How Web Ad Attribution is Gamed

Just about every media buyer would agree that above-the-fold ads are more valuable than those at the bottom of the page. But often those ads on the bottom, which are far less likely to be seen by people, are priced higher. This a huge problem created by improper usage of algorithms and misaligned incentives in a broken Web ad attribution system.

The problem stems from the fact that the entity to serve the last ad before a conversion is given full credit for that conversion. That means the last ad on a page, even if it is tiny and not viewable, receives full credit because it is the last one to load. This is a complete waste of advertiser dollars — and a risk to the exciting developments in programmatic ad buying.

The mis-attribution is tantamount to what happened in the bond market, prior to the economic collapse. Towers of B-minus paper were packaged together and rated as AAA by the rating agencies. That house of cards eventually came crashing down, and so might this one if credit is not assigned to real value creation, rather than to faulty attribution tricks.

The last-ad attribution system has created perverse incentives for a raft of players in the industry. The demand-side platforms, trading desks and ad networks that game the system most successfully are deemed the winners — and the best steward of advertiser funds. In such a system, an ad in a high-quality site like Forbes, with perfect contextual relevancy for a number of advertisers, is deemed no more valuable than a sleazy site with the same cookie and untold number of ads on the page, many of which can’t be seen.

Algorithms are not the problem; neither is programmatic buying, or exchanges. The problem lies with how these assets are used. They are like nuclear power, which can be used for good or evil.

The real source of the problem lies with media buyers, who, with some exceptions, often concentrate solely on price. In this cost-per-action mindset, premium inventory like Forbes and Time is usually automatically at a disadvantage in algorithmic buying because creating editorial quality is treated by the system as an added tax. This means that the selection pool narrows to shady long tail sites, gaming sites with tons of ads, and other dubious vehicles. The only portions of premium sites that might fit the bill are the bottoms of the pages, where tiny, below the fold ads live. In effect, the algorithm has determined that the top of Yahoo’s home page is less valuable than the very bottom, where three tiny ads reside.

To make matters worse, the entity that has the majority of the budget will always appear to outperform competitors on the plan that have smaller portions of the budget. This is the case even if the competitors in reality perform more effectively.

As a campaign runs over time, algorithms identify pockets of inventory that are the most desirable based on the last view attribution logic. Because they are competing for the inventory in an auction, the competition drives up the price of these pockets. As a result, whoever is willing to bid the highest for these pockets has the advantage. Clearly, the entity with the largest budget can bid the highest, perhaps even higher than it should, to win. But since low CPAs require the average cost of media to be low, these high priced pockets have to then be balanced with really cheap inventory, much of which could be useless. It creates a lot of waste.

The more effective way to run the campaign is to make a determination of who is going to be on it, and then give an equal portion of budget to each entity. Each will run a portion of the campaign and explore a wide pool of inventory initially. The competition among them will have little to no effect on the cost of inventory at this stage because the pool is so wide. It’s more like an ocean. As the campaign runs, say over the course of one to two weeks, depending upon the budget, a clear winner, or perhaps a couple of winners ought to emerge. At this point, the entire budget should be given to this winner for the duration of the campaign, so that it eliminates the competition for the pool of inventory that the algorithms have by now narrowed down to. By doing this, the advertiser does not compete against itself (through multiple bidders) for the same inventory. The same inventory can be bought for less.

Unfortunately, the process that prevails currently often gives a “look” to multiple entities, often at the advertiser’s suggestion. However, that “look” varies in magnitude, and each entity does not consistently receive the same portion of the budget. The system becomes set up to be gamed from the beginning, often incorrectly rewarding value creation.

Sunil Sharma is an independent strategy consultant in digital media and an expert in programmatic buying and trading. He was until recently director of trading at Adnetik, the independent trading desk. Follow him @sunilinboston.
Digiday Top Stories