A tiny 14-man startup in Munich called Proximic wants to give Google a run for its money. But it is not going after search. It is coming from behind with an attack on AdSense. Proximic has signed deals to syndicate product listings from both eBay's Shopping.com and Yahoo's Shopping Network as contextual ads on other Websites.
This is the first time that either Yahoo or eBay has syndicated these listings. In one fell swoop, Proximic will gain an ad inventory of 50 million ads (20 million from each eBay and Yahoo, and 10 million from other sources). Proximic estimates that Google, in contrast, has an inventory of about one million unique ads. Proximic's ad network based on this massive inventory will launch at the end of January.
Prospective Web publishers will be able to place an ad widget on their sites. Proximic will index the sites and serve up contextually matching products as text ads along with contextually relevant content links (see demo screen shots below). The ads and contextual Web links will also appear in a sidebar for anyone who has downloaded the Proximic Firefox add-on.
So if you are reading about Mideast peace efforts, books about the Mideast might appear in the ad widget. If you are reading about the Mozilla Foundation, you might get an ad for Firefox track jackets from the Mozilla store, along with a link to Twitter message about Mozilla hiring the guys from Humanized. (This is what actually came up in the Firefox sidebar when I tried it. Proximic also populates its widget results with the content links from 900,000 RSS feeds it has indexed and the top 500 or so Websites.)
What makes Proximic different is that it does not come up with contextual matches based on keyword search like Google or Yahoo would. It doesn't use semantic or statistical methods either to figure out what a page is about. “Semantic systems are not able to scale,” sniffs Proximic co-founder and CTO Thomas Nitsche, a former computer chess champion. “If you hold more than one million documents, you run into a problem,” he concludes. Semantic search, he thinks, is too slow at this point for ad serving.
Instead of keyword, semantic, or statistical approaches, Proximic uses proximity analysis. Nitsche is vague about exactly how it works, but it boils down Proximic's algorithm translating each body of text into a pattern of characters that then becomes represented by a mathematical vector. Matches are done through traditional vector analysis. Or, as Nitsche explains:
We look at patterns of letters. We get a profile. The profile is a vector. We compare two vectors, and compute proximity by pattern distance. We can generate proximity between texts. The text can be one word, two words, 15 words, or a complete page.
Using this method, Proximic can also create matches between product listings and Web pages, thus opening up what is now an inventory of product search results to the world of contextual advertising. In tests, Nitsche says Proximic is seeing click-through rates as high as 1.5 percent, which is much greater than the 0.25 percent or less that is typical for an AdSense campaign. Of course, Proximic has to split any ad revenues it makes with Yahoo and eBay on one side and the Web publishers on the other. Proximic plans on giving participating Websites 70 percent of any revenues after eBay and Yahoo take their cut, leaving it with a very small piece of the pie. The only way to make up for it is by generating much higher click-through rates (by improving the relevance of the ads).
It is an ambitious undertaking for a German startup which has raised only 3 million Euros from Wellington Partners in 2006. But what Proximic is trying to do is combine contextual search advertising with affiliate marketing. By analyzing the patterns of characters on a page, it is creating a machine intelligence of sorts. The nice thing is that Proximic does not need to leave cookies on anyone's browser or track you across the Web. It makes its own judgements about what is contextually relevant based on what you are reading.
If Proximic is successful in matching relevant products, imagine what it could do as a general-purpose search engine? But Nitsche knows better than to try to take Google head-on in search. He is happy, for now, focusing on the advertising end.