Yesterday I discussed the concept of normalization of CTR by position and how this factors into Google’s revenue optimization algorithm. Today, let’s discuss the “sandbox.” The sandbox is basically a holding pen for new advertisers or new ad text. Google does not fully show your ads when you initially launch a campaign, instead testing your performance in small bits until it has statistical significance about you. This is the sandbox.

My theory (again, without any supporting evidence!) is that sandboxing has several elements:

  • The average CTR of all ads on a keyword: Thus, if the top position for the keyword “blue widget” has historically gotten a 20% CTR, and your ad has 3 clicks after 10 impressions, Google might be able to quickly determine statistical significance and move you out of the sandbox. An ad, however, with a historical .001% CTR might need thousands of impressions to make the same conclusion.
  • The likelihood of new ads beating historical CTR performance: If a popular keyword has thousands of new bidders every week but historical data suggests that almost all new bidders will not be able to create ad copy that outperforms the current bidders, Google requires very strong evidence for you to get out of the sandbox (and the opposite would be true if new ads typically beat existing ads);
  • Your historical account performance: If your account has very low CTR throughout, the sandbox penalizes you, and vice versa.

All of this amounts to a risk/return analysis for Google. Google always wants to increase revenue by allowing higher performing ads into the auction, but they want to balance that return with risk mitigation against lower performing ads that reduce the revenue per thousand impressions.

One interesting point about this theory is that it essentially considers two factors totally independent of your actual behavior, and just one factor that is account-specific. In other words, your control over whether you could ever actually show up on a given keyword is largely determined by the historical data Google has amassed before you entered the auction.

This truth, combined with confusion about CTR normalization, has led to the notion that you can – or must – “buy your way into the auction.” The notion is that if you bid $50 CPC on a keyword that would cost an existing advertiser $.20, you can push your way into even the most competitive keyword auctions. My sense is that this strategy works for overcoming sandboxing on the most challenging keywords, but has no impact on Google’s scoring of your CTR. So if you are stuck in position #40 on a top keyword, Google is basically saying “we don’t think any new advertisers can make us more revenue on this keyword, so go away.” In such a case, a huge bid gives Google incentive to lower the barriers to entry and give you a shot at competing. Of course, once you lower your bid down to the level of other players, you must still overcome CTR normalization like everyone else!

Any questions?


  1. David White January 7th, 2011

    Hi David,

    It is nice to see you blogging again.

    I think that every reasonable marketer must agree to your basic premises when stated in this way:

    1. Google normalizes CTR according to position
    2. Google treats new ads differently than old ones

    But, I do not think claims that you can “buy your way in to the auction” should be dismissed as foolish. I have seen (albeit in 2007 / 2008) cases where a newly launched keyword would maintain stable metrics for a few weeks and would then see a simultaneous decrease in CPC and increase in position (i.e., going toward the top of the page). At the time, based on launching several new campaigns, I found it more effective to bid high and first and then gradually go lower than to bid low and try to sneak up.

    There is an alternate theory that is consistent with Google’s rational motivations that still allows for the concept of buying your way in to the auction: what if Google makes a small deduction when estimating an ad’s CTR in higher positions? By doing so, Google would accept a slightly lower expected RPI in exchange for reducing the volatility of the RPI.

    For example, you may sit at position 7 with a CTR of 1%. Google’s true estimate of your CTR at position 1 may be 5.1%, but Google could choose to use the estimate of 4.5% in the auction calculations.

    The current ad at position 1 is a sure thing, barring Humean objections and statistical noise. If it performs at 5% today you can expect it to perform at 5% tomorrow. On the other hand, promoting the ad from position 7 to position 1 is not a sure thing. Google may estimate it to have a CTR of 5.1% at position 1, thus beating the incumbent ad. But the actual CTR at position 1 will vary against the estimate of 5.1%. Rather than promoting the ad from position 7 to position 1 to reach for the extra 0.1% expected benefit, Google may prefer to take a sure thing.

    If this is true, the implication for the advertiser is that it’s easier to start at the top and cut back than to start at the bottom and sneak up. In the example above, the incumbent at position 1 gets credit for the full 5% CTR, but the ad at position 7 must have an estimated CTR of, say, 5.5% before it would be promoted. So, if you think that the top positions would be economical for your campaign, you would do better to bid high at first, win the position you want, and gradually allow your bid or actual CPC to reduce over time.

    So, this theory is consistent with Google’s rational business motivations but would still allow some room to game the system by bidding high to “buy your way in to the auction.” But, same as your arguments – only speculation and no proof.


  2. davidzhawk January 7th, 2011

    Hey David, always good to get a comment from you!

    I think I actually agree with you. I was not disputing the notion that you could “buy your way in” – I was disputing the idea that by bidding highly you could both buy your way in *and* artificially inflate your CTR.

    For keywords where Google has high statistical significance and a very high RPI, indeed, the only way to show up toward the top of the page would be to bid so far beyond the expected RPI that Google sees value in giving you a shot.

    I think your notion that Google may allow testing at slightly below their expected RPI makes sense too, as it is a rational trade off between risk and return.

    So I think we agree, I just may not have articulated it well in my post!

  3. Terry Whalen January 10th, 2011

    David, thanks for the post. I think I agree with both of you – but at the end of the day, I’m not sure it changes the way I approach new keywords or campaigns. I still focus on CPA – I will generally bid up when CPA is lower than target, and bid down when CPA is too high. If a keyword or ad group isn’t getting enough click spend to evaluate whether things are working or not, I’ll increase bids. If there is no spend or very little spend, and no history of spend or conversion data, then the keyword may as well not even be in the account – I bid those up.

    I don’t really think I’d try to bid super high on anything unless all my previous bids had the result of zero spend, or not enough data to judge conversion rate and CPA. Thoughts?

  4. davidzhawk January 11th, 2011

    Terry, I generally agree with you. I think the issue here is really around getting into an auction that has really high ad rank. In your example, you are suggesting that you would slowly bid up a keyword to try to get some impressions, but ultimately still manage to your CPA objective. I agree with this approach. The key here is that you might have a keyword with 0 impressions in your account but in reality may be able to get you 1000 clicks a day. The only way to know this is to take a bit of a risk and bid high enough to “buy your way in” which may also result in a temporary CPA higher than your goal. But at least you’ll know quickly if this keyword is a hidden gem or a dud!

Leave a Comment

David Rodnitzky
David Rodnitzky is founder and CEO of 3Q Digital (formerly PPC Associates), a position he has held since the Company's inception in 2008. Prior to 3Q Digital, he held senior marketing roles at several Internet companies, including (2000-2001), FindLaw (2001-2004), Adteractive (2004-2006), and Mercantila (2007-2008). David currently serves on advisory boards for several companies, including Marin Software, MediaBoost, Mediacause, and a stealth travel start-up. David is a regular speaker at major digital marketing conferences and has contributed to numerous influential publications, including Venture Capital Journal, CNN Radio, Newsweek, Advertising Age, and NPR's Marketplace. David has a B.A. with honors from the University of Chicago and a J.D. with honors from the University of Iowa. In his spare time, David enjoys salmon fishing, hiking, spending time with his family, and watching the Iowa Hawkeyes, not necessarily in that order.