NONE: ONLINE-ADS>> DEBATE: 4/17/98 - measuring impressions, digest # 05

ONLINE-ADS>> DEBATE: 4/17/98 - measuring impressions, digest # 05
Fri, 17 Apr 1998 16:37:03 -0500 (CDT)

How should our industry measure an impression?
4/17/98, digest # 05

This is special "debate" mailing from the Online Advertising
Discussion List. Anyone on the list is free to send in a response
to be considered for inclusion.

You will receive this mailing in addition to your normal
Online Ads posts/digests/reports.

web site supporting list:
moderator: Richard Hoy
posting address:

subscribe/unsubscribe instructions are at the end of this message

This week's Online Advertising Discussion List sponsors:
ValueClick and ICONOCAST

Got unsold ad Inventory??? We've got advertisers!!!
ValueClick is your hands-off solution


"The quality of ICONOCAST, the brevity and frequency is absolutely
spot on!" - Marc Phillips, Director,
E-mail with "subscribe iconocast" in SUBJECT


Panelists' Responses:

1.) Tom Shields - NetGravity, Inc
2.) David Zinman - AdKnowledge, Inc.
3.) Keith Pieper - MatchLogic
4.) Jim Meskauskas - Hawk Media

Online Ads List Responses:

5.) Joe Savelberg <> - caching banners and actual numbers
6.) Jay Luker <> - Robots, Spiders and Honesty


Date: Wed, 15 Apr 1998 20:08:20 -0700
From: Tom Shields <>

Yesterday I made a case for one counting standard. And in my
first posting, I made reference to different "families" of
impression measurement that are in use today. In this
posting, I'll try to argue that the "ad download" family of
measurement is the right one to standardize on at this time.
In a future posting, I'll expand more on exactly what
method in the ad download family I favor.

I define a counting method as in the "ad insertion" family
if it counts at the time the page is delivered. Usually
this happens when a page is dynamically constructed, and an
ad inserted, before it is delivered to a browser. However,
this method also includes "static" ad insertion, where an ad
is simply hard-coded into a page, and then the page is
requested by a browser. This method is used by a large
number of sites, because it is easy to implement, and
relatively accurate.

The main problem with the "ad insertion" family of methods
is that they don't work for third party ad serving networks.
All these methods count when the browser first requests the
page, and an ad network never sees this request. If one of
the "ad insertion" family of methods were to become the
standard, then all of the current ad delivery networks would
be out in the cold. I consider this to be a "showstopper",
and it effectively eliminates all of these methods from
serious consideration as a universal standard.

I define the "ad view" method as client-side counting and
reporting. In this case, the client needs to have the
intelligence to report back after an ad is requested and
displayed that it has succeeded. Several companies have
introduced new ad media types that provide this capability -
to my knowledge, all require either Java or a plugin to
accomplish this. These methods are potentially the most
accurate, and one from this family may eventually become the
"best" way to measure impressions.

However, at this time, most clients do not have built-in
technology to report client-side behavior. And, all of the
media types that do provide this capability are based on
proprietary technology, which could make the industry
beholden to one company - I do not believe this is good for
the industry as a whole. Finally, none of these media types
have established a major market presence. These reasons
combine to convince me that, at least for the next year or
two, no measurement from the "ad view" family will suffice
as a standard.

I define the "ad download" family as measuring when the
browser requests some part of the ad, as opposed to the
page. Most ads today are simple GIF images, and many large
advertising networks have sprung up based on the fact that
the images are requested separately from the pages, and
therefore can be requested from another site. In my
definition, this family of counting methods includes
measuring any kind of media that can be directly requested
by the browser, separately from the main page. For example,
this includes ads requested by APPLET, EMBED, or SCRIPT
tags, and ads consisting of HTML served into frames,

Methods in this family can be used by sites as well as
networks, although it may require a bit more work than the
"ad insertion" method they are more likely to use. And,
some of these methods may require additional server
resources and bandwidth, if the site does not already defeat

An argument can be made that some ads do not have a
"download" component. Specifically, ads that are just
clickable text, or HTML tables and forms, may not have an
image or other media that is downloaded separately from the
page. These ads are not countable by this mechanism. There
are several possible responses to this: embed a tiny GIF
consisting of a clear dot for counting purposes (the German
solution); fall back to counting page impressions for these
types of ads only; or deliver these ads using a browser
based download mechanism that involves IFRAME or JavaScript.
I believe that this type of ad makes up such a small
percentage of the total advertising on the internet today
that despite this problem, download methods remain the most
viable alternative.

By this line of argument, "ad insertions" are not tenable
because networks cannot measure that way. "Ad views" are
not available yet because clients don't support them. This
really only leaves "ad downloads". Note that without
specifying the method exactly, measurements within one
family can still vary by 50-600%, because of caching and
other factors, so our problem is still nowhere near solved.
Next time, I will talk about specific download methods.

-ts- Voice:(650) 655-4774
Tom Shields, VP, CTO Fax:(650) 655-4776
NetGravity, Inc 1700 S Amphlett Blvd, Ste 350, San Mateo, CA 94402-2715

-----------------------------POST NUMBER 2

Date: Thu, 16 Apr 1998 19:02:23 -0700
From: David Zinman <>

Add the fine name of Laura Mitrovich to the list of agency
media professionals endorsing the goal of measuring an
impression as close to the complete delivery of an ad as
possible! Can we say this case is closed yet?

In previous posts, I argued that we should count the very
last action taken by the server when sending an ad graphic
to a browser. I also explained the technical distinction
between counting an HTML page transfer (pageview), which
happens first, and an ad graphic transfer (adview), which
happens last.

One of the only complaints one might hear about counting
adviews is that ad graphics can be cached, causing counts to
be lower. This post covers some of the caching issues around
counting pageviews and adviews. I will make the case that
nearly all of the caching that affects adviews affects
pageviews as well. In the one area where caching affects
adviews and not pageviews, it can be circumvented easily.
Let me illustrate by explaining some of the ways caching can

There are two kinds of caching: browser and proxy caching.
Both affect counting in the same way. A server can count an
event only if it is notified that an event occurs. With
caching, however, a browser can display a page or ad WITHOUT
EVER CONTACTING the server. For publishers, this can result
in lost inventory -- for advertisers, inaccurate results.

Why does caching occur? Some instances of when caching of
pages or ads can occur include when: (1) a viewer hits the
back or forward button and their browser gets the page from
cache; (2) a viewer clicks on a link and their browser gets
the page from cache; (3) a viewer goes to a page where an
item has already been cached from a previous page (this
could include an ad graphic, the contents of a frame, or
other files).

The first two cases affect counts for pages or ads in the
same way. The third case is more specific to ads, but can be
circumvented by a decent ad management system. This is done
by a process known alternatively as cache busting, cache
confusing, or cache counting. Basically, it boils down to
making the browser believe that each ad graphic is different
from the ones already in cache, and forcing a quick
notification of the server to take place. This can happen in
a variety of ways, and it is not always necessary to force
the browser to reload an ad, depending upon how the cache
defeating system is set up. The result is the same, however:
the most accurate way for servers to count impressions!

In my next post, I'll offer tips(!): some questions that
marketers and agencies should be asking when they place web

-----------------------------POST NUMBER 3

Date: Fri, 17 Apr 1998 11:49:38 -0600
From: Keith Pieper <>

I believe this response fully encompasses what we have all
discussed thus far, and objections made to our technologies
and methodologies. I hope it's not too long for everyone.

Again, MatchLogic believes that an ad impression should be
measured when a user's browser resolves and displays an
advertisement on a user's screen.

First and foremost, we are striving for the same ideal all
others do - optimal accuracy and optimal comparability. In
addition, MatchLogic believes the issues of accuracy and
compatibility are integral with issues of management and
consistency, ultimately enabling optimal management of the
user experience.

We recognize that there will never be anything that is 100%
absolute, because 100% is ideal but not practical (i.e.
there will always be images off, stop during loads, robots,
GFX off, scroll-always, etc.) Our goal is to provide optimal
measurement - as close to 100% of eyeballs seeing an ad as

However, accurate and consistent counting simply enable
optimal management of the impression opportunity. In other
words, proper counting and measurement are fundamental - but
the larger issue is how you manage that impression, from
targeting, to delivery, and ultimately response. And we
believe our technology (namely TrueCount) and counting
methodology is the best application of any existing
technology and methodology that brings the industry optimal
accuracy, consistency, and comparability to ultimately
effectively manage the relationship with each individual end
user, regardless of where they are on the internet.

Cacheing An industry disagreement surrounds cached ads. Some
simply count impressions as ads served from an "ad server" -
this ignores cached ads. The simple fact of ignoring cache
is inaccurate counting. Our study included over 200 web
sites and millions of impressions, concluded that cache
accounts for an additional 176% more impressions, reaching
as high as 674% on some sites.

Others argue they refuse to pay for cached ads, as the
cached ads are already accounted for in reports, and
including these cached impressions would reduce click rates
- without including cached impressions, the industry is
getting an inaccurate measurement and a false sense of

The issue of discounted click-rates is important if you
believe click-throughs are the ultimate benchmark for
effectiveness. However, relying on clicks is another false
sense of effectiveness reality. The ultimate benchmark is
the relationship with the end user - and without managing
cache, you can not manage that relationship effectively.

Even more, others argue that magazines don't charge for
pass-along readership, and therefore neither should the
internet charge for cached impressions. First, other mediums
are not as measurable and therefore accountable as the
internet; second to achieve the industry's common goal of
optimal accuracy, cached impressions must be counted. Again,
managing cached impressions enables an optimal relationship
with the end user - which is just as applicable to web
publishers as it is to advertisers.

TrueCount is our patent pending cache counting technology.
Leveraging TrueCount, an advertiser or site can get an
accurate count of their ads - whether served from ad server,
proxy cache or browser cache - allowing them to effectively
manage the end user relationship. TrueCount measures things
such as the user hitting the back button (the ad is loaded).
TrueCount leaves no stone left unturned.

Our Counting Methodology (This should answer Mike's (from
PhotoLabel) questions. Another industry argument exists over
whether to count "images requested" (ad download-current
industry standard) or "images transferred" (ad views-when
browser finishes loading image). Our methodology is, of
course, "images transferred". Counting based on "images
requested" is not ideal for two reasons - one, it fails to
account for delivery of an impression, and delivery is what
everyone wants (accuracy); second, it fails to account for
cache and being subject to local site cache policy.

"Image requests" are first routed through browser cache,
then proxy cache, and finally to ad server, in that order.
If the image request finds the image in either cache, the ad
server never sees (or counts) the ad request (impression).

By counting "images served", we are counting not the
request, but the actual delivery - whether it be from
browser or proxy cache, or ad server. And this is why so
many industry leaders say that counting based on "images
served" or "transferred" is impractical - because the vast
majority of ad servers are unable to count cache, therefore
making their counts inaccurate, and leading to a false sense
of market understanding.

Finally, to increase our accuracy of counting "images
served" and come closer to optimal accuracy by defeating
more counting variables, we only "count" and ad when the
final data packet of the ad image is sent (i.e. a complete
file has been sent). This addresses exactly what David
Zinman of AdKnowledge suggests - the last data packet sent
IS the last action taken by the ad server.

This is the closest the industry can come to having the
browser inform the server that the ad was actually viewed.
As long as the browser keeps accepting data packets (the
user doesn't hit stop, reload, GFX off or link-over) and we
keep sending them, we can assume the transfer is being
completed, but is not 100% complete until the final data
packet has been sent, which is when we take the count. This
methodology overcomes problems of partial ad loads, a page
loads but not the ad, when "images are off,"

Simply, to reflect as high of degree of accuracy as
possible, you can't measure completed transfers or "final
packet sent" without measuring cache, because some ads are
served from cache. And without counting cache, you fail to
be accurate. And failing to be optimally accurate as
possible will prohibit consistency. Consistency enables
comparability. And that includes comparability with other
mediums. And ultimately this leads to effectively managing
the end user relationship. And we believe our solution has
optimally solved issues of accurate and consistent counting
and delivery. That places us in a position to focus on the
real issue at hand - optimizing the end user experience from
impression through final response.


Keith Pieper
Idea Architect, MatchLogic
303-665-4007 x232

-----------------------------POST NUMBER 4

Date: Fri, 17 Apr 1998 09:24:15 -0700
From: Jim Meskauskas <>

On Wed, 15 Apr 1998 "Hespos, Tom" <> writes:

> How will our definition deal with the notion of spider and
>crawler hits? I have seen logfiles for sites where a
>significant number of ad requests were logged by IP
>addresses of known spider and robots. I'm not talking about
>1-2 percent of traffic. I'm talking 10 to 30 percent.
>Should the ads requested (and transferred) by spiders,
>crawlers and robots be counted as ad views? Or should sites
>and third-party servers filter out known crawlers using IP
>address filtering?
>For one, I think that advertisers should not have to pay for
>it everytime an Infoseek spider comes through to re-crawl a
>site, for example.

Ad banners served to robots and spiders DO NOT constitute an ad
view, and an advertiser should not be expected to pay for the impression
generated. I think "impression" implies a certain amount of intentional
sentience. The word itself suggests as much. My ad makes an IMPRESSION
on a viewer. An advertiser isn't after robots and spiders; an
advertiser is after humans. If the TV is left on in a room full of
cats, my ads are not generating any impressions that I'm interested in.
The only way to deal with this, as Tom Hespos suggests, is to filter out
known crawlers. This shifts much of the onus onto the sites and could
negatively affect the bottom line, but it the long run it will
strengthen the integrity and increase the value of that site's

Jim Meskauskas

-----------------------------POST NUMBER 5

Date: Thu, 16 Apr 1998 23:29:39 +0200
From: Joe Savelberg <>

Sean Pfister <> wrote:
>If there were an estimate for graphics off, media planners
>might want to weight down the scheduled impressions. Maybe
>with a caching estimate, publishers can weight the delivery
>back up. (and researchers will continue to have job security
> :) )

If you want to get better estimates about caching, you could
place a non-existing image next to your banners and display
it at 1 x 1 pixels.

>From my understanding, non-existing files cannot be cached,
but they still show up in your log files.

<IMG SRC="bannercode.gif" HEIGHT=60 WIDTH=468><IMG
SRC="nofile.gif" HEIGHT=1 WIDTH=1>

"nofile.gif" doesn't exist on your server, but each client
with images turned on will try to download that image, thus
generating an error in your log files. As proxy servers
cache only existing files (such as your banner), they
shouldn't cache the error response. Analysing the hits for
the (almost invisible, 1 pixel) non-existing file will give
you the actual number of times that your page was shown with
images turned on. Comparing this number to the number of
impressions for your banner will give you an estimate about
the number of cached requests.

If you also have the client/browser information in your log
files, then you could write a script to check the number of
different browsers used per IP number.

For example: You've got 10 visits from the same IP number,
but there were 4 different browser versions used. You can
conclude that at least 4 different users accessed your page
through a single IP number (e.g. a proxy server). As there
are so many flavours of browser versions around, you should
get some interesting data... verify all the request coming
from AOL's proxy servers and you'll be surprised how many
different browser versions are used. This gives you a better
idea about how many people access your site through a proxy
server. But it is still just an estimate. (Corporate proxy
servers won't give you a big difference in numbers as
companies probably try to standardize on the same browser
version on all computers.)

By combining the log data for the non-existing gif, the
number of banner impressions, the number of page views and
the browser data, you should be able to get a good estimate
of the number of visitors to your site.

Just my 2 cents worth,


| jochen savelberg |             EasyScopes - <> |
|  webmaster_at_night |     The fastest and easiest index to your horoscopes |
| | Daily, Weekly, Monthly... horoscopes grouped by sign |
| New Year's Eve 1999 in Times Square - Join the Party (I will!):         |
|                |

-----------------------------POST NUMBER 6

Date: Thu, 16 Apr 1998 17:54:51 -0400 From: Jay Luker <>

Tom Hespos wrote:

> How will our definition deal with the notion of spider and > crawler hits? I have seen logfiles for sites where a > significant number of ad requests were logged by IP > addresses of known spider and robots. I'm not talking about > 1-2 percent of traffic. I'm talking 10 to 30 percent. > > Should the ads requested (and transferred) by spiders, > crawlers and robots be counted as ad views? Or should sites > and third-party servers filter out known crawlers using IP > address filtering? > > For one, I think that advertisers should not have to pay for > it everytime an Infoseek spider comes through to re-crawl a > site, for example.

Don't forget that robots and spiders will also show up on click-through reports, increasing CTR percentages just as much as ad requests. For over a year now Andover.Net has had a system of removing all traces of robot hits and clicks from our server log analyses, basically throwing out data coming from any IP address that registers more than 1000 pages. We felt that it was better to err on the side of over-delivery in fairness to our advertisers.

Unfortunately, this strategy also filters out a measurable number of ad clicks that were being generated as the robots indexed our pages and links. Once that magic CTR number dipped, you better believe advertisers started to holler. When we calmly explained that what we were doing actually benefited them, do you think they gave a rat's behind? You'd be surprised at the number of buyers who didn't understand what we were talking about. "Whadayamean, The Spiders are clicking on my ads?"

There's been talk of dropping the robots filter altogether, because accurate reporting doesn't seem to be top priority to buyers. They just want that big, shiny number that they can take to the boss everyday, and when the conversion rate at the end of the campaign doesn't match what they were expecting, they'll blame it on the audience and choose not to renew.

Anyway, to get around to the point I guess I'm trying to make, if a good, fair, honest standard is ever reached on how to measure traffic, clicks, etc., I think the biggest hurdle may be getting everyone to comprehend the long term benefits of implementing it. It's going to be painful. Publishers who adopt the standard may begin to hear the word "Ouch!" much more often around their offices. On the other side, publishers who choose to quietly continue their current methods of traffic measurement will look more attractive to potential advertisers who are only looking at each day's bottom line.

| Jay Luker | Creative Project Manager | Andover.Net | | |

======================================================================== ------------------------------------------------------------------------ This week's Online Advertising Discussion List sponsors: ValueClick and ICONOCAST

Got unsold ad Inventory??? We've got advertisers!!! ValueClick is your hands-off solution


"The quality of ICONOCAST, the brevity and frequency is absolutely spot on!" - Marc Phillips, Director, E-mail with "subscribe iconocast" in SUBJECT ------------------------------------------------------------------------ ========================================================================

SUBSCRIBE INSTRUCTIONS To subscribe to the Online Advertising Discussion List, Fill out the form at:

UNSUBSCRIBE INSTRUCTIONS You may leave the Online Advertising Discussion List at any time by sending an email message with the word UNSUBSCRIBE (just the word) in the body or content of your message to:

if you get the digest version, and

if you get the non-digest version.

Online Advertising Discussion List

a publication of The Tenagra Corporation


With an archive of more than 14,000 postings, since 1996 the Online Advertising Discussion List has been the Internet's leading forum focused on professional discussion of online advertising and online media buying and selling strategies, results, studies, tools, and media coverage. If you wish to join the discussion list, please use this link to sign up on the home page of the Online Advertising Discussion List.


Online Advertising Industry Leaders:

Local SEO with Video
Houston SEO
Austin Web Design

Add your company...

Local SEO with Video


Online Advertising Discussion List Archives: 2003 - Present
Online Advertising Discussion List Archives: 2001 - 2002
Online Advertising Discussion List Archives: 1999 - 2000
Online Advertising Discussion List Archives: 1996 - 1998

Online Advertising Home | Guidelines | Conferences | Testimonials | Contact Us | Sponsorship | Resources
Site Access and Use Policy | Privacy Policy

2323 Clear Lake City Blvd., Suite 180-139, Houston, TX 77062-8120
Phone: 281-480-6300
Copyright 1996-2007 The Online Advertising Discussion List, a division of ADASTRO Incorporated.
All Rights Reserved.