button, we’ve connected the concepts inside that data. Now, when you enter a search, it’s not just looking for that word. It’s actually connecting that word to all the other words and phrases that could be related to it, and then connecting you with documents that have a similar concept. It’s a much different scenario. I agree with you that when there’s simple applications that have a known set of answers, that’s a great place for keyword search. We don’t play too much in those areas. >>>