Steven Marder is CEO and a co-Founder of Eurekster, Inc. With significant experience in the traditional media, digital media and technology industries and as a serial entrepreneur, Steven has almost two decades of professional experience as a business development/ corporate development executive, venture catalyst, strategic advisor and attorney.
Steven co-founded and is on the Board of enterprise search company S.L.I Systems, co-founded and managed the North American operations of multinational Gramercy Venture Advisors, co-founded indie recording company Metropolis Records and was Chairman/CEO/ co-Founder of public company eMemberDirect, Inc./PetPlanet.com. In addition, Steven served in executive roles at global venture catalyst Double Impact, Compton’s NewMedia/ Tribune Company and EMI Music. He is a graduate of Columbia College/ Columbia University and St. John’s School of Law.
In this interview, Steven and I talked at length about Eurekster’s social search portal product (Swickis) and about the concept of social search in general.
Eric Enge: Can you start by telling the readers what a Swicki is, and how it might help their website?
Steven Marder: Sure. A Swicki is a custom search portal that is basically on the topic of the publisher’s choice. Any publisher or consumer can build a specialized search engine around any topic, and that can be around a topic that they have a website on, or a webpage, or a blog, or a profile page. It could be around any interest whether it’s a vertical interest like sports, or gardening, or journalism. It could be around local community; or it could be Atlanta sports, or it could be around a group of people or social, based on social connection.
Eric Enge: Right. So, the process as I understand it is that the person who wants to create this search engine has to start by knowing what sites they want to have in it, right? It’s not the whole web now; you are actually constraining what the search engine covers.
Steven Marder: Yes. So let’s take a step back for a second. So, that person has a topic of interest, right? It’s interesting to them, and it’s a topic of interest for the community that they serve. Again, whether it’s five people that read their blog or five million people that go to their site. You have to have a topic, and it can be broad, or it can be really specific. Then, to build a Swicki you need to understand if you want to present web content, such as text content, images, or video. It depends on what type of content is most appropriate for that topic of interest, and for that very specific community.
For example, a site like MTV.com is most likely going to be more focused on presenting video type content, and other types of sites are going to be more focused on text. Some of them are just going to want to leverage universal information, such as text, and image, and video. Then, as the users use the search engine, it basically learns from that, and learns what the users find to be most relevant for them.
The next layer down, as you had mentioned, is to then understand within that topic what the starting points are. If you want all types of content such as web, blog, images, and video, you will want to filter some of those sources, and you will want to pick ten sites or fifteen sites that I really like around sports, or around gardening, or surfing. I want to pick those sites so that the system can prioritize that content, because for me as a publisher, or as an editor, I am deeming that content to be more relevant for my community.
Then, at the same time you can also indicate sites that are really not appropriate for your community, and editorially keep them out. They may just be good sites that are not appropriate for your community, or competitive to your business strategy. Initially, what you are doing is you are picking the sources of content and you are filtering that content, and you are creating what’s somewhat known as a custom search engine.
As part of that you create the initial buzz cloud, or the Widget. You as the publisher, just like when you are creating your page, are aggregating content, and designing your page, and you are thinking okay, what’s interesting and relevant for my community of users? You are picking some keywords, or video content for that buzz cloud.
Of course on top of the buzz cloud is a search box, and you are designing it. You are picking colors that you like, and fonts, and all that good stuff; so we give you that control. You build the custom search engine, and then you build that customized Widget buzz cloud.
Then, you say okay, I want to put the Widget on my site, my blog, my webpage, my profile page. And then, you put it out there, and then the really interesting stuff begins, because, as users use it to search and discover content, it learns from that user behavior.
Eric Enge: Yes. How does that work, how does it do its learning?
Steven Marder: There are a couple of ways it learns. As users come in, and either go into the search box, or click on something in the buzz cloud, they are presented with a search results page, which again is already going to be very unique. So, even the first user within your new Swicki, or new customized search engine is going to be provided unique search results based on your filtering and your guidance, right? The first piece is that we track it the user’s implicit behavior, which is basically just a bunch of users using the Swicki, and we analyze that behavior. We then behaviorally re-rank the search results based on that behavior.
Eric Enge: Right. So, if you have a hundred people who do similar searches and given more or less the same options and they are always picking the third site rather than the first. The third site might get moved up.
Steven Marder: That’s exactly right. Some of our deepest technology, and thinking, and experience has been around analyzing that click stream. It’s not just pure clicks, because as we all know as searchers, we click on lots of different things before we find what we are most interested in.
So, we leverage all that behavior, we take that kind of community collaboration and collaborative filter, and then we apply that against the baseline set of filtered results. Then, we behaviorally re-rank, and that’s dynamic, and automated, and it’s constantly happening. So, the engine is constantly learning from that community contribution. This first piece is implicit, as the user of the Swicki does not have to do anything.
We also recognized that there are a small portion of users that want to actively contribute, that either have a perspective, an expertise, or a passion to contribute. We see this with other services like Digg and Reddit, and others.
To take advantage of this, we wanted to open things up and leverage that small percentage of users, and let them contribute, exactly like Wikipedia. 2% to 3% of Wikipedia users actively contribute, and then 97% are reaping the dividends from them.
So, what we did is we opened up to the community some explicit features where we allow users to come in and vote for results, or vote against the results. We also allow users to contribute content.
Eric Enge: New resources.
Steven Marder: Exactly, and so from our perspective that’s another layer of input into the system to make the system smarter, and to provide more relevant results.
Eric Enge: Yeah. So, an example might be that they know that there is really great content on a particular website that’s not in Swicki as far as they can tell. So, they would submit that to the Swicki.
Steven Marder: Yes. When you think about it, you’ve got enthusiasts out there. If you use a surfing example, you’ve probably got a surfing enthusiast as maybe the publisher who has got a surfing blog. Then, you’ve got a bunch of surfing enthusiasts who already are going to that blog, and to that engine.
They are cruising around and looking for content, and they are finding really relevant content in other places. For example, perhaps there is a certain beach in Hawaii that just has a great break, and there is lots of information around that beach online. But, if you just went and did a basic search at Yahoo or Google, you are never going to get access to that specialized content, or it might be the hundred and twenty-fifth result, right?
What they can do is they can explicitly contribute that content, or a specific page on that content. Let’s just say it’s the twenty-third result in the Swicki, and a bunch of these users start going there, and a few vote on it. All of a sudden you can pop up much higher, which is to the benefit of the entire community.
Eric Enge: Right. Do you crawl the entire web on a regular basis, or when someone defines a new Swicki, do you do a custom crawl on the sites that are selected?
Steven Marder: Great question. So, to take a step back on that, what we decided is Yahoo, and Google, and Ask, and others have done a fantastic job in going out there and organizing all of the web’s information. We said instead of crawling the web and trying to also do that, let’s leverage those feeds as a baseline, because they are all pretty much general search feeds.
So, what we do is work with the best of breed for organic indexes, for web search, for blogs, for images, and for video. For video we work with Blinkx, and for images we work with Picsearch, and Ask, and others. On the web search side, we work with Yahoo, and Ask, and others.
We bring in all these different feeds, and then we allow those publishers to filter those feeds. We then do some things to ensure freshness of content. This is important, especially in the blogosphere, where the blog indexes definitely have issues around freshness of content.
Getting that fresh content, something that’s been posted today, into these main indexes, is a challenge. From our perspective, if you are on a blog and you are doing a search, you certainly want to access all the freshest information; because sometimes that’s the most relevant and the most recent info.
What we’ve done is that when you build a Swicki, we’ve got a system that does crawl your site to upgrade the information, to make sure we have the freshest information. Shortly we are going to be rolling out some additional backend technology, that’s going to make sure that the most freshest content, and the freshest information will be prevalent in that engine. We think that’s really crucial in providing the full experience.
Eric Enge: Right. So, for blogs you do a certain amount of your own crawling just to ensure freshness?
Steven Marder: Exactly.
Eric Enge: For other types of sites where freshness isn’t so much of an issue, you rely on the indexes from people like Yahoo and Ask.
Steven Marder: Yes, that’s right. The index is the baseline, then we apply some tools to bring in the freshest content; and then we allow the community through their behavior to modify the results even further.
Eric Enge: Right. Of course, there are other types of sites where freshness could be an issue, such as popular news sites like CNN, which have the same temporal nature of a blog, but they are not blogs. Is that something that you also address?
Steven Marder: In the past, we haven’t addressed it as much in terms of the news oriented sites. We’ll be rolling out some features shortly, when we come out of beta, which will address that a bit more closely.
News content is a certain type of content, which tends to be very fresh. Then, you’ve got of course the deep search content. We really believe that most communities really want the ability to see both, that fresh content, but also have the ability to dive deep.
Eric Enge: Can you talk about your approach to universal search?
Steven Marder: We’ve always believed that engines ask users to make too many decisions upfront. In a tabbed world where you can choose site search, image search, video search, etc. That presents too many choices for users, and yes choice is good to a point. But of course, data always shows that whatever the default is, what users tend to do.
From our perspective we said (a): let’s give the publisher some control over that, over which types of content to prioritize, and then, (b): let’s present the users with different types of content. This can be images, video, and text. The text data could be coming from the blog world as well as traditional websites. Why not give them that blend, and then as they use it, let the community along with the publisher further define what types of content that they are most interested in?
If it turns out that it’s really mostly news content, or it’s really videos, then the system will leverage that passion and that focus in their behavior to present that. We’ve always believed in what now is considered universal search; and our approach is still a bit different. We tend to not necessarily want to chunk it, so we blend it in throughout the results and let the community re-rank both the types of content as well as the specific content.
Eric Enge: Right. Google is currently the other one who tries to integrate it, and the other search engines seem to segregate it into different areas of the screen.
Steven Marder: Yes, that’s correct, at least for today. It is changing, we’ve been at this for some time, and we’ve learnt a lot around the types of content that users want to focus on. We are definitely emphasizing more of the fresh content going forward, and in making sure that that fresh content is really available to users, and available not just from a search perspective, but also available within the buzz cloud.
We’ve always believed that search is not just a keyword type of experience, but there is definitely an element of discovery in it. That’s where we see the buzz cloud is really helping to facilitate that discovery.
Eric Enge: Right. If you are trying to research a new topic it’s useful to see what other people who searched on that topics were interested in.
Steven Marder: In some ways it can be research, and just broadening your awareness. Other times it’s just fascinating to know that other members of this community that I belong to are interested in this topic, or these bunches of topics. And, that’s where serendipity definitely plays a role.
Eric Enge: Right. Who do you see as your main competitors?
Steven Marder: On the custom search side where you can create this filtered search engine Yahoo and Google have products that allow you to do some of that. We them as somewhat competitors for distributed search solutions. But, we’ve really created a product that we think is fairly unique. It is leveraging some bits and pieces that other players are providing, but we think that our package is pretty unique out there. So, we think that anybody who is providing a distributed search product that is integrating various types of content, and then applying a social component on top of that would be a direct competitor, but we don’t really see anybody doing that.
Eric Enge: Right. So, Google and Yahoo are indirect as opposed to direct competition.
Steven Marder: Sure. Social search is a key area to talk about. We are considered by Danny Sullivan, and others, to be a pioneer in social search, because our approach has always been very different. We’ve always been very focused on powering search for publishers of all shapes and sizes. We’ve not been about creating a general search engine, or a general social search engine. That’s where we see most of the players out there right now who are leveraging social search methodology, and different types of methodology. But, all applying some human component to search, almost all of them are very focused on basically taking on Yahoo, and Google, and Ask, and building a destination. Acquiring consumers and end-users and owning them, and that’s not we’re about. We’re really here to again empower publishers to provide that highly specialized search experience. We believe that allowing them to leverage the passion and the collective intelligence of their communities is exactly the way to go.
Eric Enge: Do you have any specific partners that you can mention, that have been particularly successful in leveraging the social aspect of Swickis?
Steven Marder: We’ve got over a hundred-thousand Swicki’s which have been created to date. We’ve got over twenty-three thousand unique publishers, and what’s been really interesting is we’ve got big guys like the Disney’s of the world, who are using our technologies to provide specialized search for Family.com.
Then we’ve got the TechCrunch’s of the world that are trying to provide a highly specialized search experience with the Web 2.0 community. Then, we’ve got thousands of these other Swicki’s and Swicki builders and owners out there that are not only leveraging the custom search piece in providing this highly filtered experience; but then leveraging their communities of different shapes and sizes.
We get feedback probably daily from very pleased publishers and builders, about what they love, and what they tend to love is the fact that they can control this thing. They can build it in their own image, and continue to moderate it and guide it, just like the rest of their site. It’s really about them and their brand, and they can customize. And so, yeah it’s been really interesting, and we see the value proposition moving well ahead with these different customers of different shapes and sizes.
Eric Enge: How do you see the overall market size for what you are doing at this point?
Steven Marder: The market is fascinating. After all, anybody with a website or with a blog tends to have some community around it. Whether that’s vertically oriented, or locally oriented, or whether it’s just around social connections. We think everyone of those should have a highly specialized search engine, or Swicki that is going to leverage that passion of that community, and the expertise of the publisher. That’s a pretty big universe when you think about the blogosphere, and when you think about the web building world.
It’s also interesting when you look at what’s happened with widgets, and the ability for people to not just create widgets, but to share and syndicate them. The front end of our product has always been a widget.
The buzz cloud is widget, and so what we’ve seen was really fascinating, is we’ve seen not just publishers with websites and blogs coming in and build Swickis and put them on their sites. But, we’ve seen say consumers who fit what we call “infopreneurs”, come in, build a Swicki, or multiple Swickis, and take that widget, and put it into what we would call widget distribution. So, for example we’ve got one infopreneur out there who built a bunch of Swickis, one of which is a Swicki on love poems and quotes.
They took that, and they built this widget, and they customized it, and they branded it themselves. And then, they took it, and they put it on Widgetbox. And, on Widgetbox all of a sudden we’ve seen that, that specific Swicki or widget subscribed to and installed over five thousand times.
What’s really fascinating from our perspective is it is really in some ways a very different use case then where we began. All of a sudden what we are seeing is consumers and users grabbing that widget, because it’s fun and it’s interesting. They are putting it on profile pages, or putting it on their own site, but somebody else built it.
Eric Enge: There are a couple of different dimensions here. There is the process of using a Swicki on your website to increase customer satisfaction with your site, and then second dimension would be to increase the level of community on your site, and the participation in a site. But then, what you just touched on now is completely different aspect of this which is ways that it becomes a marketing tool in a direct sense.
Steven Marder: Yes, that’s right. This touched on syndication and the distributed nature of the product and the community. Say you’ve got a site on surfing, and you build a Swicki and put that widget on your site to reinforce your brand and satisfy your community of end-customers.
At the same time you can allow some of your users, whether those users are consumers with a profile page, and they are surfing enthusiasts, or whether they also have a blog, or a site on surfing, or whatever, to grab that Swicki and put it on their page or on their site.
They are doing it for the same reasons, because it’s a benefit to their group or community, or it’s just a badge of some kind on their profile page, because they are a proud surfer. For you as the builder or the publisher now it’s an ability to have your Swicki help market and promote your content, and your brand. And it’s really a tool to build your community, but off your site as well as on your site.
That’s really fascinating and extremely cool for us, because we see infopreneurs who are understanding this. They are building these things even without a site; they are building them because they know that they provide value to those that are using them. What they are doing now in the widget distribution world is putting them out there, and allowing people to grab them. So, they are marketing and promoting those specialized search engines or those Swicki’s and allowing different people to grab them; and they are building a business around it.
Eric Enge: Right. How about the search volume, how has that faired in past year? In an interview with Grant Ryan (Chief Scientist of Eurekster) last year he told us that you were doing about five hundred thousand searches a day.
Steven Marder: Yes. At this point we’re doing around eight hundred and fifty to nine hundred thousand searches per day. As you may recall as well, that’s globally as well. We have search activity in probably about fifteen countries. And, we’ve also got people building Swickis in Asian languages and European languages as well.
In terms of the unique users, one of the latest Nielsen reports we’ve seen says we are interacting with or touching over nine million unique users on a global basis.
Eric Enge: Are there any upcoming new community features planned?
Steven Marder: Yes, we will be rolling out some additional features that will leverage community interaction. We have always believed that our products are publisher guided, but community powered. We continue to work on making it easier for us to leverage that community empowerment and engage the community to participate in building or making these Swickis better. We will be rolling out some fairly significant features in the near term. When we get a bit closer to that, we’ll chat with you about that, because I think you’ll find that really interesting as well.
Eric Enge: Right. I would think that one of the things that would be very nifty to do is to start creating overt ties with other social networks.
Steven Marder: Yes. As you may recall when we began we, one of our first big deals was with Friendster. What we would do is look at you and your social network out three degrees, your unique network. We would watch all the search behavior within that network, and then we’d leverage that as metadata to provide a customize search experience.
In a lot of ways that was true social search or an early version of social search. It just wasn’t necessarily guided by a publisher, and there was no builder. Interestingly now, a few years later, there is OpenSocial and Facebook and the concept of the social graph. We’ve always believed that we can leverage these communities to provide an even more specialized experience.
For us it’s a real interesting time, and that in some ways not just the market, but the availability of APIs and access to certain types of information on existing networks out there, really opens up some things for us. And, in some ways it really may allow us to kind of see our initial vision through.
Eric Enge: Right. There is the relationship between the individuals that you can leverage, and that’s one of the things that I think will open up dramatically as these APIs become more available.
Steven Marder: Yes.
Eric Enge: I think that will have a dramatic impact, and then there was actually the possibility of conducting searches through the community itself.
Steven Marder: Exactly. There are a couple of things around that. There are the relationships in the connections that we can leverage. Within that, where it gets even more interesting from our perspective is when you overlay topics of interest on top of that. If you’ve got a big community, then all of a sudden you can know that community down to one enthusiast at a time and how they are connected.
Leveraging that passion and behavior, whether it’s a vertical interest or a local affiliation, it gets really, really interesting. Most of these platforms and networks are opening up or at least trying to open up, that does give us more of an ability to leverage our technology with that extra data.
Eric Enge: Right. You could see an individual really wanting to tie into a variety of different Swickis. They might be active in contributing to a Swicki if they are particularly passionate or knowledgeable about that topic. On the other hand they might want to simply have access to a Swicki that they think gives them better results for their favorite sports team.
Steven Marder: Exactly, right. That’s a really good point, and that’s always been our view. We are all members of a handful of communities and interest, and when it comes to kind of the early definitions of personalization, it’s always been hard figure out how we, as a technology or system. analyze you; figure out all these different interests of yours through behavior to then provide you with a personalized experience. History has shown that’s really hard.
Do you even want to provide that information because of the privacy issues with it? To your point, I mean as you cruse around, you may be a member of various communities. As you interact within those very specific communities, that behavior can be used very directly and succinctly to create a better experience not just for you, but others in that specific community, and leveraging whether it’s social connections or vertical connections to make that better, it is very powerful.
In a lot of ways this was our original vision. It’s very exciting for us to feel like the market and the APIs and a lot of other things are coming together to allow us to make it happen.
Some people will do that sports search quickly and efficiently from wherever they are, and then for the community at large to leverage that behavior is cool. That’s where we think the distributed nature of all this is so fascinating when you look at the web and the multiple connections that are out there, and leveraging them. Ultimately search will become less about where you are when you search. That’s where social media has been so fascinating.
That’s been our philosophy from the beginning. It’s away from centralized publishing, it’s really about decentralizing it. It’s taking the power of centralized publishing away from the one publisher, and it’s allowing the community through their behavior to help each other to build more community, and also to improve the use of search and provide access to content, products, and services through these great technologies.
Eric Enge: Right. What about video Swickis. I know we talked a little bit about universal search before, but it seems like that you have a focus on pure video Swickis as well.
Steven Marder: Videos are an interesting type of content. Video search like core video search is still really challenging for all the issues that I think even the readers know in terms of determining what the video is about. We have always thought videos are the type of content that we want to apply our social search technology to. We do bring that into the universal search environment, so video can just be one type of asset, or one type of content that we serve.
But, we’ve recognized that there are some communities or some publishers that just want video content and want to create this highly customized video Swicki. We wanted to allow them to do that, and then also because video is so much about discovery, more then it is about deep search for most people, we took the discovery benefits of our buzz cloud and applied it to video.
When you look at our video buzz cloud it’s really cool. It’s fun and compelling, and it really kind of gives you a feel for that type of content. Even for us here at Eurekster we sometimes end up getting sucked in. You’ll find yourself looking at lots of different videos through the buzz cloud. You initiate the search, and then you are on the results page, and you’re playing videos; and all of a sudden you are sitting in your office, and you’ve just watched a dozen videos, and you remember you need to get back to work.
It’s not just about applying our technology to video and video search. We’ve done a lot of work to leverage our relationship with Blinkx. We are really focusing on that buzz cloud, and how do we make that compelling; how do we make it sharable and how to give our publishers and our customers the ability to syndicate that. That becomes really interesting content, especially around entertainment type areas, or pop culture type areas of interest; and so all of a sudden when you put that compelling video buzz cloud on a profile page it’s not just informative, but it’s fun.
Eric Enge: How have the video Swickis been received?
Steven Marder: It’s doing well. We’re pleased with it. Our view is that video is big and getting bigger. There is still such a long ways to go in terms of improving results as well.
We are seeing some good traction; we can probably send you some links to some good video Swickis as well (here are the links: Social Media, Family Guy Clips, The Simpson’s, Top Travel Destinations, and Live Music Performances). In the longer term when we look at video we try to think about how we can help.
That may be through the implicit behavior or through the explicit behavior of users. If we can leverage a lot of this community engagement and community behavior to then further refine what videos are about or to specifically tie videos to queries; it’s really helping us build up this set of metadata. We think over the long term can really help video search overall.
Eric Enge: Right. Thinking about universal search again for a moment. One of the challenges is when you take a diverse set of content is trying to figure out how to rank a video versus an image versus a text webpage. I would imagine that certainly initially, the challenge would be much the same for you as it is for other search engines that are trying to do that. Overtime the community elements would allow you to adjust any “errors” in the algorithmic determination.
Steven Marder: Yeah. For us the great thing would be tying in those feeds, whether it’s Blinkx or others. Then we can refine it and leverage publisher guidance, and their communities to further refine that and really make it more relevant and usable from both the search and discovery perspective. As you say, baseline video search is still so challenging. We think that applying the social components to video can be even more compelling in terms of bringing relevance to it.
Eric Enge: What is the business model that you use?
Steven Marder: We help our publishers drive more traffic, certainly more search related traffic and more page views and page search opportunities. Then we generate revenue and share that revenue with publisher. It’s a revenue share model with most of our publishers, and we are serving up sponsored links through our ad network partners, and we are also testing some display advertising as well.
We believe again there are opportunities to provide highly focused and targeted advertising, more advertising opportunities for advertisers to communicate directly with the specific end-users or members of these communities. As we generate this incremental revenue we want to share that with our publishers. So, today when you build a Swicki, you can share the revenue that’s being generated through that Swicki, which can be CPC or CPM revenue.
We are try to optimize those advertising fees as best as possible to the benefit of end-user, to the benefit of advertisers so they can connect directly with the appropriate end-users, and then to the benefit of publishers in terms of revenue.
Eric Enge: Right. Who are your ad network partners?
Steven Marder: We are working with Ask, and we allow our users to use Google AdSense, and we have a relationship with Business.com for business Swickis. We’ll be rolling out a local advertising feed soon for locally oriented Swickis, or geographically targeted Swickis.
We are constantly trying to work with various ad networks to provide the most optimized advertising for these Swickis. Because, when you think about it and you really take a step back and look at it, we are creating highly specialized inventory.
We’ve got over a hundred-thousand search engines, and each one of those is serving a unique community. It really provides a highly targeted opportunity for advertisers. Our challenge is to partner with ad networks that have the capability to leverage that metadata, and leverage the specialized nature of that inventory. That’s still a work-in-progress to be perfectly honest, but our primary focus has been building the inventory first.
In the longer term we think the Holy Grail is in taking these twenty-five, thirty million searchers a month, and beyond, and allowing the most appropriate advertiser to serve that specific end-user the right ad content.
Eric Enge: Thanks for taking the time to speak with me today.
Steven Marder: I appreciate the opportunity. Thank you.