to more data about user behaviour should improve the targeting (though one of these respondents indicated it was not clear how the ad server could use this data across publishers as this is forbidden by the contracts). Indeed, the notifying party has indicated that such use of the data was contractually not feasible as publishers themselves requested such use to be limited. Publishers insist that data use restrictions are included in their contract terms. The theory of harm relying on data combination (and access to CPI) implies that the new entity would have to renegotiate these terms despite the fact that customers seem to value highly the confidentiality of their datasets (or alternatively that the new entity would have to fraudulently use this data). In any case, the type of behavioural targeting that lies at the core of these direct network effects is an emerging technology which neither DoubleClick nor Google have developed, contrary to a number of competing firms (such as Yahoo!'s ad network BlueLithium or AOL's Tacoda network).
The presence of strong indirect network effects lies at the core of most third- party complainants' theories of harm. These network effects are such that an ad network becomes more attractive to advertisers as the number of publishers increases (and vice versa). The reason put forward is that liquidity is key to success in online ad intermediation and more liquidity is achieved through scale. With a higher number of publishers and advertisers participating in an ad network, the probability and expected value of a match increases. Through the foreclosure strategies described above, if the AdSense network is able to attract additional publishers (or inventory), it will reach a critical size while denying the necessary scale to competing ad networks. According to some complainants, the presence of indirect network effects provides additional incentives to engage in foreclosure strategies as rival networks are more likely to be weakened. While the presence of these network effects is theoretically compelling, the evidence gathered during the investigation suggests that these may not be as strong (or at least, not strong enough to lead to "tipping").
The market investigation has provided evidence that there is significant entry and competition in online ad intermediation as well as evidence on the prevalence of multi-homing and the ability of ad networks to compete even with a relatively small number of partners on the publisher side. The prevalence of multi-homing suggests that the participation by a publisher or an advertiser to an ad network (e.g. AdSense) does not imply that they are unable or unwilling to participate in another ad network, that is to say their participation to an ad network is not exclusive. This may result from the fact that the fixed cost of joining an ad network is either very low or non-existent (as indicated by the notifying party). Multi-homing is also enabled by the interoperability of the ad serving technology allowing publishers and advertisers to provide instructions across several networks. For example, RightMedia's Direct Media Exchange is a free ad network manager allowing publishers participating in several ad networks to automatically display ads on the network paying the highest price per impression (in real time bidding).
306. Data provided by the notifying party shows that more than half of DoubleClick's customers used at least 2 ad networks156 (and sometimes more
See submission of the notifying party's economic advisers of 13 September 2007, page 4.