Imagem: Estevan PelliIt is no coincidence that when someone clicks to purchase a television set from a virtual store, the site suggests that the shopper buy a blu-ray player. This random action is backed by a sophisticated technology called recommender system, mastered by only a few companies around the world. Zunnit Technologies, established in the city of Belo Horizonte in 2009, is one of these companies. The company has transformed the knowledge generated at the laboratories of the Federal University of Minas Gerais (UFMG) into a commercial product. This technology is used to monitor users browsing on the internet and thus automatically define their profiles and interests. Personalized contents – such as news, services, and products – are offered to users, based on this information.
To this end, Zunnit installs a code in the client”s site, which can be a virtual store or a content portal. In addition, the system counts on robots (special computer programs). Zunnit”s robots are called Web Focused Crawlers; these robots sweep the sites in search of contents related to the user”s interests. “The robots can accurately sort relevant digital contents from contents that are not that important,” says Nívio Ziviani, professor emeritus at the Department of Computer Sciences (DCC), of UFMG, where the knowledge that generated this technology was developed. “Our robots are highly selective and help provide quality recommendations for internet browsers.”
In other words, this technology makes it possible to know what the user is doing, what he is clicking on, which topics and product the user is most interested in. “Based on this information, the system uses computer techniques, such as the retrieval of information and data mining, among others, to become acquainted with the given user”s profile. It can identify whether the user is a man or a woman; it can verify where the user is located and identify which is the best product to recommend to this user,” explains Lesley Scarioli Júnior, executive director of Zunnit. “If the user is interested in television sets, for example, what can be recommended along with the television?”
The technology developed by the company goes beyond this. It is able to interpret the context of the browsing. “If the user is browsing through a Formula 1 site, for example, and the site contains an article on Paris which contains the words Paris Hilton, the system is able to understand that the words refer to the hotel and not to Paris Hilton, the socialite,” explains Scarioli. “This semantic issue is complicated and difficult to solve, but our technology is able to solve it with 90% efficacy, which is a very high percentage.” This is different from what search engines such as Google do. On Google, the user looks for something specific and, to this end, types the word or words of whatever he is looking for. “A state-of-the-art recommendation system, on the other hand, surprises the user with something he does not know or expect, but that pleases him,” explains Ziviani.
The earliest research studies conducted at UFMG on this topic began in 2008, at the Information Treatment Laboratory (Latin) of the DCC. That study generated master”s theses by Ziviani”s students. Zunnit is not the first company that Ziviani helped establish. He is also a successful entrepreneur with a stake in two other technology companies. One of them is Miner, founded in 1998 and sold to the UOL Portal, owned by Grupo Abril/Folha de S. Paulo, in 1999. The other company is Akwan Information Technologies, founded in 2000 and acquired by Google in 2005; Akwan became Google”s research and development center in Brazil. In mid 2008, Ziviani decided to open up a third company, whose objective was to commercially exploit the recommendation knowledge of systems generated at Latin. “The intellectual property is fully owned by UFMG, but it is important to transform the knowledge generated by research studies into wealth for society,” he says.
In November 2011, UFMG and Zunnit entered into a pioneering technology transfer agreement. The agreement established that UFMG would transfer the knowledge generated at Latin to the company. In return, the UFMG became a shareholder in the company, with 5% shareholding interest. The UFMG will receive the profit from these shares. In other words, the university will have a share of the profits, just like any other shareholder, but will not have any decision-making power. Furthermore, the university will receive an amount equivalent to its shareholding interest if Zunnit is sold.
The idea of founding this company materialized thanks to the existence of a business angel (person or people who have the capital to invest in start-up companies.) In this case, the investor was Scarioli. Ziviani met the investor through friends in common. Scarioli had the capital and the willingness to bet on knowledge that could generate profitable technology. This is how Zunnit was created, albeit with different objectives than its current ones. “Our first idea was to recommend books to web browsers, according to the topics they were looking for on sites they were browsing,” says Scarioli. “For example, if someone was looking for information on the Formula 1 car race, the system could recommend a book on this subject to the person.”
In time, the directors of the company became aware of market demands and changed the focus, re-directing it to the company”s current line of business. “Nowadays, our core business is e-commerce; we make recommendations at virtual stores,” says Scarioli. “Amazon (the American virtual store) was the first company to do this, in 1998. Amazon was very competent in this respect which is why it is a global reference for this technology. In fact, 35% of its sales stem from recommendation.” Scarioli adds that Netflix, which rents on-line films and videogames, uses this technology extensively. 70% of Netfix sales come from recommendations.
However, neither Amazon nor Netflix sell their technology to third parties. Zunnit started to sell its product in December last year. Nowadays, its clients include big companies, especially content portals, such as UOL, Sky and Busk, a social network focused on news, and linked to the Globo group. “We are also conducting negotiations with several e-commerce sites, one of our business focuses,” says Scarioli. “This is the major market for our technology. This line of business turns over approximately R$ 20 billion a year, while the advertising market, for example, only turns over R$ 2 billion.”