Forum Moderators: martinibuster
Smart pricing doesn't work at all as described or intended. If it did the results wouldn't be so seemingly random or fluctuate so wildly.
I haven't seen wild or seemingly random swings that I would attribute to smart pricing. On a daily basis, I do see some fairly large swings, but I can usually connec them to changes in traffic or in advertisers. And those swings have died down as my overall AdSense impressions have grown.
So I'm wondering about this. If the smart pricing factor is updated weekly, as has been suggested here, then its impact should only be noticeable from week to week, and not day to day.
What have other people seen? What kind of changes in earnings or EPC has smart pricing caused, and how do you know it was smart pricing, and not something else?
And what kind of site(s) do you run? That may have an influence--I run a small, fairly focused, content site. Perhaps that kind of site is less vulnerable to fluctuations...
One week the EPC for one site will be 1/2 what it was the week before with the other site doing about normal. The next week they might reverse themselves with the high one going low and the low one going high. Other times they will both be either high or low.
This cannot be attributed to the standard responses such as advertisers lowing their budgets or pausing their campaigns - by and large the same ads and advertisers are on both sites.
Traffic drops or increases aren't the culprit because these wild changes take place during long periods during which there is no noticable difference in traffic levels.
Clickthrough rates can't be the problem because they almost never deviate by more than 1.5 percentage points from the daily average.
If smart pricing worked as decribed, clicks from a site wouldn't be worth $x today, $x/2 the next week and then back to $x the week after. The value would drop (or rise) until it found its proper level (as determined by the smart pricing algo) and stay in that neighborhood. It doesn't.
Also, if smart pricing worked as described it would affect both of these sites in exactly the same way at the same time. It doesn't.
My site is a travel-planning site with about 5,000 pages of editorial content on many different subtopics.
My busiest and most profitable site got hammered by smart pricing shortly after its introduction. EPC's fluctated rather wildly but the general trend was down, eventually to a level where I determined that I could do better by going back to regular affiliate advertising. I removed the AdSense code from my 8k+ pages and replaced it with affiliate links.
Literally the next day I got phone calls and emails from several advertisers wanting to know why I had removed Adsense from the site. I explained that smart pricing had reduced the income to a level where it no longer made sense to run Google ads on the site.
They were flabbergasted, and most of them bought advertising directly from me for more than they were paying per click to Google and their ads are still running many months later.
After Google introduced site targeting I placed the AdSense code back on that site to see what would happen and within a couple of days EPC and dollars went to an all-time high and it has stayed in that ballpark for months with little fluctuation at all, even with the other sites acting like a yo-yo.
The smart pricing algo didn't have the same high regard for the quality of clicks from that site as the advertisers who are making a conscious decision to target it - and they must like it because they're staying.
I see day-to-day EPC variations of up to +/- 10% (rarely more) from the monthly average. I haven't observed any week-to-week variations that would suggest changes in smart pricing.
You have one of the most professional and useful sites on the Internet - no doubt the advertisers love targeting your site.
My eCPM was remarkably steady up until the start of January and then halved, and has been pretty steady again at the lower level. Interestingly I saw the same sort of change for other (CPM) ad networks I am using, so I have simply put this down to post-Christmas belt-tightening by advertisers (especially as one of the ad networks has me down in the "Digital Cameras" type of category).
I have noticed that I have been able to buy a LOT of AW clicks for my site a lot more cheaply than before (all at 1p or 2p, ie GBP0.01/0.02 or USc2 or USc3, still with good coversion percentages) in about the same period, so I think that the market is just much slacker nowth.
So, if eCPM volatility is meant to be an indication that SmartPiercing(TM) is broken, then it does not seem broken to me, even if I'm being paid less at the moment.
Rgds
Damon
I'm also intrigued by your assumption that the stability in EFV's earnings is due to advertisers targeting his site. Couldn't his earning's stability be due to overall good results for ads on his site, coupled with a good breadth of content?
EFV, do you think that most of your ads are site-targeted? that most of your ads are appearing on your site due to keywords? or some of both?
AdSense has stated that smart pricing is applied across an account, and not by site,
Where can this statement by Adsense be found?
birdstuff, the experience you describe is very interesting, but since AdSense has stated that smart pricing is applied across an account, and not by site, I am still wondering how you KNOW that the fluctuations you see are caused by smart pricing?
I don't know with 100% certainty, but when you rule out all other variables there is nothing left except smart pricing - the only variable that we cannot "see" or measure.
I'm also intrigued by your assumption that the stability in EFV's earnings is due to advertisers targeting his site.
Surely Google wouldn't omit from their algo the most concrete and easily measurable piece of evidence of the "value" of a click from a given site: the number of advertisers who make a conscious decision to place their ads on a particular site and leave them there if they like the results... If they do omit this important variable then smart pricing is in even worse shape than I thought - and the gods must be smiling on my site.
My opinion is that it is not SmartPricing that causes huge fluctuations. That issue might arise when the topic of your site is borderline between alternate arguments. One with high paying ads and the other one with the opposite. Rollercoasting between the two, AdSense tries to guess new niches. Sometimes it finds a good one. Some others...
That would make sense but the evidence in my case rules it out. The two sites mentioned in my first post are on a very tight but very popular niche subject and the ads are always 100% dead-on topic (except in the very rare instances when the mediabot goes nuts for a day or two) and the same ads always show on both sites.
That would make sense but the evidence in my case rules it out. The two sites mentioned in my first post are on a very tight but very popular niche subject and the ads are always 100% dead-on topic (except in the very rare instances when the mediabot goes nuts for a day or two) and the same ads always show on both sites.
Do you know for sure (ie. you have tested) that the ads you see are those your visitors see? What about people from different regions of your country, or the world? Even different computers?
What are you assuming about adsense and how it works that might be wrong?
Not meant as a personal criticism, but there's such a strong tendency to, in the absence of real information, make up assumptions. Newbies should be aware about all these "conclusions".
Surely Google wouldn't omit from their algo the most concrete and easily measurable piece of evidence of the "value" of a click from a given site: the number of advertisers who make a conscious decision to place their ads on a particular site and leave them there if they like the results
AdSense certainly omitted this from smart pricing for some time. If I remember correctly, site targeting only started a year or so ago, AFTER smart pricing had already been running for several months. For that reason alone, I doubt that smart pricing factors in site targeting. And again, this is based somewhat on my experience. As far as I know, my site has not been site targeted, and yet my EPC has not declined since site targeting was set up.
And I'm not sure I agree with your take on the basis for site targeting. Site targeting is paid for on an impression basis, not clicks, so even if some advertisers target sites hoping for good clicks, others are surely looking for impressions only.
I still want to hear what EFV says about this.
birdstuff, there has been some speculation that AdSense may apply a sort of site quality score to publishers (similar to what they do with advertisers), independent of smart pricing. I wonder if your sites somehow sit close to a "tipping point" in that scoring, and sometimes tip one way, and other times tip the other way?
Re someone else's question about smart pricing being applied account-wide, sorry, I can't give you a reference. But I do remember ASA saying so on this forum. And it is sufficiently widely accepted as being so that there have been discussions here in which people complained about how stupid it was for smart pricing to be applied account-wide, and how to deal with that if one has multiple sites.
Do you know for sure (ie. you have tested) that the ads you see are those your visitors see? What about people from different regions of your country, or the world? Even different computers?
I fully expect to see different ads than the ones seen in other countries, but the niche is so narrow that the ads should be consistent there as well, even if they are different from the ones I see - they shouldn't be changing from week to week.
Regardless of the country where the ads are shown, my sites are in a very tight niche and entirely in English so the ads should be targeted the same way everywhere. If not, there is a major problem with the mediabot.
AdSense certainly omitted this from smart pricing for some time. If I remember correctly, site targeting only started a year or so ago, AFTER smart pricing had already been running for several months.
This is true, but the reason it was orignially omitted is obvious - site targeting didn't exist at the time. Site targeting is such a powerful and accurate indicator of advertiser confidence in a site that it's simply inconceivable to me that Google wouldn't have added it to the algo almost immediately upon its release. Actual advertiser "opinions" on the quality of a site should surely trump any other variables used by an algorithm.
The "tipping point" theory is an interesting one, but I fail to see why a site should ever "tip back" the other way once a direction is established, regardless of how close the site is to the "tipping point". That "quality score" shouldn't change unless the site itself changes. Besides, smart pricing itself should preclude a "tipping point" unless changes are made to the site to make it more or less valuable to advertisers over time.
[edited by: birdstuff at 9:38 pm (utc) on Feb. 19, 2006]
So I see.. Usual eCPM in week one, usual eCPM in week two, 35% lower eCPM in week three and finally usual eCPM in week four on. And this happend around the 15th of the month in Jan and Feb.
I sent Google an email last month to inquire about it but only got a form email response that eCPM varies.
Looking back at last year's reports I see more daily flucuations until January of this year. I must say that my monthly earnings went up every month for the past 5 months and then on Jan of this year they seemed to plateau, but the plateau only happened because of one week of bad earnings.
Page views haven't dropped. Clicks haven't dropped either. Only eCPM drop by a full 35% for one week... the 3rd week for two months in a row.
Paul
I'm also intrigued by your assumption that the stability in EFV's earnings is due to advertisers targeting his site.
I don't know how they'd target my site with AdSense CPC ads, unless they'd learned through experience that ads for certain obscure keyphrases didn't generate much traffic from other sites.
Couldn't his earning's stability be due to overall good results for ads on his site, coupled with a good breadth of content
I think that's the most likely explanation. My site does well with affiliate sales, so it's reasonable to assume that AdSense conversions are decent. And having a good breadth of content is like owning shares in an indexed mutual fund: It tends to flatten revenue dips and peaks, since the site is drawing ads on many different topics from many different advertisers.
Having a topic that reaches an international audience may help, too, because British or Australian readers are likely to be surfing the Web even if American readers are watching the Super Bowl or shoveling snow along the Eastern Seaboard.
EFV, do you think that most of your ads are site-targeted? that most of your ads are appearing on your site due to keywords? or some of both?
I think most of my ads are traditional CPC contextual ads, because (1) nearly all of the ads that I see are standard AdSense units with multiple ads, and (2) ads are usually targeted to the page's content.
For what it's worth, I doubt if smart pricing has much to do with up-and-down fluctuations in earnings. Smart-pricing discounts shouldn't change all that much once they're in place, because conversion rates (or anticipated conversion rates) for a given site or type of content aren't likely to bounce up and down like a yo-yo.
Smart-pricing discounts shouldn't change all that much once they're in place, because conversion rates (or anticipated conversion rates) for a given site or type of content aren't likely to bounce up and down like a yo-yo.
Now we get to the crux of the problem with smart pricing. There is no reliable way to statistically determine conversion rates for even a simple majority of Google publishers because relatively few advertisers bother to track them. Some do, but most do not. Statistical anaylis is useless when the sample size is insignificant. The fact that they use conversion rates in the algo at all given the limited amount of data available makes the entire process unreliable.
I think I remember that EFV is a CTR skeptic, but he and I do tend to agree that smart pricing is based on factors that aren't going to swing around a lot.
You don't, and I suspect that a large part of the reason why we disagree has to do with our different experiences. The pricing part of AdSense seems to work differently on our sites. Is that smart pricing, or something else?
What I can say with 100% certainty is that Google is at least giving the impression that smart pricing is severly broken on my sites and it isn't due to varying CTR, impressions, different advertisers or conversion rates. If one of these were the culprit then my two widget sites would be affected the same way most of the time - but they aren't. They are more or less in lockstep with each other in all the metrics - except EPC.
The original post asked for a description of what we see in regards to smart pricing - and the last paragraph pretty much sums it up.
I agree. All of our own individual theories about how AdSense works (or doesn't work) are based upon our own experiences and testing. We can rule out some things in most situations, but we can never be 100% sure about anything as long as Google keeps the inner workings a secret.What I can say with 100% certainty is that Google is at least giving the impression that smart pricing is severly broken on my sites and it isn't due to varying CTR, impressions, different advertisers or conversion rates. If one of these were the culprit then my two widget sites would be affected the same way most of the time - but they aren't. They are more or less in lockstep with each other in all the metrics - except EPC.
Not only can you not be 100% sure about anything, but you can't be 10% sure about anything.
As for your final comment, I would suggest that while you may be 100% sure about your conclusion, I can't see that you know enough to have that level of confidence. You are making a sensible conclusion on the basis of what is outside the black box (and where you have data), but there may be 1000's of other variables or reasons within the black box that you are completely unaware of. Maybe what you observe is a result of an x factor within the black box? You can't possibly EVER rule that out. Hence you can't rule in that smartpricing is a cause of anything through a process of elimination.
We could hypothesise dozens of possible mechanisms in the black box that would account for almost any pattern of data. But we don't know. To ignore that, and to look for solutions only where we have a little light of information is nohow going to get us to any valid conclusions, no matter how fun it is.
At the same time the sites can be viewed as two control samples. Both have enough clicks to be statistically relevent and both have same content type. They have mentioned same basic advertisers on each page. So lets rule out the obvious...CTR is stable and revenue drops. Is a factor of ecpm. Now the big question is what is controlling that? Advertisers fluctuations? ad serving rotation...or smart pricing. Hearing a lot of nay sayers here but so far bird's theory is applicable. What other reason's are there? Be sure to read what bird has posted about similarity of sites and dramatic earnings swings. May not apply to all here however by far the best comparitive sampling of google perhaps changing payouts I have read.
Sorry, but once again, there's a set of unwarranted conclusions here. First "Both have enough clicks to be statistically relevent and both have same content type."
If WE look at what we can fathom about these two sites, they appear "to US" as having certain relevant characteristics. You are assuming that these characteristics are relevant to GOOGLE'S algorithm, and that no other variables interact with the variables we identify about these sites that we do NOT know about. It's false logic. It "may" be true, but it's as likely to not be.
We don't KNOW what variables are used by google to serve ads, determine payouts, or almost anything. We don't know what variables interact with what. You can't assume that two sites are functionally equivalent (treated the same way) by google just because they appear to be equivalent to you (or me).
In other words, because we don't know what variables are relevant to google and how they hypothesize they interact to yield a result, most of what we all write about these subjects is fictional, based on so little understanding of the black box that we could be 100% wrong in the conclusion.
It's unknowable. To us. Perhaps even to google for any specific site or set of sites.
The latter is a characteristic of algoritms. An algo of any complexity in terms of number of variables and the interactions included in the algo among those variables may not be understood in its specific application to a particular situation, even by those that designed it. Even WITH an understanding of how it's "supposed to work".
So, if even google can't predict how an complex algo will function with a specific site or sites (without actually applying it), reverse determining the algo by looking at a site or sites is even more impossible.
What I'm saying here is that all the speculation is fun, it's a natural desire to understand, but if the concern is to increase revenue, all the speculation is just that, and that we'd be more productive in simple empirical trial and error, and controlling the few variables we can control. We can't generalize about the "system" from here.
Finally you said: "What other reason's are there?". The answer is use your imagination to conceive of aspects of algos that might apply that we know NOTHING about.
Here are a few:
1) Google introduces several random factors into the mix to prevent fraud and protect their algo info. Those random factors kick in on given sites at different times, dramatically altering results for specific periods of time.
2) Google uses some sort of visitor behavioral information to alter the cpc invisibly, so that unless two sites have exactly the same people visiting they cannot possible be considered equivalent.
3) Google alters the cpc for a particular visitor depending on some behaviors of the previous five visitors.
4) Google uses some sort of income "smoothing" process based on some unknown model of how websites should generate legit income, and that is evaluated hour by hour. When income from let's say the last x visitors is too high, it compensates by reducing the income for the next 100x visitors for a particular site.
5) Google uses some form of spreading the most "valuable" ads across a range of sites -- like a quota system. The criteria it uses to calculate the "quota" could be almost anything from the ones we'd think would be obvious to things like IP address, how old the site is, how many visitors from Papua it gets...anything.
There's an INFINITE number of plausible (and not so plausible) factors to explain what is going on in the black box that would also explain ANY data we have. We talk about smartpricing because we have been told it's one set of factors that can affect revenue. We don't know about the other infinite possibility of factors that we haven't been told about.
Just because we know about one, doesn't mean it's important, or relevant to a particular result. To focus on what we know about to explain something and ignore all the unknowns doesn't make sense to me.
We could hypothesise dozens of possible mechanisms in the black box that would account for almost any pattern of data. But we don't know. To ignore that, and to look for solutions only where we have a little light of information is nohow going to get us to any valid conclusions, no matter how fun it is.
Other factors real or imaginary cannot be considered, only what we know to be factual. The examples I cite are real. Anything else is pure speculation.
Sure, it's possible that Google has thousands of items in their magician's hat but so far we have seen no evidence of any of them, and even if they exist they do not dis-prove my theory that smart pricing is broken - they might even support my contention that the system is unstable. Perhaps they are trying to juggle too many items at once and keep dropping some as a result. Again, pure speculation that doesn't fit the facts as we know them. The bottom line is that smart pricing isn't working as intended - the number of tricks in the hat is irrelevent.
So, if even google can't predict how an complex algo will function with a specific site or sites (without actually applying it), reverse determining the algo by looking at a site or sites is even more impossible.
If Google can't predict how their own algorithm will function with a specific site they're in deep trouble. Come to think of it maybe that's why the results tend to be random...
Could we drop the metatheory, though? I'd just like to hear some more examples.
they exist they do not dis-prove my theory that smart pricing is broken.
I don't want to sound snarky here, but a) Do you know what smart pricing is in its detail? b) Do you know how it functions specifically? c) Do you know how and when it is applied?
I can live with the idea that "something" is broken somewhere becuase [insert data]. But your comment seems about as meaningful, and more important useful as saying that "the universe is broken".
As for your comment about google being in trouble if it doesn't know how its algo will interact with a specific site , I am wondering whether you've actually ever grappled with developing or understanding an algorithm of even moderate complexity that has multiple variables interacting. My guess is not.
The whole POINT of many algorithms is that you don't HAVE to know how each situation in the micro universe will be handled so long as the overall purpose of the algo is realized in the aggregate.
I think maybe not understanding how algos work, are developed and function on a complex scale is one reason why people believe they can fathom what's happening in the black box, because they believe the processing of the black box is simple, linear and predictable for every scenario. It's NONE of these.
