So the experiment will slide down the ad-balancer for a % of traffic?

No you can compare settings to find the optimal one. I have an experiment running now %75 vs %100. I'm curious to see the result, given the current situation. (LOL morphing to sobbing....)

I hadn't noticed the "Experiment" feature. Thanks for sharing. (I'm trying 90 vs 80 percent estimated earnings right now.)

Trying 60% vs 100%... results in 5 days.

Isn't there really only ever point using this, if you have backup ads? Or is the thinking it can improve user-friendliness and therefore have people staying on the site longer?

Because otherwise, isn't it just the RPM that's improving, but the overall earnings remaining the same?
Fewer clicks with higher CPC = more clicks with lower CPC.

I'm thinking long term, it's a matter of sustainability.
If you think short term, go on and flood your users with ads.
I like to show ads to users that click them, and don't bother anyone else.

If less ads = same money, why not?

Side benefits:
- no need for adblockers in the future
- no ad spam parasites
- better user experience

Regarding the topic: day 2, 0% variation.

@Steviec79, there are many advantages to using the Ad-Balancer, beyond backup ads.
1- It helps combat ad blindness, showing ads, specially low quality ads will lead to users zoning them out and ignoring them. Place a high quality ad that is relevant to the user alone on a page and the user is more likely to click it.
2- elminating spammy low quality ads. Blocking ads using the Ad-review center is futile as the spammers have developed techniques to avoid being blocked like using mutliple AdWords accounts and target urls. But one thing that remains consistent is that spammers are cheap bast...rds. So by blocking the lowest paying ads you blocks the spammiest ads. Furthermore blocking an ad in ad-review, blocks one ad, but leaves room for next ad, that is the next highest paying ad, so assuming that price determines quality by blocking ads your simply allowing worse quality ads to bubble up. Ad-balancer removes inventory from the auction process, thus reduced supply for equal demand should increase bid prices.
3- page speed, ads slows pages down and chew up the user's band width, why show ads that pay nothing you rather speed things up for your users, then the user might comeback next time.

These are the three main reasons I am sure there are more. But now you do not need to take my word for it. Run an experiment and see the outcome, the addition of the experiment feature directly on the ad-balance pages makes it incredibly easy. Like that you will know if it works in your specific situation. It is a great feature but it is not for all sites.

here are some more threads that discuss the topic:
[webmasterworld.com...]
[webmasterworld.com...]

Nick - some valid points there.

The thing about the Ad Balancer experiment is that it seems to only show you the RPM. Not the overall earnings. It would be better if you could see more information. So it's possibly better doing the experiments away from the tool.

What's wrong with RPM, it is the only standardized metric. You cannot compare any other metrics such as CPC or CTR as these are subject to biases.

RPM only tells you part of the story.

If you choose the Ad Balance at 50% so it shows ads only 50% of the time, and because it's showing better quality / higher bid ads, the RPM goes from £1 to £1.50 then your RPM has improved by 150%.
But the earnings could have dropped overall by 20%.

What it should also show is Estimated Earnings. So you know what the Control has earned and what the change has earned.

To use the Adsense Experiment tool, it only tells me what my RPM is - well that's not telling me whether I should change it or leave it how it was originally. Because RPM is only part of the puzzle.

If you choose the Ad Balance at 50% so it shows ads only 50% of the time,

No it shows the top 50% of ads based on earnings. There is no relevance to time, it could show all those ads in 1 hour and then not show anything the rest of the day (this is an exaggeration but it illustrates the point).

Your general point about RPM is a good one, but I'm not sure your right (or wrong), I need to think about it.

Sorry, I did mean the top 50% based on earnings.

The only way RPM alone would be the only metric you need when split-testing is if Google is 100% bang on accurate when you move the slider and it shows your earnings still at 100% - I'm not convinced.

I'm not sure they can know that for certainty, as higher CPC ads may not be clicked on enough to balance the earnings out. This is why they need to show Earnings before the Ad Balance change and after. So it's a fair A/B split test.

When Google say your earnings will remain at 100% when the slider moves down to, say, 75% - if that is for definite, then the RPM by itself on the Experiment, would suffice as you'd know to stick with the highest RPM.

I've just come to the end of an Ad Balancer experiment:

The Control (100%) beat the Variation (90%). The drop to 90% coverage dropped the RPM by 4%.

Even though it's a small % I find it a bit odd that dropping the coverage to 90% also actually lowered the RPM.

@Lagonda can you tell how is going the experiment ? I just launched mine

After 3 days, there's already a result.

Original
Ad fill rate: 60%
Variation
Ad fill rate: 100%
—
RPM
-7%
—
Ad session length
-4.2% to -1.5%

Even if it was +7% RPM with 100% ads, I would choose 60%.
Long term, long term...

So 2 experiments with different results. And as I mentioned earlier, there's no way of knowing for sure that a lower Fill Rate produces higher actual earnings.

A higher RPM can still have lower overall earnings.

@steviec79

If you choose the Ad Balance at 50% so it shows ads only 50% of the time, and because it's showing better quality / higher bid ads, the RPM goes from £1 to £1.50 then your RPM has improved by 150%.
But the earnings could have dropped overall by 20%.

I have taken sometime to think about this and what you are saying is true. But earnings are heavily biased, mostly by traffic. I can easily earn $10 with 10,000 page views but it would more difficult to earn $10 with a 1000 page views. So comparing these two earnings figures one would think that these scenarios are the same. If one compares RPM for these scenarios and it becomes immediately apparent: RPM of $1 for 10k PVs versus RPM of $10 for 1k PVs.

The other bias that earnings is subjected to is size. A 2% difference on $100 is $2, where as 1% on $10k is $100. Now if one had a report that said the difference was $100 most people would say "that's a no brainer I'll take the $100" but if the report showed a difference of $2 one may say "meh it's only $2, I'll stick to the status quo". Where in reality the $2 outcome is 2x better than the $100 outcome.

Remember that the A-B testing is carried on a small sample of ones total impressions and it be would scaled accordingly, like $100 vs $10000 above. To then make the sample comparable to ones total earnings one would need to convert the earnings to RPM and then multiply RPM by impressions. So you'll end with RPM regardless.

The one thing that is missing from the results is some indication as to whether or not the results are statistically significant. A 4% change or I'm currently seeing 2% and even 7% may not be statistically significant, meaning that the test cannot explain this difference, it might actually be due to randomness.

"RPM of $1 for 10k PVs versus RPM of $10 for 1k PVs."

You've lost me a bit here. RPM is per thousand.

@steviec79

First to quote text you can use the word quote surrounded by square brackets [] and then /quote in [] at the end of the text like this

[ quote]Some text here[ /quote]

RPM = earnings / number of impressions * 1000

For $10 earned with 1000 impressions:

$10 / 1000 * 1000 = $10

For $10 earned with 10000 impressions:

$10 / 10000 * 1000 = $1

Why use Ad Balance if Google isn't showing ads anyway?

Good question.

To minimize the spammy ads.