By Jeremy Schlosberg

   Ever since emerging as an advertising medium in October 1994, with the delivery of the world’s first banner ad on Wired Magazine’s HotWired web site, the internet has been heralded as a medium with unprecedented potential for measurement and accountability.
    But the availability of hard numbers for certain online actions—whether it is page views, ad requests or click-throughs—belies the fact that all too often these numbers have little practical meaning for media buyers.
  "Very early on the web was touted as the most measurable of all media," says Peter Black, vice president of marketing services for BPA International, a media auditing service. "But it didn’t pan out that way."
    It certainly did not. The key item of measurement, the ad impression, is subject to a number of different and conflicting measurements.
    In part one of our series, Draft Worldwide’s Paul Benjou noted that he’s seen discrepancies up to 90 percent between the impressions a site claimed were delivered and what the ad server said were delivered.
     This may be an extreme, but discrepancies ranging in the 10 to 40 percent range are widespread, according to industry insiders.
    What’s being done about it?
    Not very much.
    "Up to about 15 to 20 percent is a wash," says Leslie Laredo, president of The Laredo Group, a company that does internet advertising sales training, internet advertising consulting and online research. Both web sites and agencies figure it’s usually more trouble than it’s worth to resolve discrepancies lower than that.
    "But 20 percent seems to be the threshold at which the pain gets too high," says Laredo. "That’s when both sides want to get together and figure something out.
    The discrepancy problem can look at first glance like an "us against them" confrontation, with web sites self-servingly claiming high numbers of impressions against agency tallies of far fewer.
     But it turns out, as Black notes, "no one’s doing this on purpose."
    Rather, ad impression discrepancies are sewn into the fabric of the web.
    Here’s how it works: A user clicks on his bookmarked link to, say, AltaVista. His browser "pings," as they say, the content server—the computer that puts AltaVista’s home page contents onto the user’s screen. 
    This in turn pings the ad server—in this case, DoubleClick—whose job it is to load the banner for ComputerShopper.com that’s supposed to go onto the page.
    It can be even more complicated then that if a third-party ad server is in turn being served by an agency ad server.
    "You can wind up with as many as five ad servers in the chain," says Tig Tillinghast, director of agency strategy at Solbright, a company that sells ad management software. 
    "And all of these ad servers may have different numbers. The agency will have its ad server and it winds up with a report saying 80,000 impressions ran. But web site says we delivered 110,000. There’s your discrepancy."
    Why are the two numbers so different?
    The simplest reason is that users don’t always wait around for ads to load. Our visitor to AltaVista may see something on the main page he clicks on right away, before the ComputerShopper.com banner even appears.
     But other things can happen too.
      "Third-party ad servers are not all compatible," says Laredo. "Requests for ads can get lost, or there can be extra-long delays for ads."
    And then there’s the question of when the impression is counted.
    "One ad server may count an impression exactly when the GIF [the graphic file with the ad] is requested, another ad server may count it just after the GIF is requested," notes Alex Flagg, director of agency sales and relations for Spark Online, a company that makes software that helps automate and manage online advertising campaigns.
     "Even small differences in the way the ad server is coded or the way they’re installed will affect the way it counts."
    And it’s not just the wrinkles in ad serving that cause trouble. 
    Web sites do not always filter out hits that are not delivering actual impressions. For instance, all major web sites have staff working continually on the pages; this activity should not be included in impression numbers, but will be unless the site filters them out.
    Also web sites are often being visited by robots and spiders on behalf of the major search engines. These too are not actual visits by users but will count towards impressions unless filtered by the site.
    Caching is one other source of mischief fouling up attempts to tally impressions.
      Caching is when a computer saves something it’s already seen during a session in case it needs it again, which is easier and faster than going back to the web page.
     "Computers do caching at different levels," explains Tillinghast. 
   "A lot of people realize their browser has a cache—when you press back button and go back to a site you viewed, you’re viewing the page from your cache, not from the site itself." In this case, he notes, an ad gains impressions that neither the ad server nor the web site can know about.
    Another type of caching that’s less widely recognized is caching done at a network level. This happens within organizations, to cut down on network congestion. 
    "Your company may serve the same Yahoo page that you saw to your colleague," says Tillinghast. This happens without an extra ping to Yahoo’s content server and the associated ad servers.
    This also happens, as few people realize, across ISPs.
     "You might be going out to the internet on AOL and someone else in the next state may get the Yahoo page you were served," says Tillinghast. "People don’t realize how complicated this is. This is why log files are basically bullshit."
     Even rampant discrepancies would not be a tremendous problem for media buyers if there were an agreed-upon process for dealing with them. 
    Too often, however, no one delineated ahead of time which numbers will be the ones upon which the guaranteed impressions will be based in terms of tracking, make-goods, and whether the entire buy is delivered.
    This is not merely a case of forgetfulness. At this point the whole underlying complication of the system begins to defeat an easy solution.
    "If I buy an impression from a site, and I’m using a third-party ad server, essentially what I feel I’m buying is the hole in the page," says Greg Smith, director of strategic services at Darwin Digital, Saatchi & Saatchi’s interactive spinoff. 
   "If my ad doesn’t get served before the user leaves the page, the publisher may say, ‘Well, I leased that page for you so if your ad didn’t get served, it’s not my fault.’ 
   "At the same time, the third-party server may say I served it within three seconds, that’s a reasonable amount of time. It’s not my fault.’
    "So I have to sit there and think of my clients’ interest. They paid for the ad and they didn’t get anything. Where should I put the blame?"
    We’ll look into the answer to this question and what might be done about the whole situation in future articles in this series.
    It’s not a pretty picture so far. The impression is such an elusive thing that many web sites prefer not to get pinned down on the issue at all, according to some media buyers.
   Nancy Ravell, interactive media coordinator for Fry Hammond Barr Advertising in Orlando, says that she often finds web site sales staff don’t really know how many impressions they can deliver, and they don’t seem to want to know.
    "So they give me the site traffic and want no guaranteed number of impressions in the contract. That's like saying, ‘Give me your money and you'll get whatever I decide to give you.’ I have had to be very strong in order to get a decent deal for my clients."
    The best web sites, says Leslie Laredo, are learning to automatically over-deliver to advertisers—which is one of the fastest and easiest ways to avoid problems. The inventory is there anyway, after all, since there isn’t a web site in the world that’s sold out.
    But this isn’t happening often enough for media departments, too many of which are finding themselves tied up in discrepancy resolution to an unprecedented extent.
    "I’d say that at least 10 percent of our time, and probably a good deal more, is being spent trying to resolve discrepancies, trying to recover impressions," says Greg Smith.
    "That’s a lot of money being wasted."
    Smith, who is one of the forces behind the New Media Consortium, a new industry group pushing for online standards, notes that the commonly accepted 20 percent deviation is when you think about it pretty ridiculous.
     "As one of our members recently put it, ‘What if I more or less paid our bill, plus or minus 10 or 20 percent?’"

-Jeremy Schlosberg is the senior editor for new media.

                     Cover Page | Contact Us