You should read this article.
After all, if you’re involved in any sort of online marketing, you probably spend time with online advertising networks and behavioral targeting providers. According to Deutsche Bank (Jan. 2005), spending on online advertising networks will make up 14 percent of all online advertising in 2005, a significant increase over the past year. Typically, such networks aggregate media from scores of individual publishers and sell impressions on a pay-for-performance or targeted CPM basis. Because the networks have access to a wide variety of relevant data, they have also been the earliest adopters of behavioral targeting.
Unfortunately, it is often difficult to peer “under the hoods” of major ad networks. Lacking visibility into how ad allocation decisions are made, many marketers similarly lack understanding of how to optimize their effectiveness on these networks. They feel trapped in the proverbial black box.
In reality, buying online advertising parallels the purchase of an airline ticket. When booking air fares, knowing some simple rules (e.g. weekend fares are generally cheaper, buying in advance facilitates lower fares, et cetera) can save you a bundle. Likewise, even rudimentary knowledge of how many networks allocate inventory can enhance your marketing success. Learn to game the system and you can beat the networks.
How things work
Despite their claims of sophisticated, proprietary technology, most aggregators utilize relatively simple yield management systems to allocate inventory. There are really only two things a marketer absolutely must understand about how things work. First, networks are constantly trying to balance supply and demand. In reality, since these networks lack perfect visibility against both dimensions, they actually are in a constant state of imbalance. Thus, when demand drops during summer months or supply constricts during the holiday season, both price and availability fluctuate. This is happening all the time. In addition, most networks allocate inventory according to expected value. Naturally, advertisers who are willing to pay more for any piece of inventory will be rewarded with greater volume.
In short, ad networks are constantly imbalanced and inventory is rewarded according to expected return. While these two truths may sound pretty basic, they can be incredibly useful to those marketers who understand how to use these rules to their own advantage.
Now, beat the system…
For those marketers who plan to work with a network and/or behavioral targeting provider, there are a number of ways to game the system. The list is long, but even a handful of tactics can dramatically improve your odds of success:
Set up a second, "sweeper" deal. Many networks and behavioral targeting providers are willing to price their inventory on a performance (cost-per-action, or CPA) basis. In these arrangements, advertisers specify a budget and the price they are willing to pay per conversion, and the network delivers results. In most cases, the bounties advertisers set are realistic and the networks are able to deliver the contracted volume. However, advertisers are missing a big opportunity. Since the network is constantly imbalanced, most advertisers who specify a fixed budget and realistic bounty fail to capitalize on supply spikes. In other words, if excess supply comes into the network, the typical advertiser cannot take advantage of softer pricing and higher volume availability.
By setting up a second deal structure, with a ridiculous, self-funding CPA and uncapped budget, opportunistic advertisers can take advantage of supply/demand imbalances. When extra inventory exists, the second, “sweeper” deal will generate conversions at an incredibly attractive ROI. When the network is balanced, those conversions will disappear but the more realistic, higher bounty primary deal will still convert. And while this is a smart tactic for performance deals, it is equally effective on CPM buys (for which an advertiser could contract for a fixed amount at a higher CPM, but then receive surplus at a much lower CPM).
Slow down, but never stop. If impressions are sold on a CPM basis, yield management is pretty easy for a network. The expected value of the inventory is simply the CPM price the advertiser is willing to pay. The explosion in performance pricing makes things far more complex. A network that is offering CPA pricing must calculate the effective yield of each impression by forecasting the probability that a particular impression will result in a conversion for the advertiser. Most networks utilize historical data to predict expected value -- assuming that how an ad converted in the past will indicate how it will perform in the present. Metaphorically, this historical data “primes the pump” on the ad serving decision engine. But since many networks have a 14- to 30-day look-back window on utilizing this historical data, any advertiser who completely turns off a CPA campaign is leaving their network with little to no relevant information around which to predict expected value. In the absence of such information, the prediction will be more conservative and the advertiser will receive fewer impressions and conversions than they would if more recent information was available.
As a result, CPA advertisers should consider shrinking their budgets rather than pulling their campaigns completely, which will facilitate stronger performance when advertisers want to ramp spend.
Move up the conversion chain. Since most yield management systems allocate CPA inventory according to expected value, and expected value is based on historical data, it’s important to know how to influence expected value. Highly considered, less frequent products are often at a disadvantage. While they carry a high bounty, the conversions occur so infrequently that the predictive power of the historical data is vastly diminished. Consider a cruise line that is willing to pay $100 per online booking. Since bookings are infrequent, a cruise line that has a one percent site conversion rate may want to consider paying a smaller bounty against a more frequent action. Assuming no conversion slippage, a $1 bounty per site visit equates to the same expected ROI as a $100 bounty per ultimate booking. Yet, to the yield management system, the additional conversion success generated from moving the bounty upstream in the purchase process vastly improves its statistical confidence and thus rewards this advertiser with more impressions.
The most sophisticated advertisers will often set up hybrid deal structures that reward different actions with different bounties, all in an effort to feed the yield management system with strong predictive data. The aforementioned cruise line could, for example, improve ROI and volume by setting up a hybrid deal that rewards $0.25 for every site visit and $50 for every booking (at a one percent site conversion rate, the cruise line now only pays $75, on average, per booking).
Experiment with impression (CPM) pricing. Many marketers are wedded to CPA pricing, when they could generate stronger performance by purchasing on a CPM basis. Since the expected value of CPM inventory is already known to a network, by agreeing to CPM pricing on highly targeted inventory, marketers can better manage volume of their targeted campaigns and potentially pocket a significant ROI improvement. In CPA pricing, the network will apply “black box” targeting to maximize its own yield. In CPM pricing, the advertiser may be able to employ its own targeting schemes to procure highly relevant impressions at a lower effective cost. The elimination of irrelevant advertising waste and the higher conversion rates on the purchased impressions may generate far better returns than a risk-free CPA deal.
In other words, any marketer that is employing primarily CPA deals should periodically test targeted CPM purchases to compare returns.
Add filters. Finally, if a targeted campaign is not generating strong ROI, advertisers should consider alternatives to cancellation. Typically, by applying additional filters, campaign performance can be vastly improved. Perhaps the easiest filter is simply to control cookie-level frequency; other levers include tightening time of day targeting, geo-targeting or gender targeting.
Rather than rejecting their targeting programs, advertisers should tighten their targeted segments. Waste will be reduced and performance will soar.
The good news: Most advertisers have no idea how ad networks allocate inventory. You are now the exception. Even with a little knowledge, you can tilt the odds in your favor.
Scott Howe is general manager for DRIVE Performance Media (DRIVEpm), an operating unit of aQuantive, Inc.