Dec 52011

The AdWords Auction and Why Google's Automatic Bid Pushes are Risky

It's the height of the holiday shopping season and marketers are working feverishly to drive record levels of revenue while the getting's good.  In a timely move -- on Cyber Monday, no less -- Google introduced the ability to set Automated Rules in AdWords to push keyword bids to the CPC required to show on the top of the page.  Common sense and a quick review of top of page vs other segmentation stats tell us that click-through rates are significantly higher for ads above the natural listings -- compared to those on the right rail or bottom of the page -- so pushing bids in this manner will certainly drive more traffic. But does it make sense for advertisers to utilize this new feature?

To help answer that question, let's back up a bit and review the mechanics of the AdWords auction.  In simple terms, where your ad ranks in the paid listings depends on your keyword's bid and it's Quality Score.  As Google states, "You always pay the lowest amount possible for the highest position you can get given your Quality Score and CPC bid."

For example, let's say my only competitor bids $1.00 at a quality score of 5. Their Ad Rank is $1.00 X 5 = 5.  If I bid $2.00 with a QS of 10, my Ad Rank is 20 and I win the auction.  In order to beat my competitor's Ad Rank of 5, I actually only need a CPC of $0.51 per click, which is well below my bid and it's what I pay.

Quality Score, by name, wasn't always part of the equation, but Google was smart to base Ad Rank on expected CTR from the beginning.  Consider a scenario where an advertiser has a very irrelevant ad, but at a very high bid.  As we'll show below, without consideration for probable CTR, that ad would show prominently, but generate few clicks, hurting Google's potential revenues and making for a poor user experience.

While other factors influence QS, it is still largely a measure of an ad's expected CTR and, for our purposes here, we'll assume those other factors are equal.  With that in mind, let's get some assumptions out there before getting to some more detailed auction hypotheticals.

From left to right, the first table simply lists the 10 advertisers in our auction, the Quality Score they have been given by Google, and their initial bids.  The second table gives the expected click-through rate for each position for the average advertiser.  This is not meant to represent CTRs for AdWords auctions universally, only to provide a reasonable baseline where top positions generate clicks at significantly higher rates than lower positions.

The final, and likely most debatable, table provides an estimate of how much higher an ad's click-through rate will be in any position given its Quality Score.  Again this is not a universal law of AdWords, based on any insider information, nor is it how QS is really defined.  Instead, these are just levels that would make sense to Google given Quality Score's impact on CPC.  For example, a QS of 7 is generally considered average, so it's given a multiplier of 1.  Now, if I had a keyword with a QS of 10 that ranks just above a QS 7 keyword that has the same bid, I would get a 30% CPC "discount" on my bid, just by the auction mechanics described earlier.  In order for that to work for Google -- generate the same revenue -- my ad would need to generate clicks at 1.43 times the rate of the other ad (0.7 X 1.43 = 1).

A Basic Ad Auction

Alright, let's take those assumptions and drop them into a world where Quality Score/expected CTR is not considered.  Here's how it would play out:

Ranking is determined solely by bid and each advertiser pays a penny more than the advertiser beneath them.  Advertiser A takes the top position, but only generates a CTR of 17.14%, lower than our expected average of 20%.  Assuming 1000 impressions for this auction, it generates 419 clicks and generates nearly $613 in revenue for Google.

Enter Quality Score

The same auction, but with Quality Score determining Ad Rank:

Advertisers A and B have swapped positions due to B having a significantly higher QS than A, giving B a higher Ad Rank despite their lower bid.  None of the other advertisers change positions, but most end up with a different CPC -- some a bit higher, some a bit lower.  The average CPC across all advertisers only goes up 2.4%, but the auction generates 10% more clicks, resulting in 13% more revenue for Google.

Top of Page Threshold

It rarely seems to be discussed in full context, but as we explained last year, Google's 2007 change introducing an eligibility threshold for showing at the top of the page, effectively made Google a ghost bidder in many auctions.  In brief, if your ad is the last one shown above the natural listings, your CPC cost will be the amount of the CPC threshold, which varies with your Quality Score.  We can view the threshold as an Ad Rank threshold as we have below:

This is the same auction as the one just above, only with the top of page threshold added.  I've made the threshold an Ad Rank of 9 for maximum impact.  Advertiser C pays the full amount of their bid and their contribution to Google's Revenue jumps from about $26 to $64 with zero additional clicks received.  The rest of the field goes unscathed.

Automatic Bid Pushes

Now we finally get back to the basis for this discussion -- Google's new Automated Rules option to push bids to the top of page CPC.  Let's say advertiser D has adopted the settings outlined in Google's blog post announcement:

Google raises D's bid so that their keyword reaches the top of the page.  To do this, advertiser D must beat out advertiser C who previously held the bottommost top of page or "promoted" position:

Advertiser D doubles their click volume, but pays three times as much per click.  The net effect on the overall auction is minimal, largely due to where the top of page threshold was initially set, but what happens when more than one advertiser chooses to adopt an automatic push strategy?

Advertiser C wasn't happy about their clicks being cut in half, so they decide to push to reclaim a top of page position.  They go back and forth with advertiser D until both end up pushing advertiser A to the right rail:

To do this, both C & D must beat advertiser A's Ad Rank of 18 by raising their own bids significantly.  Advertiser D's CPC is now 489% higher than it was before they decided to push to the top of the page and they have only doubled their clicks.

A final hypothetical, what if there are more advertisers pushing to a promoted position than there are positions available?  Here's how that might look:

Assuming C,D,E & F all adopt Google's example settings and allow their bids to be pushed to $10.00 and that the top of page Ad Rank threshold shifts with the actual competition in the auction, we see an explosion in click costs.  Advertiser E, with the lowest Quality Score at 6, sees their bid jump to the full $10.00, setting the Ad Rank threshold the other three must beat at 60.

Google's revenue haul has risen 352% from the basic Quality Score auction above, while clicks delivered to advertisers have only increased 6%.

Use With Caution

It's difficult to think of an advertiser for whom or a scenario in which Google's example settings above -- increase bids to up to $10 on all but deleted keywords to reach the top of page -- will be a winning strategy.  Fortunately, Google does provide a number of segmentation options to limit the advertiser's exposure to runaway CPCs program-wide.  Google itself suggests that, "you might want to only raise bids for keywords that have a good quality score (for example, Quality score >= 6), so as not to bid up keywords that need to be optimized." Sure, that would help, but most will probably only want to consider using these rules on a handful of individual keywords that have special value for branding purposes.

If Google wants to see greater adoption of push to top of page rules, they really need a lever that restricts the bid change to a certain percentage -- i.e. Don't push my bid by more than X%.  There's a certain margin of error in all bid calculations, so if we can generate a lot more clicks for a small CPC increase, it could be worth it.  Even then, problems would remain.

The big one: Unlike the simplified hypotheticals above, a keyword can actually appear across a number of positions at the same bid due to differences in QS for individual auctions so the tradeoff between clicks and costs is a smooth curve rather than a discontinuous, staircase-like graph where a relatively small CPC change can result in one big and persistent increase in clicks.  Since Google isn't making push-to-top calculations for each auction -- the finest change is hourly based on that day's data -- they're really just moving you up the same click to cost curve you can see in Bid Simulator and you're probably better off doing that assessment on your own.


9 Responses to "The AdWords Auction and Why Google's Automatic Bid Pushes are Risky"
Mark, There is one more nuance that must be talked about. The auction process isnt exactly what Google tells you. The first stage is that your bid determines if you qualify for an auction or not. If you have a very high bid you qualify for more auctions but those auctions are not necessarily relevant for your business. The effect of this is that you can get a lot more impressions at the same position for a higher bid and CPC but your conversion rate typically drops. I have covered this in detail here If one doesnt know this , then the auction process can be very mysterious to some. For instance, you bid $1.73 and get position 2 at a CPC of 75 cents. You think the CPC is too high so you drop the bid to 1.08. You might end up with a very similar position like 2.2 or so but your impression volume plummets and your CPC is now 19 cents. (this is real and recent data by the way). The reason this happened is because the lower bid led to the advertisers bid qualifying for a different pool of auctions. Unfortunately, Google does little to explain this step in the auction process with the result that many advertisers think that if they want a certain CPC they can bid CPC+ 1 cent. Not true. Well.. my 2 cents ! George did a recent blog post on position distribution. It will be great if someone did a follow up analysis on a keyword that was at position say 1.1 at a certain bid and then bid much higher to say position 1.05 and then look at the position distributions. I am pretty sure, heck I am sure, that the position distributions will be very different. Sid
Mark Ballard says:
Great points Sid. The decline in traffic quality you would expect from entering additional marginal auctions is further reason to be wary of allowing Google to push your bid to the top of page CPC. This is a bigger concern for terms that are on broad match and/or running on the search network, but applies for any term. We discussed weird position effects in a post awhile back (, but it would definitely be interesting to take another look now that we have the actual position of each click. To your larger point, and I hoped to get this across in my post, the auction is certainly much more complicated than my hypotheticals. Google isn't guaranteeing a top of page showing with this tool, but they aren't going out of their way to explain how it isn't guaranteed even with a certain bid. As I wrote, they're really just moving you along the potential click to cost curve. And, I should have added, probably giving you lower quality traffic in return.
Moshe says:
The underline assumption is that all advertisers will follow suit, and raise their bid for all keywords the may trigger the ad, and that there are only 4 potential competitors for every auction. That is not necessarily true. A more real scenario is that there are many more advertisers, while only one or two will execute this strategy.
Tom says:
Thanks Mark, a very good write-up that will help to explain the issue to clients. i'd also like to add onto Siddharth's point something i've noticed recently (but haven't had time to investigate in detail yet): since Google started showing ads at the bottom of search results a month ago, the race to bid to the top seems to be even less relevant, unless you're not you're an ROI-focused advertiser. For exact matches (so eligibility for the auction shouldn't play a role like it does for broad matches) i've noticed that reducing bids can get you better CTR and more importantly, better CVR than the 'top' positions. Google doesn't make it easy to segment 'other' ad positions so that's why my theory is currently just based on empirical data. But it seems that not entering the race to the top for competitive terms and having your ad shown at the bottom, to the user who bothers to scroll down gives you a more qualified user. My thinking that users who scroll down obviously didn't find anything appealing at the top, and are happy to click on an alternative and that is a very different user behaviour than how users interact with ads on the side. I've noticed this because we have clients in emerging markets where data doesn't always follow the smooth curve described above but is a lot more discontinuous, which sometimes is actually a good thing for spotting tipping points. Anybody has any thoughts on this?
Mark Ballard says:
Tom, I wouldn't be surprised that CVR is higher, if just slightly, at the bottom of the page than it is at the top. Higher CTR would be unexpected though, unless the bid reduction is causing a shift from the right rail to the bottom of the listings. You didn't mention whether these exact terms are opted into the search network, which can significantly alter the resulting average performance. Higher bids should lead to more search network traffic which is usually not as well converting and may have lower CTR. We're looking into bottom of page performance ourselves. Like you said, it's not easy to segment, but currently the ad position ValueTrack parameter will tell you if your ad was clicked in top, side or other (bottom) position. It's a start at least.
Tom says:
Hi Mark, yes, i was surprised by the higher CTR as well, i was wondering if there's any way google could not count the impressions if the ad is below the fold? that would explain it...
Karl says:
If they were not counting impression below the fold it would make sense that the CTR would improve for the same relevancy reasons the CVR would. If the above is true and you had a client with a limited budget it would actually make sense to target only these positions so long as you could get the required volume, it would therefore be more value to advertisers if Google brought in a tool to allow this instead. Strange they have brought in a top of page tool after removing positional bidding because of the issues it caused?
Tom says:
I actually mentioned this to a google rep today and he confirmed that impressions below the fold are not counted. that said, i've also had google reps recommend increasing bids to improve positions and increase CTR in order to have a better QS, while it's widely known QS is adjusted for position. So i'd still not take that as a definite...
Mark Ballard says:
Yeah Tom, I would probably check with another Google rep. I'm nearly certain that below the fold impressions are counted. Maybe the confusion surrounds the ability to exclude ads from showing on the Google Display Network when they would be below the fold.

Leave A Comment