Start-Ups Seek to Cash In on Web-Video Ads
By KEVIN J. DELANEY
March 2, 2007; Page B1
Several start-up companies are hoping to cash in on the exploding viewership for online video with systems that can match advertising to video content.
Google Inc.’s enormously successfully online-ad system works by identifying key words users are searching for or seeing on Web pages and placing ads alongside them that are targeted to the same words. So a consumer who reads a Web site that includes the words "Canon" and "digital camera" might see ads from Google for retailers carrying Canon cameras.
Now a handful of start-ups is deploying advanced technology to take a similar approach to advertising that appears with online video clips.
Most agree the market for such ads has enormous potential as blue-chip advertisers look for some of the same branding boost they get from TV. Web publishers also say they quickly sell out ads linked to professional and semiprofessional videos. But advertisers are just beginning to tap the marketing potential alongside the sea of amateur clips that consumers put on Web sites such as Google’s YouTube.
There is little consensus, however, on the best means to scan videos for content, how to display the ads or how to target them to consumers who will be most receptive to them.
The start-ups, with such names as ScanScout Inc. and YuMe Networks Inc., are trying out various high- and low-tech tools to scrutinize the content of online videos, ranging from software that generates transcripts of audio tracks to human editors in India who try to verify that videos’ creators have accurately characterized their contents.
Closely held ScanScout has some of the most ambitious plans. The Cambridge, Mass., start-up uses technology to recognize words spoken in the audio tracks of clips. It then lets advertisers choose to have their ads appear at the moment in the videos when specific words are spoken. The company tested the program with video sites late last year and plans to make it widely available next month.
In a demonstration, ScanScout’s system displayed an ad for a sports-car brand at the bottom of a video clip as the people on screen were discussing that type of car. Consumers can click on the ad to pause the playback and see a video commercial, or be taken to the advertiser’s Web site. By analyzing a video’s content, ScanScout can help companies steer clear of subject matter and language that make them uncomfortable, potentially increasing the confidence of traditional advertisers in buying ads on amateur videos. It can also place multiple ads on a given video clip as what is being discussed changes, something that could let Internet companies tap additional ad dollars.
"This becomes a new revenue stream on top of what [online video sites are] already doing," says Sarah Fay, president of the Aegis PLC digital marketing subsidiary Isobar U.S., which has advertiser clients that plan to try ScanScout. Some experts also predict that similar ad-targeting systems will eventually be used in television set-top boxes to match commercials with TV shows.
ScanScout takes a commission on ads it sells that are carried on other companies’ sites. Borrowing from the search-ad model, it charges advertisers only when a user clicks on an ad, with the flat fee set at roughly 50 cents or less a click to start. Its investors include Georges Harik, a former Google executive who helped build the technology behind some of the search company’s ad systems.
ScanScout is already gaining some traction with customers. Online video site PureVideo Networks Inc. of El Segundo, Calif., says it plans to begin testing ScanScout, probably with standup comedy videos on its StupidVideos.com site. Blip Networks Inc. of New York has tested ScanScout on its blip.tv video site since the fall, and Chief Executive Mike Hudack says he believes it is more effective than just matching ads with the descriptive information the videos’ creators supply.
Brian Buchwald, general manager of an NBC Universal digital venture who has been briefed on the system, says he is impressed with ScanScout’s approach. "If the technology does what they say it can do and it improves over time, it becomes a business execution question for them," he says.
Closely held YuMe Networks, meanwhile, relies on information supplied by video creators and other data sources to group video clips into categories, such as "automotive." Advertisers can then select the category alongside which their ads will appear. YuMe, Redwood City, Calif., says it verifies the information using speech-recognition technology that picks out words in the audio track. It also employs people in India to make sure that videos purporting to be about sport-utility vehicles, for example, are what they say they are. Internet executives say amateur video creators often lie about the contents of clips or are sloppy when they supply information.
PodZinger Corp., a Cambridge, Mass., subsidiary of BBN Technologies Corp., uses technology to generate word-for-word transcripts of the audio tracks of clips that appear on its partner sites. It then classifies videos based at least partly on the transcripts, such as by grouping together clips related to professional basketball. Advertisers select categories of videos they would like to advertise in, and in some cases they can identify specific spoken key words they would like to target.
Some say applying a key-word targeting approach for online video clips may not be the most effective way to go. Online video search start-up Blinkx says it recently tested targeted ads with video from Britain’s Independent Television News Ltd., but it says identifying key words in the text generally didn’t produce better results for advertisers than just identifying the type of content and likely demographic traits of viewers.
Suranga Chandratillake, Blinkx’s founder and chief technology officer, also says there currently aren’t enough video advertisers to provide ads that match many specific key words. For example, even if you identify the words "Canon digital camera" in the dialogue on a video clip, there may not be a video ad to go alongside.
Online-video company Brightcove Inc. says it is selling targeted ads based on the data supplied by the video creators rather than the audio or video information itself. But Brightcove founder and CEO Jeremy Allaire says there is often a disincentive now to applying more specific content targeting to video advertising. Sites with high-quality videos can garner premium prices for online video ads using the traditional online-ad model, with advertisers paying for the number of visitors rather than the number of visitor clicks. In some cases, the start-up ad systems could generate less revenue as a result.
ScanScout co-founder and President Waikit Lau says his company lets advertisers pick categories of videos they want to target, in addition to having their ads appear alongside specific key words. He says ScanScout will at least initially tap nonvideo ads, such as graphical and text ads, to appear alongside videos to compensate for any shortage of video advertisements. And Mr. Lau says Web sites can run their high-priced ads as well as the ones from ScanScout, since its ads appear only intermittently during the clips.
Looming over the start-ups is one worry: that some of the current Internet advertising giants may jump into the game. Microsoft Corp., for one, says it is researching such video-ad targeting. Joe Doran, general manager for Microsoft Digital Advertising Solutions, says technology that ScanScout is using to try to analyze video images is "pretty interesting stuff."
Yahoo Inc. says it doesn’t match ads with video clips analyzing the words spoken but declines to say whether it is interested in the area. Google doesn’t currently let advertisers target their ads in this way either, instead targeting ads based on information provided by the video creators. (Dow Jones & Co., publisher of this newspaper, is testing distributing nonbusiness video clips combined with advertising from Google. It also uses Brightcove technology to distribute video online.)