Brands are worried about the company they keep online. Preventing inappropriate or offensive material from appearing next to corporate messages is a top advertising industry priority today, and a necessary step if web advertising is to flourish. Yet ad agencies hadn’t been able to finesse the technical challenge. Protecting brands online requires mining, rating and filtering data from billions of web pages, with high confidence and at blinding speed.
Thanks to NYU computer scientists, under the leadership of Foster Provost of Stern, AdSafe Media solved the puzzle. AdSafe Media uses computational data science, pattern recognition, machine learning and crowdsourcing technology, to scour and analyze billions of Internet URLs. The patent-pending process, by computers with human input, interprets URLs’ text, image and video content and rates them for suitability—all at the very moment a viewer is on a particular webpage.
Professor Provost led the data analytics from the beginning and continues to advise. He brought over a team from NYU, and AdSafe employees from NYU today number seven out of a staff of 25, including two Stern MBAs and a graduate of Arts & Science. Provost designed and developed AdSafe’s data science architecture, with NYU-Poly Ph.D. student Josh Attenberg, during the summer of 2009. Professor Panos Ipeirotis, the George Kellner Faculty Fellow in Stern’s Information Systems Group at IOMS, was brought in to enhance human calibration. Attenberg now helps to lead the company’s data science efforts.
Today, AdSafe fills a floor on Mercer Street in SoHo, intentionally chosen for its walking distance from NYU, which makes meetings fast and easy. Coriolis Ventures incubated AdSafe with $500,000 in seed funding. To date, AdSafe has raised about $7 million more from angels and VCs, including Founder Collective and Atlas Venture. Though it’s not profitable yet, about 100 recognizable national brands use the service. AdSafe also works with most of the global ad agencies. Meanwhile, scientists are benefiting: AdSafe’s massive data sets are available for experiments and have resulted in about a dozen academic papers.
