Peer39: Interview with Amiad Solomon
Founded in 2006 by an innovative group of scientists and engineers, Peer39 was voted by MIT’s Technology Review as one of the Top Ten Web Startups to Watch in 2008. Semantic Universe editor Tony Shaw recently spoke with Founder and CEO Amiad Solomon about the company’s innovations, the FTC and what we may expect in the future.
Tony Shaw: Over the last few years, it seems that your business has changed. Can you update us on what Peer39 is currently doing?
Amiad Solomon: For the past three years we’ve focused on building semantic technology relating to advertising – particularly internet display ads – which is an important distinction because there are many areas in which semantics can be applied. In the beginning, we were building our own network, a semantic ad network and providing our technology to publishers and partners. In 2008 we decided to shut down our network and not continue building an ad network because we felt there were enough ad networks out there. However, we knew we had exceptional technology and a real differentiator, so we concentrated on building relationships and licensing our technology to a wide range of publishers, networks and exchanges.
Our technology is being used as the backbone of our clients systems for two main reasons. The first is for matching content to ads. We have a pretty robust targeting tool that sends signals into the ad server to match and understand every URL or every page. The signal is then sent into the ad server to match every ad to every impression. The best part is that we don’t have to deal with privacy issues because we’re not using cookies. We make sure that the accuracy and precision recall are presented in an optimum way as it relates to the advertising world – specifically mapping our capabilities to what the IAB or internet advertising is interested in. Secondly, we provide this technology for the same type of customers for the brand protection piece. We are able to match each advertiser level and create a safe environment for those brands while filtering out negativity – such as articles about disasters – that would not be suitable for specific advertisers.
We are also signing large publishers, new exchanges, and networks that pass us sets of millions of URLs. We are capable of returning results in real time which is extremely important because they have to decide whether they want to buy those URLs in the exchange.
Tony Shaw: Sounds like business is doing well in the face of pretty tough times.
Amiad Solomon: Yes. Shutting down our network was a tough decision, but it turned out to be one of the smartest decisions we’ve ever made and I think it’s a big differentiator for us. We are in a unique situation in that even though we work with different companies, publishers, and networks, we’re not perceived as competitors to any of them. Q4 was our best quarter ever. The semantic message of advertising is becoming a reality!
Tony Shaw: Are you selling ‘semantics’ or are you finding that the terminology simply resonates with people? Are they more hooked into terms like contextual analysis?
Amiad Solomon: Many people in advertising are saying “yes, we’ve heard of semantics” but they may not fully understand what the term means. When we explain the concept, they don’t automatically default directly back to contextual. Semantics is becoming mainstream in the same way that behavioral targeting is. It’s definitely becoming something that people know or think they should know about.
Tony Shaw: What are the key issues that are of concern to the FTC or to the stakeholders that are pushing the FTC?
Amiad Solomon: I’m not a big expert on the behavioral area, but I do know that the main concern is that since there are a lot of ways to target ads on the internet, there’s a big fear that the good and bad end up in the same basket. There are companies using unlawful targeting strategies.
Tony Shaw: Are they doing that knowingly?
Amiad Solomon: Yes. Companies like Yahoo conduct themselves legitimately by allowing users to determine whether or not they want to share cookie data. However, there are smaller companies whose visitors don’t know how much of their information is being collected. One company that was shut down was hooked up with the ISPs to track users through their accounts.
There is a lot of fear internet-wide about what is permitted and what’s not permitted when you surf the internet. How much of the data of the cookies are you allowed to use and share with others? Peer39 is benefiting from the concern not only in congress, but also with marketers – particularly pharmaceutical marketers.
Coupling cookie data with offsite data and information about prior activities has to be regulated even if regulation is simply enabling people to opt out. Many companies are lobbying hard in Washington for that to happen. But that has more to do with behavioral targeting.
We only look at content, which is similar to what Google does. They look at context and not the cookies. There’s a lot of value in coupling all that data, one on top of the other. For example, you know that a person is interested in sports because he came from ESPN five times in the last month and you know he’s reading a specific article about a particular basketball player on a specific website, so coupling that data is a great targeting tool.
Tony Shaw: What is the mechanism for tying in historical data with the semantic content of the site? Is Peer39 able to provide a better experience for the user because only content is being analyzed? How then do you take advantage of tying in that historical data without essentially undermining that integrity argument?
Amiad Solomon: First of all, we do not provide cookie data at all to our customers. That is a big concern. For instance, Newsweek is a client of ours. They have a big database of their users, what they’re interested in, and what they’ve done. I assume they track all the information about users through cookies which is 100% fine. Problems would arise if Newsweek were to sell that information to a third provider.
We provide data at the user level on the page. We then determine what the page is about and send that information to our clients. The clients then use that data to target by coupling their data with the users’ data. We don’t track what the user did before and I think that’s a very big advantage. Semantic technology sounds much cleaner to our customers today because everything we do is within regulation. Even after regulations have been set, the data we provide will be within those regulations.
In terms of analysis, as an example, for Publisher X, we would look at their news articles which are typically sold for a low CPM. Not many advertisers are interested in the news section. Most are interested in the travel section; therefore, they pay more to be there. So what our technology does is it sits on the pages of the news and technology sections, analyzes those pages, and sends signals to the partners that this page is not only about news, it’s about health, business, and technology and those signals in effect create the situation where the right ad from a technology vendor or a health vendor is associated with the right content there. The result is that they’re able to move the pricing for that page from remnant pricing which is south of $1 to direct prices of $10, $20, $30. We have partners who are selling their pages for $90 and $100. If you can increase the amount of pages they have in those sections they will see increased revenue from it. They need clarity of content. Once they have that information, they can make a lot more money either by selling it directly or by ensuring that the ads are relevant to specific pages.
Tony Shaw: So those $90 and $100 amounts are CPM?
Amiad Solomon: Yes, for 1,000 views.
Tony Shaw: Does the FTC have some understanding of what semantic technology can deliver in this context?
Amiad Solomon: I’m not sure. I think we’re out of the debate. We’re not the problem. It might be an interesting idea to educate them about the semantic space as an alternative to other technologies. Right now we’re more of a solution than part of the problem because our technologies – all the semantic players – are not using cookie data on a semantic level. It might be interesting to tell them about it, but I’m not aware of any initiatives in that regard.
Tony Shaw: Do you have a sense of the time frame with regard to policy making? Are they looking to do something imminently?
Amiad Solomon: I think it will take months. There was a lot of talk about doing it in 2009, which didn’t happen. There are a lot of recommendations and if you follow it you see a lot of articles about people from the FTC talking about what’s prohibited and what’s okay. I’m hoping it will be resolved in the first half of 2010 although it might take all of 2010 to resolve the situation. The solution could be something very drastic in which everything would be prohibited and the semantic space would benefit even more, but to be realistic I think there will be a compromise somewhere in the middle. Some of the methods we’re working with will not continue and some will be allowed. I think in the semantic space we’re okay either way. Anything prohibited through behavioral can enhance the targeting using semantics. I don’t think it will be totally banned. There will be some kind of compromise because of the fact that a lot of people are using cookie data. Any changes would affect a lot of industries if all this data is prohibited.
Tony Shaw: What do you see for the future of Peer39? What’s the next front?
Amiad Solomon: I actually think that 2010 will be a very, very significant year now that we’re closer to not needing to explain what ‘semantics’ are. Data will become everything in 2010 and recent acquisitions in this area have confirmed that direction. Like Adobe buying Omniture. A lot of the big companies – both search companies as well as portals – will be moving dramatically into the semantic space by either developing their own technology or via licensing. In looking at the evolution of targeting on the internet, there was geo targeting and there was textual targeting. Then, behavioral targeting came in and took the big spot and everyone knows the next big thing will be semantics. At Peer39, we are focused on distributing and managing to leverage our technologies at many crucial points in our industry and that’s kind of our prediction of 2010.
Tony Shaw: Its fine for the US and Europe to make these policy decisions, but on the internet it’s very difficult to define legal jurisdiction. Servers can be placed anywhere, users and advertisers can be anywhere. Advertising engines can be anywhere. How are we going to satisfy US or European laws in this server environment where these players could be anywhere in the world?
Amiad Solomon: Good question. I agree 100%. On the internet, it’s much more difficult to enforce regulations. I think the advantage in terms of privacy issues is that, at the end of the day, what people are going after, especially in the US, are US companies. I understand you could also question “what’s a company”. Any companies that have business in the US are being carefully observed and looked at in regards to how they use any user data. I am sure there are questions such as “what’s the right mechanism to it” and “how should it work” and so on, but I think that the key would be to follow where the business is held and where they have business and track it through that. There are examples in other industries such as gambling. In this case, the bad ethics, I think, will disappear. I think you’ll always find one or two companies that are doing something wrong but as an industry I sense that it will be much more about what company has been doing. Businesses acting off mark will be called out on their practices and forced to change. I do agree there’s a real issue of where the servers are, where the users come from, and where the agencies come from.
Tony Shaw: One of the biggest issues confronting large advertisers is who owns the data? These lines have not been defined in the law as yet. It’s one of the areas in which the political system is dragging badly behind the technology. Do you have a sense of where this will end up?
Amiad Solomon: I think that’s probably the hottest topic now. Exactly who does own the data? The publishers, for instance would probably say “we own the data because we provide the content”. But then, the users might say “no, we do because we’re the users”, and maybe the advertisers would say “we own the data because we pay for the ads that provide all this”. So whoever you speak to has their own opinion of how it works.
This is something that has to be resolved this year. I don’t have an answer as to who really owns the data, but we do hear different views from all three parties and at the end of the day, there’s truth to the fact that if someone provides content for free he should have the ability to target relevant ads on order to pay for the content. Of course, whomever it falls to in the end, the data has to be used in the right way and the right manner taking privacy issues into account. In general I think everyone is pushing toward resolving all aspects of your questions.
Tony Shaw: Briefly describe the SemanticProtect service.
Amiad Solomon: Semantic systems can be configured to automatically recognize general terms that are undesirable, such as profanity and sexual language, and detect phrases in text associated with negative events, such as disasters and crime, which may reflect poorly on a brand. For example, semantic targeting would avoid targeting a golf-related ad to content about Tiger Woods’ latest mistress because it understands the true context of the page is not about golf. Upon detecting ‘undesirable’ content, Semantic technology automatically engages preset functions to replace an ad. The space can be left blank, filled with a public service announcement or an in-house ad, or sent to an ad network.
Tony Shaw: What is the ‘semantic’ element? How does it work?
Amiad Solomon: As opposed to other brand protection services that help keep brands away from keywords, semantic brand protection can shield a brand from certain ‘categories’ of content, such as accidents or terror. By understating the true meaning and sentiment of a page, semantic systems can make sure that brands are everywhere they want to be and nowhere they do not. For example, contextual-based brand protection may block a hospital ad from appearing alongside content with the term ‘cancer,’ but miss the opportunity to serve that ad near content about a new clinic that helps to treat the disease.
Tony Shaw: What does an advertiser need to do to its web pages, press releases, ad messages, etc? Do they need to put their own RDFa into their data?
Amiad Solomon: We can integrate with partners in a variety of ways – pixel based integration via the publisher’s web page or within the ad itself; or via our API. It is a very simple, quick process which requires minimal manual labor. We send back results in the client taxonomy or language, so there is no need for customization or tuning on the publisher’s end. These steps enable an advertiser to target the most effective content for their brand or campaign, avoid objectionable content, and allow us to provide in-depth post campaign analysis.
Tony Shaw: Who are your early customers? What are they saying?
Amiad Solomon: Our early publisher customers are seeing the benefit of using semantic targeting to better monetize their own content. As the market is evolving, we are moving to different models with different types of partners, including ad networks, exchanges, and DSPs.
Tony Shaw: Are there additional applications that you foresee beyond ‘brand protection’?
Amiad Solomon: Our core solution is in our ability to understand the meaning and sentiment of a webpage, and target the most relevant and appropriate set of ads to that page. Clients can choose to target to or away from specific semantic ‘categories’ of content. The real magic in our platform is in the ability to allow advertisers to access and advertise in the content that is most favorable to their brand and message, with the most transparency possible.
Tony Shaw: Have any metrics been evaluated previously about the cost of an inappropriate ad placement? What are those?
Amiad Solomon: We are not sure of any specific metrics or studies on this – that is a great question for us to explore with our partners.
Tony Shaw: Advertisers normally pay for placement, impressions, clicks, etcetera. How does an advertiser pay to NOT get these? How do you demonstrate success in NOT placing an impression?
Amiad Solomon: Just as demographically-targeted advertisers pay a bit more for their ads to appear in front of an internet user from one city and not from another, those that use semantics also pay a premium to have their ads appear next to appropriate and desired content and not near damaging or inappropriate content. Peer39 does not get paid based on how frequently an ad avoids negative content; we simply charge a slightly higher CPM for the brand protection that we offer. We have many different business models with our partners, but ultimately, advertisers are paying for our granular semantic classifications, and then deciding to target against or away from certain content ‘categories’ on their own.
- A Tale of Agile Development... of a Standard
- Introduction to: Open World Assumption vs Closed World Assumption
- Summary of 11th International Semantic Web Conference
- Elephant Hunting for Pygmies: How To Introduce Big Data and Linked Data to Your Company