I was wondering if anyone cared to comment on this.
IMHO, Account history ,URL history & Keyword history are the most important elements for adword success ,one side effect of this is it frustrates and discourages new users.
Whether this a good thing or bad thing i dont know :)
<Added>
I am not sure if Mr.Cringely or his reader know its against the adword policy to flood the SERPS with multiple similar ads for the same service from the same company!
-Dr.X
Do no evil, except as needed...
But then when you're on top you really have to expect people to take potshots at your motto. And when your "corporate model" is the same one that the church uses for sainthood you're just asking for trouble.
The experiment may be flawed, but the theme of Cringley's article about Google is one that I'm hearing a lot these days:
Just my opinion.
-Dr.X
Advertiser 1 goes in and bids $2.00 on a new keyword K (I used my name) Note that there was no clickthrough history on this keyword.
Advertiser 2 goes in and bids $1.00 on the same keyword K.
Predictably, Advertiser 1's ad appears in the first position in minutes.
Advertiser 2, on his machine, then clicks on Advertiser 1's ad.
Advertiser 1, on his machine, then clicks on Advertiser 2's ad.
Question 1: What do you think Advertiser 1 paid?
Multiple choice:
(a) $2.00
(b) $1.75
(c) $1.01
(d) $1.00
Question 2: What do you think Advertiser 2 paid?
Multiple choice:
(a) $1.00
(b) $0.75
(c) $0.50
(d) $0.05
Answers:
Question 1: (b)
Question 2: (d)
Surprised?
This experiment replicates more precisely the same intuitions behind Cringely's article.
I repeated the experiment on another keyword and found a second measurement: $1.73.
I then did an experiment with 3 advertisers bidding $15, $12 and $9 for a keyword where lots of advertisers were already bidding ("postgresql hosting"), and found extremely unpredictable click prices for the 3 advertisers: $10.76, $10.09 and $5.69.
I had no theory about how Advertiser 1 was charged $1.73 or how the first and second advertisers were charged $10.76 or $10.09. Something was very wrong.
I then ran another experiment where I advertised on 8 terms with 4 advertisers (each bidding $1 - $2/click) and got 32 separate click measurements (8 per advertiser). The resulting observed CPC that Google charged each advertiser was in fact consistent with the formula
CPC[n] = ( CTR[n+1] * ( .01 + MaxCPC[n+1] ) ) / CTR[n]
that can be inferred from various AdwordsAdvisor posts.
I then wrote an Unofficial Adwords Position and Pricing Estimator (link not permitted to be included here).
The heart of the remaining mystery is how the "predicted CTR" (CTR[n]) is computed. No precised published information is available (I would appreciate details) In my analysis, it is not possible to infer this by sheer experiment -- e.g. how to understand why Advertiser 1 was charged $1.73 and not some other number.
I have concluded that it is unlikely that there is an honest answer to come from Google. Consider what would happen if it was revealed that Google's CTR[n] measurement was flawed:
(1) advertisers would have to be refunded massive amounts of money
(2) any new method of measurement would result in significant revenue shortfall, inconsistent with GOOG's projected EPS numbers
(3) Google's stock would drop precipitiously
Thus Google has every incentive to not reveal any information to its advertisers or to its investors. As is precisely the case.
Insiders selling billions? Clouds of secrecy? Overcharged customers? Investors cheering on a New Age company? We have seen this already: Enron.
Best wishes.
google = crooked
You don't have to use google.
Why is making money considered evil? Isn't that what everyone is trying to do online? Are they evil because they have a system that makes them a TON of money?
That said. Google's evil.
It's about using AdWords.
This forum covers everything about AdWords advertising. The discussions can range from general discussions about AdWords news, working with the Google API, and working with the AdWords system. Posts about writing ads, AdWords display, CPC, and CTR algorithms are welcome.
Please refer to the AdWords Charter [webmasterworld.com]
I set a max CPC of GBP0.01 for a word which no one else is using (the one cringely tested....) and yet Google makes it inactive unless I:
"Increase quality or bid GBP0.55 to activate"
What does increase quality mean? The quality of what? The keywords, the ad text, the landing site?
Where does the figure 55p come from?
I'm the only one using Cringely's test word for campaign so why can't I bid 0.01p?
At least overture is straight foreward and what you bid is what they take.
Oh, how wrong you are.
While the bidding may be transparent, Overture/Yahoo's traffic has changed dramatically over the past six months (all negatively, mind you). I know this is an Adwords forum, so I'll keep what could be a long rant brief: Advanced Match is nothing like it was before a May change (where the spigot's were thrown wide open) and their search network expansions are junk.
My Google CTR and conversion for search traffic has stayed relatively consistent (both gradually climbing) for well over a year. While Google's pricing model may have changed, their traffic is - at least - consistent in it's quality. That's worth a few cents per click.
[edited by: poster_boy at 4:23 pm (utc) on Sep. 27, 2005]
This new ranking method is just plain crazy? There is no rhyme or reason to it? And I have yet to figure out how to get any kind of traffic from it.
To the best of my knowlege, your ad has to have your keywords in it to lower your CPC. Otherwise it isn't targeted enough and it becomes inactive.
Then you have to raise the minimum bid high enough to rank high enough to get a page view.
Then you have to get a high enough click through percentage which is impossible if you're not ranked high enough and your ad goes inactive.
Come on Google, not everyone is a rocket scientist!
Your formula has several interesting (and partly doubtful) implications. Let me reproduce it here for convinience
CPC[n] = ( CTR[n+1] * ( .01 + MaxCPC[n+1] ) ) / CTR[n]
I guess n+1 is the lower bidder ad position, and n is the higher bidder ad position, right? You didn't define it but it is the only possibility that makes sense.
You formula implies:
1. Case CTR[n+1]=CTR[n] makes sense. Higher bidder pays 1 cent more than the max bid of the lower bidder.
2. The higher CTR[n], the lower CPC[n]. Makes sense either and confirms my observations.
3. If CTR of the lower bidder sux, i.e. CTR[n+1]~=0, the higher bidder gets the clicks for free CPC[n]~=0. I doubt that. If G can not make money on the lower bidder, why should it "choose" not to make money on the higher bidder? Many people on this board report they are the only bidders on the KW, but the min CPC can be as high as $5.00. Thet click the ad and are actually charged $5 per click. Not consistent with the formula...
4. Define:
GoogleProfit[n] = CPC[n]*CTR[n]
MaxGoogleProfit[n] = maxCPC[n]*CTR[n]
Your formula implies
GoogleProfit[n] >= MaxGoogleProfit[n+1]
i.e. Google's algo will choose such CPC[n] that its profit for higher bidder n will be always higher than the max profit of the lower bidder n+1. The beauty of the above result is that "GoogleProfit" is still the only thing that matters. Keyword Quality?! Give me a break! We should think in terms of "GoogleProfit" variables in order to understand the new system. My observations confirm these results so far, although for me it looks more like GoogleProfit[n]>=min(constant, MaxGoogleProfit[n+1]).
arikgub
Math aside, do you agree or disagree with his assertions?
CPC[n] = ( CTR[n+1] * ( .01 + MaxCPC[n+1] ) ) / CTR[n]
CPC[n]: how much Google will charge for a click on Ad n
CTR[n]: Google's *estimate* of Ad n's clickthrough rate
CTR[n+1]: Google's *estimate* of Ad n+1's clickthrough rate
MaxCPC[n+1]: Advertiser n+1's bid
and Ad n is in a higher position than Ad n+1
Normally people us the term "AdRank" to refer to CTR*MaxCPC. That is:
AdRank[n] = CTR[n]*MaxCPC[n]
AdRank[n+1] = CTR[n+1]*MaxCPC[n+1]
You said:
1. CTR[n+1]=CTR[n] ..
I don't know how to make sense of this formula. However, the Higher bidder does NOT pay 1 cent more than the max bid of the lower bidder.
2. The higher CTR[n], the lower CPC[n]...
Yes. This is "Remember:" qualitative rule of thumb Google (AdwordsAdvisor) advertises, which is useful when you are rich enough to not care. The kind of "Remember" qualitative rule of thumb you want to know is more like:
"CPC[n] will change drastically when a new advertiser comes in"
3. If CTR of the lower bidder sux, i.e. CTR[n+1]~=0, the higher bidder gets the clicks for free CPC[n]~=0.
Rounding is if you want a qualitative understanding, but the point of understanding the formula is to understand exactly what happens when Google decides to charge its customers some money. I don't "free" phone service -- I pay $.10/min.
If instead of plugging in ~0 you plug in numbers like:
CTR[0]=5.0; MaxCPC[0] = 1.00; AdRank[0] = 5.0
CTR[1]=0.5; MaxCPC[1] = 0.50; AdRank[1] = 0.25
then what Advertiser 0 pays is:
CPC[0] = ( CTR[1] * ( .01 + MaxCPC[1] ) ) / CTR[0]
= ( .5 * ( .01 + .50 ) ) / 5.0
= $.051
I don't call $0.051 ~=0 when you buy a lot of them.
Google also has changed it so that there is a "MinBid" on each keyword K, as many advertisers are now vociferously complaining about on this board. So the actual formula is really:
CPC[n] = max(MinBid[K], ( CTR[n+1] * ( .01 + MaxCPC[n+1] ) ) / CTR[n])
Sorry if I didn't include that in my formula. You are quite right to point out the missing MinBid[K] component.
That aside, the mystery is the *means* of computing CTR[n] and CTR[n+1]. Everyone knows its something like the ratio of clicks to impressions, but over what time period and whose ads? The specious reasoning that Google gives Advertiser n is:
- we won't tell you MaxCPC[n+1] and CTR[n+1]
- we promise you CPC[n] < MaxCPC[n]
and most people make the following inferences:
A you can't know what AdRank[n+1] is
B you can't know how CTR is computed
C Google is being "smart" about not telling its customers how its algorithm works
A is false, because you can invert the formula: AdRank[n+1] = CPC[n] * CTR[n]
B is false. Google simply isn't audited enough; while the formula is disclosed in an extremely roundabout way and there is PR on this very forum, there are virtually no checks and balances in place so that advertisers can know what CTR[n] measurement system is in place.
C is true or false depending on whether you believe a customer has a right to know what it is paying for, or whether Google is "smart" to invent something pretty close to a casino.
In a better world (one where C is false), advertisers would be able to download from Google's website, certified by a third-party not paid by Google, explaining precisely how the algorithm works:
(i) how CPC[n] is computed (as in the above)
(ii) how CTR[n] is computed
I think advertisers are in the dark about (i) and (ii)
In the meantime, as Cringely's column says: as in a casino, the house always wins.
>You said:
>1. CTR[n+1]=CTR[n] ..
>I don't know how to make sense of this formula.
>However, the Higher bidder does NOT pay 1 cent more
>than the max bid of the lower bidder.
Well, whatever the CTR estimation algorithm is, I assume that after an ad is running for a long time your estimated CTR is equal to your actual CTR, right?
Then, it is not excluded that CTR[n+1]=CTR[n]. And according to your formula the higher bidder will pay 1 cent more than the lower bidder.
>Google also has changed it so that there is >a "MinBid" on each keyword K, as many advertisers are
>now vociferously complaining about on this board. So >the actual formula is really:
>
>CPC[n] = max(MinBid[K], ( CTR[n+1] * ( .01 + MaxCPC>[n+1] ) ) / CTR[n])
The algo of setting MinBid is another interesting issue. In my campaigns I clearly see MinBid fluctuations as a result of CTR fluctuations.
My wild guess is:
MinBid = MinProfitGoogleHasInMind/CTR
where CTR is the estimated CTR you are talking about.
Using data from my campaigns I can say that the MinProfitGoogleHasInMind in the above formula is not that high. With actual CTR above 2% I'm getting MinBids of $0.03.
I used to complain a lot about MinBid myself, but not anymore. Even if I get high mins when starting a new campaign, I just bid whatever I think I have to. As the trial mode is clearly there despite what Google is saying, I do get few impressions untill G thinks the estimation of my CTR is reliable. Then, if CTR is high the MinBid falls significantly. If it is low, the campaign is lame and I stop it anyway.
>That aside, the mystery is the *means* of computing >CTR[n] and CTR[n+1]. Everyone knows its something
>like the ratio of clicks to impressions, but over
>what time period and whose ads?
This is wild speculation again but I believe the estimated CTR is updated recursively with each impression in a manner
CTR[i] = a*CTR[i-1]+(1-a)*HIT[i]
where
i - impression counter starting from i=1
HIT[i] - equal 0 if no user click on impression i, 1 if user cli9cks on impression i
CTR[0] - is the initital CTR taken based on the historical performance of your account and the keyword
As variable i grows the initial CTR becomes less and less performance and the actual CTR (based on HIT function) is more and more important.
a - "forgetting" factor less than 1
At least, that's how I would calculate it if I was Google.
>In a better world (one where C is false), advertisers >would be able to download from Google's website, >certified by a third-party not paid by Google, >explaining precisely how the algorithm works:
>(i) how CPC[n] is computed (as in the above)
>(ii) how CTR[n] is computed
>I think advertisers are in the dark about (i) and >(ii)
Agree.
arikgub
How can I put math aside if all his assertions are based on the mathematical model he suggests?
In general, it seems to me that the model describes the CPC behavior adequately, but I am a novice in adwords. You may want to know the opinion of more experienced advertizers.
arikgub
I'd like to add that adsense is similarly frustrating. They must have an algo in place to not allow you to improve your earnings by very much percentage-wise no matter what you do. And while the adwords advertisers appear to have noticed paying more, the adsense publishers have not noticed G sharing the windfall.
Strong absolute statements by you; my experience is a counter-example.
My AdWords costs (especially overall cost-per-conversion) have gone down dramatically, and my AdSense income continues to rise (slowly and with the usual rollercoaster temporary reverses!).
One sample does not an axiom make...
Rgds
Damon
