I wrote this article in April 1997, in response to problems I perceived in existing advertising networks.Ad Targetting in 1997-98

written April 13, 1997 by Mark J. Welch

Over the past year, I’ve watched dozens of “ad networks” come and go, and perhaps the most surprising thing is the amount of raw, blind copying.

The Commonwealth Network announced a model whereby it promised to pay a fixed amount for every ad impression; a half-dozen companies leapt in with similar promises. All have failed (including the Commonwealth Network, which resorted to lies and simple refusal to pay amounts owed to stay afloat).

Internet LinkExchange unveiled a “banner exchange” model that was copied by more than a dozen companies. Although LinkExchange survives essentially unchanged today, most of the “copycats” disappeared, unable to pay for the CPUs and bandwidth required to administer even a simple banner exchange network.

And dozens of “advertising brokers” and “ad reps” have lurked on the sidelines, seeking to cash in by linking wary advertisers with hungry web publishers.

It’s Time to Add Value.

Every ad network already recognizes that it can “add value” to raw traffic in “banner advertisements.” But so far, none has effectively segmented the traffic, and few have even tried. Most “established” vendors, like the Commonwealth Network and LinkExchange, are crippled by their early adoption of essentially useless category systems. LinkExchange compounded its problem by assigning a single category to an entire web site, rather than separately registering each page. Every ad network has also been burdened by the staff time required to “police” affiliates who assign incorrect categories or use prohibited content or techniques to increase traffic or “impressions.”

A few companies have sought to “add value” to the “banner exchange” and “ad network” concepts. For example:

• LinkTrader had created a model whereby a web site accrued its impressions over time, but could “save up” traffic in order to implement a larger campaign over a short time, or to shift all accrued impressions to specific days of the week. LinkTrader also allowed each advertising “campaign” to use a different ad banner and link URL. Unfortunately, on the day after I wrote this article, LinkTrader abandoned those procedures, instead opting to automatically trigger the use of all impressions daily (removing all timing control from the web publisher).
• SmartClicks (still in “beta” testing) designed software that allows each web page to be separately categorized; the software tracks the performance of each banner advertisement (based on the ratio of displays to click-throughs) in each category. An advertiser can then designate specific categories for their advertisement, or can instruct that SmartClicks automatically display ads more frequently in categories that it has “learned” are more effective. My own experience, after two weeks of allowing random rotation of ads, was the discovery that my ad was performing well in an unexpected category; on reflection, I understood why, but I would never have selected that category without first seeing its success.
• LinkExchange currently allow advertisers (but not its unpaid members) to designate specific categories for paid advertising, but only from a very short list of broad categories, and with no assurance that the selected category actually applies to a specific web page where the ad is being displayed.

There are some major flaws in the existing models used by all ad networks and exchange programs.

• Categories: No vendor has yet adopted an acceptable system of categories. For example, I struggled to find the correct SmartClicks category for my estate-planning web site, and finally selected “Stock Market/Financial” as the best of a bad lot. In my opinion, a vendor must devote substantial energy to creating, maintaining, adapting, and enforcing a system of categories for web pages. Each page must be separately registered, since many web sites include a variety of web pages often with unrelated content. And the vendor must adopt a system to verify the proper assignment of categories. For larger sites, this will entail actual review of the site or page and confirmation that the correct category or categories have been assigned. Ideally, the category system must allow for dynamic extension, with sub-categories added to distinguish sites and pages whenever possible. And if the ad-rotation software tracks the effectiveness of each ad in each category and on each page, the software should automatically detect many incorrectly categorized web pages based on discrepancies in click-through rates. Each vendor should consider whether to implement one system of categories, or whether it makes more sense to implement multiple category systems: for example, utilizing SIC codes, yellow pages categories, or SRDS categories in addition to other category systems. Finally, the ad network must address the issue of content changes at a web site: some system must be created to insure that new content, or gradual changes to content, are reflected in the category assigned to the specific page. A system must also be adopted for web sites with “dynamic content” (such as newspapers with new articles each day, or sports sites whose primary sports theme shifts with the seasons).
• Dynamic Display of Ads: Once categories are in place, the network must allow advertisers to select specific categories for ads to appear in, and the network should automatically adjust the display of ads within the sub-categories and specific sites within that category, to optimize the performance of the network (e.g. increase click-through rates). However, the advertiser and web publisher must each retain a “veto” power to prevent the display of certain advertisements on certain web pages or categories.
• Dynamic Pricing: Currently, most banner exchange networks and paid ad networks give web sites the same benefit in exchange for each ad display, but most vendors charge a premium to paid advertisers who want to target specific categories within the network. Strangely, the premium is the same for each category chosen. But as the AdBot auction in early April demonstrated, there is greater demand, and a premium price can be commended, for web sites in certain categories. Thus, I believe that each ad network should dynamically adjust the pricing for each category based on demand and “market forces.” In turn, the network should adjust the benefit offered to members (otherwise, web sites in more desirable categories will leave the network to collect their own premiums). While some ad networks and ad representatives do share the premium with web publishers, none of the “banner exchange” networks currently does so. Instead of offering a fixed ratio for banner exchanges, each network should dynamically adjust the “value” of each banner display. This could be done by use of credits — for example, each adview in a low-performing category might be purchased for one credit and sold for two, while adviews in better categories might be purchased for five credits and sold for ten — or the adviews could be assigned “wholesale” and “retail” prices.
• Other Targetting: Variations in the “performance” of each ad are not based solely on the subject or “category” assigned to each web page or site. Other factors also apply, and the ad network of the future must assign values accordingly. The domain type and geographic locale of the visitor (to the extent it can be determined), the time of day when the visit is made, the day of the week, and the operating system or browser used may also command a premium or discount.
• Registration: Thus far, no ad network has made effective use of “user registration” (e.g. gathering user demographic information to be used to further target advertising). However, I anticipate that some networks will begin adding select demographic data by carefully testing user response and addressing privacy concerns. For example, visitors may be willing to disclose some very broad pieces of information (zip code, sex, age, marital status).
• Campaign Control: Currently, most ad networks provide little or no control to the advertiser (either paid or exchange) regarding the execution of an ad campaign. LinkExchange gives affiliates no control over the timing or targetting of exchange ad displays, although paid advertisers can designate specific subject categories (and perhaps times and days); this might offend some affiliates if they perceive that they receive only the least desirable adviews. At one time, AdXchange admitted that most of its paid and exchange adviews allocated each day ran out before 5 p.m., so that only “house” ads appeared in the evening. LinkTrader allows an affiliate to “hoard” accrued ads and release them in a single campaign, but there is a lag of several hours from the “order” until ads begin to rotate. Most ad networks require staff time to manually change ad banners and the link targets for the banners. And currently, all ad networks require direct staff time to accept and process advertising orders.
• Portfolio Management: In addition to providing “campaign control” for the advertiser, an ad network should provide a convenient way for web publishers to manage a portfolio of web pages or sites. Since effective targetting requires that each page be separately categorized, and a web publisher may have hundreds or even thousands of web pages to manage, it is essential that “portfolio management” be made as simple as possible to encourage accurate and current registration of each page. One alternative to registering each page separately could be to provide a mechanism for the web publisher to use a single ad code, with parameters identifying the subject or category of the page; perhaps the X-URL could then be used to identify the unique page, or the web publisher could assign her own unique codes to each page.

Business Plan: Each ad network must put together a detailed business plan that identifies key issues involved in operating an ad network. The following elements must be discussed in any business plan for an ad network:

• Startup costs: the costs (for staff; for hardware, software and bandwidth acquisition; for customization and setup of the ad network, including customization (or development) of the ad network software, and integration with other modules; for publicity to entice affiliates; for promotion of the network to advertisers; and for maintenance through the initial startup period and until ad revenues are received).
• Transaction costs: The ad network must identify the “cost” of each transaction involved in the operation of the network. These include the cost to process each adview request and to serve each ad banner (in CPU time and bandwidth); the cost to acquire each new affiliate; the cost to review and categorize each new web page added into the network; the cost to “support” affiliates; the cost of soliciting advertisers; and the cost of acquring and processing each order for advertising. It is essential to recognize these transaction costs in setting “minimum” traffic levels for participation (for example, stating that only web sites with traffic of 1,000 impressions per week are desired, and only web pages with traffic above 100 impressions per month; and setting a minimum advertising order at 1,000 or 10,000 impressions). The business plan should identify ways to reduce these transaction costs through the use of online forms, CGI scripts, and automated credit-card processing for advertising orders.
• Profit: The business plan should identify the source(s) of profit within the ad network. Presumably, most networks will profit through the sales of paid advertising, compensated per impression or per click-through. If web affiliates will be compensated with exchange’ impressions, the business plan should reflect the number of impressions actually available for resale. The business plan should also address the issue of the disposition of “undesirable” impressions (e.g., adviews that are predicted to generate very low click-throughs or which are not desired by paying advertisers). If the network will seek to utilize all available “desirable” adviews for paying advertisers, and will allocate “less desirable” adviews to unpaid members, the business plan should address the issue of whether affiliates will accept this (or whether dynamic pricing’ may be used to alter the overall patterns of traffic and desirability of certain adviews). The business plan should also address any anticipated revenue sources other than advertising sales (for example, sale of demographic data or advertising effectiveness information).
• Comparison: The business plan should specifically examine existing competitors (including Petry, LinkExchange, LinkTrader, and SmartClicks), and should address how changes by those vendors might affect the new network’s business (for example, if competing vendors added new capabilities such as targetting or “campaign control,” how would that affect the business projections?). A key issue is an analysis of the total available “banner advertising inventory” on the Web, and its allocation to various market players. (Over the past year, many web sites have registered with a dozen or more ad networks, shifting sites and pages from one network to another several times.)
• Cheating: Unfortunately, any business plan must address the issue of web publishers who act unethically or fraudulently to artificially inflate the benefits received from participation in the network.

Comments are closed