Google’s Ever-Evolving Quality Score

Before we delve into Google’s latest Quality Score changes, it’s worth noting that Google announced this week it’s terminating the Google-Yahoo agreement that would have resulted in Google ads being distributed to Yahoo, its sites, and (potentially) its partners.

While the original deal was rife with flaws, a deal that was fair to advertisers might have been possible if Google and Yahoo had deduped their advertiser base by display URL, which would have assured that advertisers and their affiliates were never bidding against themselves across the two networks. The last elements of the deal circulated before the pullout represented a partial solution; they called for an opt-out for Google advertisers who didn’t want to have ads syndicated. In reality, however, if most of a advertiser’s competitors stayed opted in, it would still have felt the impact of the changes within the Yahoo marketplace.

I hope Yahoo redoubles its efforts to improve Panama, particularly by opening up behavioral, textual, or graphical display inventory priced on a CPC (define) basis tied to prior search behavior. If consumer privacy issues can be addressed (and Yahoo’s users, after all, will experience better advertising when they opt in), this will be the single largest opportunity Yahoo has. Currently, the search engine has tons of poorly monetized inventory that gets sold on the RightMedia and BlueLithium platforms. Yahoo’s APT project is a step in the right direction, but it needs to be simple for search marketers to opt in within Panama and have API (define) access to control those behavioral ads.

Google’s Quality Score

A couple weeks ago, Google once again tweaked the Quality Score formula that controls what you get for what you pay (or how much you’ll have to pay to get what you desire). In the early days of AdRank, which operated in a way based purely on predicted CTR (define), there weren’t dozens of other factors involved in Google’s position calculation for a given bid. However, Google didn’t have a simple task, either. Normalizing the data correctly to put all ads on a level playing field to calculate the predicted effective CPM or AdRank was (and is) a non-trivial task.

Now that there are many more variables included in Google’s Quality Score (and therefore in the position calculation), Google has come full circle and taken a fresh look at how it calculates and normalizes existing CTR data to arrive at a more accurate predicted CTR. The initial announcement was a bit unclear, and many people took it to mean that Google had never normalized for position. But Google clarified the post to indicate that it emphasizes “that AdWords has always accounted for the influence of ad position on CTR and removed it from the Quality Score. This specific improvement updates this system to make it fresher and more accurate.” Google apparently made data from positions that were randomly assigned higher than the AdRank would have normally suggested the basis of the calculation. Now Google uses normalization tables and the CTR data at the current positions to calculate the true normalized score.

A more accurate Quality Score, and particularly an accurate predicted normalized CTR, is fair to everyone. You don’t want to be outbid by someone who doesn’t really deserve the spot. Coincidentally, improved predictions on CTR also make Google more money per thousand searches. By improperly calculating true predicted CTR, it misranked advertisers, costing it money in opportunity cost.

I don’t have any specific insight on how the algorithm was making suboptimal predictions regarding predicted CTR and, therefore, calculating an incorrect Quality Score. However, one significant thing I’ve noticed over the years is the brand domains of well-known companies never seemed to get the credit for their stickiness when they appeared lower down the page. Consumers seek out brands they know and trust. So the normalization table one would create using all advertisers wouldn’t hold true for strong brands and would be too kind to URLs that are considered a poor fit. Eventually reality will catch up to the predictions, but we all hope Google is getting better at predicting true relevance.

In addition to the supposed improvements in estimating CTR, the new system factors in the Quality Score calculation’s freshness. This is great, because before a poorly performing ad or listing could be dragged down by its history. I’m hopeful that such a freshness factor will allow recent history to outweigh past poor performance.

Google’s systems are still a bit opaque for my liking, but anything that improves the auction model’s relevance and accuracy is good for the ecosystem as a whole.

Meet Kevin at Search Engine Strategies Chicago December 8-12 at the Chicago Hilton. The only major search marketing conference and expo in the Midwest will be packed with 60-plus sessions, multiple keynotes and Orion Strategy sessions, exhibitors, networking events, and more.

Related reading