INTRODUCTION

Online video normally refers to video content viewed on a PC and delivered over broadband Internet connections. Several factors have contributed to increasingly being recognized as explosive growth of online video:

1. Video can be created and edited easily and at very low costs
2. Online video platforms provide close to zero cost distribution (from content producer's perspective)
3. Broadband penetration (and bandwidth) and PC technology and ownership in the developed world has reached a point where online video is a viable choice for consumers
4. Bandwidth costs and storage costs have fallen to a point where online video platforms are able to provide free hosting to content producers.

TRADITIONAL VIDEO

Traditionally, commercial video content was created by entertainment companies such as Viacom, Disney, News Corp, EMI group, Sony Entertainment etc and distributed through TV broadcast, cable or more recently IPTV. The distribution networks were closed in the sense that content creators could not get to consumers unless they had negotiated and gained a channel on the distribution network. This arrangement essentially required scale both at the level of content creators and the video channel owners (such as ABC, NBC and CNN etc).

REVENUE MODELS

1. Broadcast video generated almost all of its revenue from advertisements.
2. Video distributed on cable made its revenue from cable subscriptions and advertisements.

ONLINE VIDEO

Online video has disrupted the traditional video value chain - whereas the large content creation behemoths like Disney still have a place, most of the online video content that is being consumed is Do-It-Yourself (DIY) video created by hobbyist or unknown artists. Because of large scale, the traditional video channels were able to expend effort to reach a large audience. With online video, the content gets popularized through word of mouth or social networks. Consumers also find content through browsing or searching. There is a proliferation of online content but most of the content gets viewed only a few times - that is most of the content viewed is in the 'long tail' in statistical jargon. The traditional video producers are increasingly concerned whether choice will triumph over quality.

THE MAJOR PLAYERS IN ONLINE VIDEO

1. YouTube/Google Video: Clear leader with 80 million unique visitors in September 2006.
2. Yahoo!Video: Recently enabled user generated video postings to be 'comprehensive'. No major differentiation from other big players except that Yahoo!Video provides search across all providers.
3. MSN Video: Mostly deriving its viewer base from professionally produced video clips, MSN Video added user generated videos this Fall with the 'Soapbox' site.
4. MySpace Videos: Primarily a social networking site that also allowed video sharing. News Corp recently acquired MySpace and negotiated a deal with YouTube for revenue sharing for cross referenced videos.
5. Metacafe: Quality control and incentives for contributors
6. VideoEgg: Ease of posting video. Partner program allows partners to add video upload capability to the partner's website
7. Veoh: Full length video upload (c.f. 100 MB limit on YouTube). Allows download of content that can be played using its proprietary client.
8. Brightcove: Wants to differentiate itself by providing users the ability to create internet TV channels.
9. Daily Motion: A French company with more than half of its video content in French. Already has 16 million page-views a day and is an example of a specialized site with local focus.

ONLINE VIDEO PLATFORMS

[Graphic sourced from Economist.com]

INFORMATION COMPLEMENTS AND MARKET SUBSIDIZATION

CONSUMER MARKET

The online video platforms provide not only access to content but also video client - either as an embedded client within the browser as in the case of Google video or standalone client as in the case of Veoh. The consumer market is subsidized so that there is no charge for the video client.

CONTENT PRODUCERS MARKET

Both corporate and DIY content producers need an online video platform that provides storage and access bandwidth as well as captive subscribers to the site. The online video platforms subsidize this market too. In fact, corporate content producers such as Warner require online video platforms such as YouTube to pay royalty on Warner-owned content. Most online video platforms do not compensate DIY content producers with the notable exception of Metacafe.

ADVERTISER MARKET

With consumer and content producers being subsidized (or compensated) in a double sided network, most online video platforms will eventually have to build their business models around advertisement. None of the players have demonstrated viability with the advertisement model as yet but it is a work in progress. The advertisers are looking for the 'eyeballs' that these online video platforms provide. However, they are also chary about the appropriateness or the legality of content and whether their advertisement will reach the right target audience. We discuss this more in the next section.

ONLINE VIDEO ADVERSTISING

The boom in online video is made possible by the growth in online video advertising, the primary means of monetizing video. Online video advertising spending is expected to total $410 million in 2006, representing 2.6 percent of total online ad spending. By 2010 online video ad spending is expected to reach $2.9 billion, representing 11.5% of total online ad spending. Although growth rates for online video are very high, online video ad spending will still be only 3.3% of total television ad spending in 2010.

The majority of the growth in online video advertising will come from advertisers shifting ad spending from television. Video is a familiar creative format and long-favored mass medium for big brand advertisers and ad agencies. Though TV may reach a broader audience quicker, online video advertising reaches an audience that's easily targeted and difficult to reach via TV. Unlike TV, online video advertising delivers consumer interaction and engagement in a contextually relevant environment that's measurable. Dynamic Logic's MarketNorms 2004-2005 data showed online video ads raised brand and persuasion metrics at a statistically significant 90 percent confidence level; specifically, aided brand awareness had an 8 percent lift, message association 38 percent, brand favorability 6 percent, and purchase intent 7 percent. In addition, viewers cannot skip online video ads when watching video content. These qualities enable online video ads to command higher CPM (cost per thousand impressions) rates of between $20-$40 which is much higher than traditional text and banner ads ($3-$10).

Video ads primarily monetize long-form and short-form premium video content which are streamed online for free on an ad-supported basis. In March 2006, ABC started streaming episodes of its hit shows such as Lost and Desperate Housewives online for free. A total of 4 video ads from the same advertiser were played at the start and in the middle of the show as part of an integrated ad campaign. In fall 2006, NBC, CBS and Fox have all joined ABC in streaming the latest episodes of hit shows online on an ad-supported basis. These shows have attracted premium advertisers such as Procter & Gamble, Unilever, Honda, Fidelity, Home Depot, and Nissan.

The length of online video ads range from 5 to 30 seconds, but 15 and 30 second ads are the most common. A study from Microsoft suggests that consumer attention span for online video ads typically last for 5-7 seconds before starting to turn to annoyance. This suggests that video ads lasting 15 seconds or less may be more effective.

Monetizing User Generated Video Sites

While user generated video content has grown tremendously online, monetization is still difficult. YouTube and a majority of user-generated video sites have shunned video ads so far, believing that viewers do not have the patience to watch pre-roll video ads (i.e. video ads shown before the start of a clip). Instead, their revenues today come from traditional banner ads, Google Adsense, and some sponsored content. Revver.com, although not yet a major player, has attained the most publicized success in monetizing user generated clips thru video ads. Revver attaches video ads from companies such as Microsoft, Universal Studios, and Time Warner that play at the end of each clip. In June 2006, a hit "Diet Coke & Mentos Experiment" clip featuring footage of elaborately choreographed fountains of Diet Coke spurting into the air after Mentos candy is dropped into cans and bottles generated over 5 million hits in a month. Its popularity was reported to have generated $60,000 in ad revenues for Revver, 50 percent of which went to the clip's producers.

However, it remains unclear how effective this model would be for the vast majority of user generated videos who do not have a high production quality like the Diet Coke & Mentos clip. Advertisers are very careful with associating their brands with any content that may harm brand perception and association, as well as illegal or inappropriate content. Delivering relevant video ads thru targeting is also difficult. One big difference between search ads and video ads is what Google, or other providers, are able to target the ads against. With search, a user is clearly telling the search engine, and thus the advertiser, what he or she is looking for. With video, there could be several "ad hooks" throughout a video clip that ads could be targeted against, and the challenge is to create a system to identify those hooks and target ads against them. "Measurement is also a problem with online video, since it is much harder to provide meaningful analytics for video than for a standard HTML page. That doesn't prevent advertisers from demanding the same level of data from agencies and publishers", said Karen Anderson, VP and director of media at Digitas' Modem Media, during a Digital Media Innovators panel at a recent Ad conference. Using a recommending system, we believe, can boost revenues on both sides of the network.

Recommendation Systems

A recommendation system is software that attempts to determine what items a user may be interested in based on the user's previous actions. For example, if a user has previously ordered a book about computers, then the system may recommend a book about PDAs. Generally, the goal of a recommendation system is to help a user find targeted products or content that he or she would have otherwise missed among a large selection.

Understand Users First

Recommendation systems begin by obtaining information about the users. This information can be obtained primarily through two methods: implicit and explicit collection. Implicit collection analyzes the products a user has reviewed, the amount of time a user has spent reviewing those products, or the products a user has bought in the past. Explicit collection analyzes how a user has rated other products, how a user has ranked a collection of items from most favorite to least favorite, or the items a user puts in his or her "wish list."

Recommend Clips Based on User Data

Once this user data has been collected for enough users, a recommendation system will determine which items will be of interest to specific users. There are three different types of recommendation systems that determine these recommendations differently: item-based filtering, collaborative filtering, or a hybrid of item-based and collaborative filtering. Below we describe the differences between these three filters. Item-based filters recommend to the user items that are similar to items the user liked in the past. With collaborative filtering, the user is recommended items that similar users have liked in the past. Finally, a hybrid can be built that combines the recommendations from the item-based and collaborative filters. Despite the wide variety of implementations, the filters must have user data before they can begin to recommend items.

Potential Problems with Recommendation Systems

Although recommendation systems can provide many targeted and sophisticated recommendations, they are still prone to numerous problems. Initially, the system may have little or no user data, which can make it very difficult to recommend items. These poor results may give the user an abysmal experience, driving them away from the service. Users may also be worried about privacy issues when visiting sites that implement a recommendation system. The users may feel that a "Big Brother" is watching over their moves. Another worry is that a user will end up becoming too complacent on recommendation systems and will not discover new content. Although recommendation systems are powerful tools, they are only one tool among several that online users have, such as browsing, searching, and friends' suggestions. Thus, users will constantly be exposed to new items through various discovery mechanisms. One final problem is that recommendation systems can often suggest items that a user is uninterested in.

Successful History

Despite the potential problems and risks, numerous companies have successfully implemented recommendation systems as part of their service. The best example is Amazon, an online retailer, which suggests products (i.e. books, DVDs, etc.) based on a hybrid filter as described above. In fact, Brynjolfsson, Hu, and Smith showed that 47.9 percent of Amazon book sales came from books that ranked outside of the top 40,000 titles. Netflix, an online DVD renter, has also implemented a recommendation system with great success. According to Chris Anderson, 60 percent of Netflix's rentals come from recommendations and they have even offered a $1 million prize to anyone that can improve their system by 10 percent. Pandora is an interesting website that recommends similar songs based on a user's previous song selections. Pandora uses a typical recommendation system but also supplements it with recommendations from hired music experts which rate songs on hundreds of musical attributes. Google News even has an obscure feature that allows you to have news stories recommended to you.

Application to Online Video

The relevant question for this paper is: How do recommendation systems apply to online video? Although recommendation systems have been applied to nearly all types of online content, no company at the time of this paper has implemented such a system for online video clips. Based on our research we are confident that recommendation systems will have a significant role in helping users discover video clips that would have likely been missed otherwise. First, users have only two ways to discover online video. They either browse web sites or search the web. Thus, if a user does not browse a certain web site, he or she will miss video content. Video search engines have yet to be perfected and often are only good for very specific searches when the user knows what he or she is looking for. A recommendation system, on the other hand, could recommend clips that similar users have watched or similar clips based on a user's previous watching habits. Second, in general much of the content on the web today is refreshing content that is changing seemingly every second. Users need a solution that will push and filter content for them based on their interests. This allows users to avoid information overload and for them to consume more information quickly. A video recommendation system could offer such a solution because it takes user data and finds potential items of interest for the user. Third, people are used to watching television while leaning back on their sofa. Since a recommendation system could suggest new videos to watch, a service that can recommend videos will better allow users to continue leaning back while watching video clips online. Of course, people should still have a choice over which clips to watch too.

Monetizing Online VIDEO

A video recommendation system fits into the online advertising market quite well too. Such a system could find relevant video clips for a user to watch as well as relevant advertising. Since the user is watching video content, targeted video advertisements would intuitively be the best option, though targeted text advertisements could be promoted too. Thus, we believe that the same technology that is used to recommend video clips can be used to find relevant advertisements for a user. A user will login to an online video recommendation service first. Then, fill out a basic profile and begin browsing and searching for video. Over time, the service will begin to understand the user and his or her interests. With this information the service would be able recommend clips based on similar video clips and similar users. The users will have new video clips available to them without having to search or browse, which dramatically lowers their searching costs.

Conclusion

Recommendation systems are now an important factor in the information economy helping to reduce both search and transaction costs and providing revenue streams on both sides of a double-sided network. The development of a recommendation system for online video will maximize one constraint on the development of the online video industry, facilitating increased production of video content by commercial and DIY producers thereby making a contribution to the expansion of the information economy.

[A bibliography is included.]