Moe, Larry, Curly & Google’s Quality Score Calculation

In this Article, Moe, Larry, & Curly will try to shed some light on the many different components affecting the Google AdWords Quality Score.

As the Pay Per Click landscape becomes much more competitive, and the launch of Yahoo’s new Panama System, one of the most important things an advertiser has to take into account in maximizing performance is the Quality Score. Given that Google weights this factor so heavily that it can drastically alter a campaign’s profitability and that Yahoo is going to create a variation of the Quality Score (which I shall refer to as ‘Q’ score) it has become a very important factor in the PPC marketing game. The Q Score, as defined by Google is based upon: a KeyWord’s CTR, ad copy relevance, keyword, relevance, and landing page relevance. The price the advertiser pays and the average position of the ad on a page is based on the Max CPC, Q score, and its relation to other advertisers.

We will start with a general definition of a Quality Score and then explain the how the different Quality Score’s affect the different parts of a Google AdWords campaign. The pricing process starts with an Google Ad Rank. This is a function of the Max CPC and the Quality Score. Here is an example of how that works:

Advertiser,Quality Score,Max CPC,Ad Rank,Actual CPC
Moe,3.50,0.75,2.6,0.72
Larry,2.00,1.25,2.5,1.06
Curly,1.75,1.20,2.4,0.44
Shemp,1.50,0.50,0.8,0.01

AD RANK
Google’s Adwords uses a variant of the Vickery Auction called a Generalised Second Price Auction. Google starts with an Ad Rank calculation as above. As you can see Moe, the first of The Three Stooges, came out with an Ad Rank of 2.6. Larry, always being the second stooge, comes in with an Ad Rank of 2.5. As one can see, Moe was willing to pay 0.75 Max CPC while Larry was willing to pay a whopping $1.25 Max CPC.

ACTUAL CPC CALCULATION
The Actual CPC of the #1 position is determined by calculating Ad Rank of the NEXT highest bidder divided by the Quality Score of the Winning Bidder + .01. So in this case it would be Larry’s Ad Rank divided by Moe’s Q Score or 2.5/3.5 + .01 = 0.72. This analysis can be also be applied to Larry, Curly, and the most often forgotten stooge – Shemp. In order for Larry to bid on the #1 position, Larry would have to increase his bid to Moe’s Ad Rank divided by Larry’s current Maximum CPC + .01 or 2.6 / 1.25 + .01 or $2.11. Since he lacks a high enough Quality score and is unable to raise his Max CPC, he still has to deal with being the number two Stooge.

DETERMINANTS OF QUALITY SCORE
The determinants of a Quality Score include Landing Page Quality, Ad Copy Relevancy, CTR on Google, as well as the CTR History based on an Exponential Moving Average. An exponential moving average places more emphasis on recent history or CTR data than long term CTR data which means that your Quality Score will be impacted much more quickly than it would have been if a standard Simple Moving Average was used.

MULTIPLE QUALITY SCORES
Many people do not know that there are MULTIPLE quality scores that effect different attributes of a Google AdWords campaign or how they are applied to your Google AdWords Account. These variations include:

1. Google Search Network: A Quality Score is determined for the Google Search Network as described above.
2. Ad Group Quality Score: The Ad Group Quality Score is determined by an average of individual keyword Quality Scores.
3. Minimum Bid Quality Score: The Minimum Bid required to activate a keyword is determined by relevancy factors described above as well as the Landing Page Quality.
4. Account Quality Score: The Account Quality Score is determined by a combination of Keyword Quality Scores, Ad Group Quality Scores, Account Spend, and Account Quality Score History.
5. Quality Scores are also computed for Ad Copy as well as Landing Pages. The analysis, optimization, and structure of an AdWords Campaign and Website Development must take into account all of these factors.

It is the COMBINATION of these quality scores and APPLICATION of these quality scores that affect how a given campaign will perform and pose various challenges for the Three Stooges as well as any seasoned PPC marketer.

BAYESIAN PROBABILITY AND INCOMPLETE INFORMATION
As a former statistical arbitrage trader, I had to make decisions using incomplete information or using Bayesian Probability or Inference Theory. Bayesian probability theory has been used in Options Trading and even Poker. The biggest problem lies in the fact that an advertiser has incomplete information to make decisions. The advertiser has his Max CPC, Actual CPC, CTR, Average Ad Position, but no actual Quality Score (although it is rumored that Google will provide this number in the near future). The only knowledge that an advertiser does have is their position relative to other advertisers, but no information regarding their CTR, Max CPC, Actual CPC, or Quality Score. In a future blog, I will attempt to explain how to use Bayesian Probability Theory in order to make smarter bidding decisions.

Please note, no Stooges were hit in the head with a hammer, poked in the eye, or set on fire during the writing of this blog entry.

9 Responses to Moe, Larry, Curly & Google’s Quality Score Calculation

  1. Loren says:

    Great article, David. This is great information for anyone advertising on Google, and probably Yahoo once Panama is released. The thing that is so frustrating for an advertiser is that Google does not make this information readily available, especially to the small guys only spending a couple thousand a month in PPC advertising. I don't think advertisers are opposed to Q scores, especially if they prove to make the user experience better and conversion rates go up, just let us know how we can make our sites and ads better to improve the Q score! The black box mentality of Google is very frustrating. Great research on this, David.

  2. Ditto! OUTSTANDING work. The BEST single piece I have ever read on this topic.

    Best of all, you have real world Wall Street experience which adds significant credence to your thoughts.

    🙂

  3. Jeremy says:

    Good summation of information. With all the various hints and allegations floating around about this topic, it's nice to see a concise and cogent breakdown.

    Hopefully pressure from articles like this will lead both Google and Yahoo to be more forthcoming about the algorithms used to make quality decisions.

  4. Jason says:

    It's refreshing to see that someone has (finally) stepped up and brought this blog some meaningful content (in layman's terms with cited references) on a relevant topic. Keep it up!

  5. Ryan says:

    Nice research. As ppc continues to evolve and grow, advertisers will and should demand this information from G and Yhoo. One of these days those guys are going to realize that transparency is the key to longevity with advertisers and publishers both.

  6. Great Post David. I am looking forward to the second part. Quite heavy for the end of the week, but I like it when the head smokes 🙂

    I came across the mentioning of BAYESIAN PROBABILITY several times in combination with almost ever area of Internet marketing.

    Can you recommend any related reading that can provide a basic understanding of the mathematical principles and if possible relate to or uses Internet marketing as practical examples?

    If it is related to Internet marketing then I would have something to relate to as somebody who was doing quite good in math at school, but not made his masters in higher mathematics 🙂

    I'd appreciate it.
    Cheers,
    Carsten

  7. We use exponential moving averages as well for both our click and revenue recency models. The question we need to find out is *what half-life does Google use for its CTR history EMA*?

    Anyone have any thoughts?

  8. Hi Chris,

    Thanks for the comment.

    If you want to figure it out I have some general ideas. First, I am assuming that you are referring to the smoothing factor not necessarily a decay rate or the time for an old value's contribution to shrink in half which can be imputed from a smoothing factor.

    The way I would consider doing it is, as follows (and this is just a high level view):

    1. Set up two different brand new AdWords accounts with no account history.
    2. I would consider setting up a 'fake' keyword that nobody advertises on such as: 'ASDFGLMD' exact match so you can control impressions, CTR, and monitor the effect of one variable at a time.
    3. Generate two different levels of click activity, measure results, change click activity, and you could probably could empirically back into some type of 1/2 life.

    I would also suspect that the 1/2 life factor could also be a function of the competitiveness of the marketplace and search volume. If you view the PPC market as being analogous to a financial market, a highly competitive, high volume keyword space is going to have to factor in information much more quickly than a low volume, less competitive keyword space.

    Best,

    David

  9. Andreas Dzumla says:

    just stumbled about this when looking up what people think and write about QBB. Fantastic summary!!