1. The subtle science of bidding Guide
The subtle science of bidding
Understanding nuances of the SEM marketplace
Contents Search keyword management and executing a bidding strategy can be complicated. Beyond the usual bid, cost
1: Bid, CPC, and per click (CPC), impression, and position considerations, several nuances in the search engine marketing (SEM)
impressions:
auction mechanism to consider when making bid decisions. This white paper explores the marketplace features
The real story
that Adobe leverages on behalf of its advertisers to generate maximum return for their advertising dollar.
1: What adwords tell you
2: The actual auction
Before delving into the details, it’s instructive to understand how Google’s Ad Rank system truly works (versus
mechanism
how we are told it works).
3: Brand keyword
management
5: Second order effects
Bid, CPC, and impressions: The real story
6: About the Adobe
Digital Marketing Suite Here is the daily data for an exact match brand keyword. The data is normalized to eliminate day-of-weeks effects.
7: About Adobe Systems
Incorporated Bid CPC Average Position Clicks Impressions
$23.38 $1.35 1.01 1,868 14,105
$17.80 $0.50 1.09 1,701 9,039
$9.42 $0.33 2.01 9,00 5,625
These numbers are surprising. Knowledge of the auction marketplace as well as information provided by Google’s
AdWords helps, but even then, these numbers don’t make a lot of sense. Here is why this data is unexpected.
• If a bid of $23.38 got an average position of 1.01 at a CPC of $1.35, why did a $9.42 bid get position 2.01? You
would expect the $9.42 bid at position 1 because it’s so much higher than the CPC of the current position.
• Why did impression volume drop when the bid changed from $23.38 to $17.80? Oddly enough, click volume
did not change much.
• Why did CPC increase so much from position 1.09 to 1.01?
• Why the huge gap between the bid and CPC?
Although Google data is being used to explain these observations, all search engines employ a similar strategy
to determine CPC.
What AdWords tell you
The Google Help pages state that the CPC paid is a function of bid, quality score, and your competitors’ bid and
quality score.
Step 1 Google calculates the Ad Rank for all advertisers in the auction in which
Ad Rank = Quality Score * Bid
Step 2 Advertisers are in descending order of their Ad Rank.
This determines rank in the auction process.
Step 3 CPC is calculated as:
CPC = (the next closest and lower Ad Rank to yours)/your quality score + $0.01
2. Here is a dataset to understand how the auction mechanics work.
Advertiser Bid CTR QS Rank=QS*Bid Rank2=CTR*Bid
A $25 15% 10 250 3.25
B $10 8% 7 70 0.8
C $10 2% 4 40 0.2
According to Google, Advertiser A has the highest Ad Rank, so A wins the auction and gets position 1. The CPC
that Advertiser pays is:
This explanation is incomplete and incorrect for the following reasons:
• Ad Rank is more a product of CTR and bid rather than quality score and bid. For this example, this is referred
to as Rank2. Because CTR and quality score are strongly correlated, this modified Ad Rank is not much
different from Google’s explanation. Although using QS or Rank2 does affect position in the above auction, it
would affect the CPC calculation. Using Rank2 yields Advertiser A an effective CPC of $5.33 (0.8/15%).
• The explanation doesn’t answer the question of why bidding substantially higher than the position 1 CPC
could still get Advertiser A position 2.
• It also doesn’t answer why a lower bid might get Advertiser A the same position but a lot less impressions,
especially on the brand words.
The actual auction mechanism
The actual mechanism is best understood in the following steps.
• Advertiser A’s bid and the competitors’ bid determine the auction marketplace. Google uses Advertiser A’s
bid and its competitors’ bids to determine which keyword match type and bid combinations participate in an
auction at a query level.
• When the auction participants are determined, the Ad Rank is calculated based on CTR or a CTR proxy and bids.
The CTR proxy that Google calculates is an estimate of CTR at position 1. So if an ad has never seen position 1,
Google estimates it. Like any calculation, the estimates could be way off, which in turn could hurt CPC.
• If Advertiser A wants to come on the left side of the page (above the organic ads), Google has an artificial
threshold to beat. So in a sense, Google’s organic results are competing with Advertiser A for rank.
• Finally, CPC is determined by the formula discussed above.
This modified auction mechanism explains the questions posed above:
• The bid and CPC are more decoupled than typically thought. The bid determines the type of advertiser
competing in the auction. A very high bid on a broad match keyword means participation in many auctions,
which means more impressions and clicks. This explains the observed drop in impressions, even at the same
position (1.01 to 1.09)
• The CPCs observed for position 1 and position 2 are independent, because the participants for those
auctions were different. Although Advertiser A bid higher than the position 1 CPC, Advertiser A could get
position 2 because Google let another high-bidding advertiser participate in A’s auctions when A was bidding
low. You cannot calculate the CPC at position 2 just by looking at position 1’s CPC. You must look at the bid
and the CPC.
Note: Adobe models CPCs by looking at the combination of bid, CPC, clicks, and impressions to get 90%–95%
model accuracy. While it appears difficult to model, it’s possible with sophisticated math.
The subtle science of bidding Guide 2
3. When making bidding strategies and decisions, consider the following.
• Bid is far more important than CPC. Bid determines CPC and the competitors in the auction marketplace.
• Position is only an artifact of the auction. It doesn’t determine anything, so don’t pay undue attention to it.
Instead pay attention to bid, CPC, clicks, and ROI. They are the true performance metrics.
• Dramatic bid changes can adversely affect the CTR estimates that search engines make on keywords as they
move positions. This can hurt CPCs.
• A strategy of bidding brand keywords high to position 1 could be very detrimental for performance. Typically,
this strategy leads to higher CPCs but not many extra clicks. Hence, effective management of an SEM
campaign requires close monitoring of all keywords in the account, including the brand.
Brand keyword management
Many SEM managers believe that brand keyword bid management is easy. Because the ROI on these keywords
is so high, many SEM managers just bid these keywords to a high level and then leave the bids alone. This
strategy can be very inefficient.
An Adobe client wanted to test this hypothesis by bidding their exact match brand keyword from 1.05 to
exactly 1.0 at the end of April. The data presented has been normalized, but the trends are exactly what we
saw.
When the bid was raised by 20%, the impressions increased. However, looking historically, the impression
volume was not abnormally high. On the CPC and conversion side, however, some dramatic trends were
observed. On the third day of the bid change, CPCs shot up three times with a simultaneous conversion rate
drop. The net result was that ROI tanked.
The subtle science of bidding Guide 3
4. Most SEM managers do not realize that there’s a big difference between position 1.0 and any other position.
For instance, take the average position of 1.05. At this position, position 1 is not achieved 5% of the time
because the search engines are experimenting with other advertisers at this position. If the goal is position 1 all
day, the search engines charge a huge premium for denying them the chance to experiment. Moreover, when
bidding very high, you participate in more auctions, many of which aren’t relevant. As a result, the quality of
clicks is lower, and conversion rates drop. While in theory this is only supposed to happen for broad match
keywords, the above example shows that this is not always the case.
You might ask, “Why do the search engines want to experiment with position 1?” One reason is that search
engines need to constantly refine the quality score estimates for all advertisers. Remember, quality score is
based on the estimate of CTR at position 1. If an advertiser has not seen position 1, the search engines base
quality score on an estimate that could be quite wrong.
Here is another example, this time from Bing. The branded keyword was bid to $1 every day. On February 11,
the bid was brought down to $0.26 due to an impression volume drop. When the bid was raised again, the
impression volumes recovered, but a much lower position at a higher CPC was observed.
When the bid was lowered, the impression volumes fell, and Bing’s CTR estimate at position 1 was inaccurate.
As a result, Bing estimated a much lower CTR than before, so they began to charge a much higher CPC at the
same position.
What is the advertiser to make of all this?
• Bid management for brand keywords is complicated.
Due to the subtleties involved, you need a very sophisticated approach to managing brand keywords.
• An off-hand approach to brand keyword bid management does not work.
The traditional strategy of bidding all brand keywords very high can be detrimental, leading to higher CPCs
and lower conversion rates. Moreover, advertisers are exposing themselves to the whims of the search
engines. Smart brand keyword management requires a highly accurate and precise bidding strategy just like
non-branded keywords. Using sophisticated mathematics, you can build keyword bid, CPC, clicks, and
performance trade-off models to 90%–95% accuracy.
The subtle science of bidding Guide 4
5. • Position 1.0 is usually very expensive.
The commonly held notion that position 1.0 and 1.2 are very similar is wrong. The search engines charge a
much higher CPC for position 1. It’s almost always better to be at position 1.05 or lower than position 1.0.
While the point is subtle, its effect on brand keyword performance is huge. This also means that you must
model out keyword performance at the high positions with a lot of granularity.
Second-order effects
Many marketers like to think that each keyword operates on its own and that its performance is a function of its
own bid, CPC, impression, and ROI trade-offs. However, the truth is that for effective campaign management, you
must look at keywords simultaneously to make smart decisions.
Simultaneous keyword management has two parts. One is the bid management piece. Assuming the exact bid,
CPC, and performance trade-off for every keyword are known, look at the trade-offs of all managed keywords
at the same time to make optimal bidding decisions. The outcome of this approach is called Portfolio Theory, a
rigorous mathematical method that guarantees the best outcome for any goal. The details of this method are
covered in the white paper “Algorithms and Optimization.” The other part is second-order effects:
understanding keyword performance trade-offs due to decisions made on other keywords in the campaign.
Consider a Google brand campaign where the bulk of traffic came from three broad-matched brand keywords.
Clearly, December 6 was a disaster. Not only did impression volume tank from 500,000 impressions to
150,000, but spend went up from an average of $600 per day to $12,000.
The first clue to what happened comes from the impressions. On a usual day, the bulk of the impressions came
from Brand 1, but on December 6, its impression volume decreased dramatically. The bid, CPC chart below
provides more insight as to what happened.
The subtle science of bidding Guide 5