- Joined
- May 7, 2018
- Messages
- 4,594
The other day @servandosilva, @jimmyvanilla, @Luke, @agentf and I got into a discussion on Skype with @SelfAdvertiser (Gal) about their Auto Optimizer feature. It turned out to be quite interesting, so we've decided that it would be good to share here on AffLift. 
SelfAdvertiser already has some good documentation about their Auto Optimiser here in their knowledge base, but sometimes you just have some additional questions...
NICK: Do you recommend using the Auto Optimizer, and what settings would you suggest?
I always recommend using Auto Optimizer. When setting up a campaign under the “campaign settings”, you will see SelfAdvertiser’s suggested settings.
Above are the default settings, and once you change the min bid, we automatically suggest max bid accordingly.
When you set the CPA goal, we automatically suggest the leaning budget per source we think is best for this traffic and targeting type.
Specifically, for RON style campaigns (push, pop, and domain), we recommend not higher than 25% of the CPA GOAL for learning budget per source.
NICK: Why is only 25% of the CPA GOAL recommended for the learning budget per source?
RON traffic is high, and there are many sources. If you sample them each with more than 25%, you will spend a lot of money on the learning phase we want to spend wisely.
In any case, you can make sure your learning budget per source matches your goals by calculating the number of clicks you could get with this budget, to see if it's enough.
Learning budget/min learning bid (CPV) = number of clicks for learning budget
For KW campaigns, we recommend the CPA GOAL as the learning budget since traffic is more targeted, and we want to give it more chance to convert.
NICK: Can you explain a bit about how the optimizer works? Do sources get paused automatically, or is it more just optimizing the bids?
The Optimizer works in 2 phases:
optimizing is always being done on a source level optimizing is being done all day long in real-time
Each source goes through a cycle. The learning phase will "end" if a source reaches its "learning budget" or has conversions.
If a source is not performing well, the bid will gradually decrease to almost 0 (you will see a lot of 0.000001$ bids) - that is equal to pausing since it won’t be able to buy traffic with such a low bid.
Note we never fully pause a source, just lowering the bid to a minimum as possible - giving it the option to reawaken if a late conversion appears.
If a source has conversions, we calculate the optimal bid according to CPA GOAL, number of conversions, etc...
So the optimizer pauses underperforming sources and optimizes bids for the ones that had any conversions at all in a way that keeps them within the CPA GOAL.
The range of min-max bid gives the optimizer the chance to raise the bid in case a source is performing well, and we should increase the bid to gain more traffic to withstand the competition. In many cases, most sources won’t reach the max bid if not needed.
So our recommendation is always to keep at least three times between min and max bid to have enough range to play.
In many cases, you will see bid being lower than the min bid in sources on optimizing phase if needed to stand the CPA GOAL. But never higher than the max bid
NICK: At what point would a source have its bid reduced to almost 0? At the end of the learning phase, with no conversions, or later on in the optimization phase over time?
At the end of the learning phase, with no conversions or if the cost per conversion is way too high, it will gradually reduce the bid to try and remain in the CPA goal, and in some cases, bids can go down to a very low bid (0.000001$).
NICK: Is the bid optimization only considering the source as a whole, or does it consider other things within the source (i.e., browser, OS, etc.)?
Yes, all targeting settings. Device, OS, Browser, etc.
The only thing not considered is the ad creative for push campaigns (see the next question).
You will see an average bid on the platform, but backside, there are many different bids per each of the various targetings.
NICK: Are bids also adjusted for each creative in each source? For example, If I had three creatives and only two of them convert and the third one doesn't, would the 3rd creative's bid get dropped so low it gets no traffic?
No, not yet. It is up to you to disable creatives yourself.
NICK: Do bids get optimized per segment within the source then?
Yes, that is the best way to optimize with SelfAdvertiser. And it happens all day long, in real-time.
NICK:
I know! That’s why we love it so much
.
NICK: I'm really curious about the 25% of the CPA goal for the source's learning budget recommendation... Why is this what you suggest if the source will essentially be paused at the end of the learning phase without conversions? What if, after 50% of the CPA goal, there could be a conversion? Won't a 25% source learning budget cause me to miss out on a lot of potential volumes?
You are right, and that is why on a KW campaign, we recommend on 1*CPA GOAL and, in some cases, higher.
But on a RON campaign, there could be many sources - thousands in many cases.
Let’s say your cap goal is $4. if you spent $2 on each source, or even more before pausing (which in some GEOs is a whole lot of clicks), you will spend a lot of money on learning.
Since RON campaigns have many sources, we believe it’s better to eliminate the non-performing ones quicker and then going through more potential traffic so you will find the best ones for your offer.
NICK: Is this "magic" auto-optimization effective on multi-geo campaigns?
No, the magic is only on a single GEO campaign since it considers all other settings, and the performance per source per geo could be very different, it could damage performance.
JIMMY: Do you have any recommendations for the CPA goal ratio - e.g. 80% of payout, etc.?
I always recommend that the overall CPA GOAL should be the payout- % Profit margin you would like to keep.
Let’s say your payout is $10 and you would like to maintain a 20% profit margin.
CPA GOAL = $10*(1-0.2) = $8
CPA GOAL is only for optimized sources, meaning sources that are converting. So, setting the CPA goal too low to try and avoid losses, will cause the optimizer to lower bids in performing sources, sometimes unnecessarily.
JIMMY: What kind of spend is the algorithm going to allow before pausing a source (e.g., 2x/3x payout, etc. with 0 conversions)? What about sources with conversions (i.e., what kind of CPA will it allow it to get to before pausing)? I’m curious to know how aggressive it is (e.g. with Propeller CPA Goal, I’ve found I’m much more aggressive than the algorithm, so I still have to do some optimization)...
We need to divide it into 2 parts- the learning phase and optimizing phase of a source cycle:
Learning phase - the system will pause a source (reduce bid to almost $0 once it reached the learning budget and didn't have any conversions). So, in case the learning budget is $2, it will be paused at around $2 spent.
Optimizing phase - sources get optimized all day long, through all targeting settings, taking into account a long period of statistics.
If a source is on average doing good in comparison to the CPA GOAL, the optimizer will try to buy more of it.
If a source is underperforming, the optimizer constantly calculates the CPA for all targeting settings, and you will see it will lower the bid and stop buying traffic.
We don't pause it straight away if the CPA GOAL is let's say $8 and the source is at $9 CPA. we try to optimize and reach the $8 CPA with a lower bid, and if we “don't succeed” we again lower the bid and try.
This happens all day long and if a source does not reach the goals it will be paused.
SERVANDO: If we make changes in the campaign during or after the testing phase, does it reset the algorithm and start from scratch? What if we do some optimization (e.g. blocking a browser version, ISP, or OS) on our end after the campaign has been optimized, is it recommended?
No, making changes to the campaign settings during the testing phase does not reset the algorithm and does not start from scratch.
A couple of examples:
SERVANDO: Are there any recommended settings for traffic distribution and frequency capping while using Auto Optimizer?
Yes, the default settings when opening a new campaign is our recommended one, at least to start with.
You can see that for RON campaigns we recommend 1-12 hours (frequency) and evenly distribution of the budget.
** Important to note that if you set up a campaign with a really low budget- $10 for example, the $10 budget will be divided throughout the day to less than 50 cents per hour.
That is why we recommend at least a $20 - $50 daily budget.
For KW campaigns we recommend 1 - 30 minutes (frequency) and as soon as possible, and that is because on a very targeted campaign, we want to try and buy all available traffic for the bid ranges you set. If you would set it on evenly, for example, you could “lose” potential traffic because of the hourly budget.
----------------------------------
Special thanks to Gal from SelfAdvertiser for answering all our questions.
If you have ANY questions about SelfAdvertiser's Auto Optimizer feature feel free to ask them here in this thread. @SelfAdvertiser would be more than happy to answer them all for you.
SelfAdvertiser already has some good documentation about their Auto Optimiser here in their knowledge base, but sometimes you just have some additional questions...
NICK: Do you recommend using the Auto Optimizer, and what settings would you suggest?I always recommend using Auto Optimizer. When setting up a campaign under the “campaign settings”, you will see SelfAdvertiser’s suggested settings.
Above are the default settings, and once you change the min bid, we automatically suggest max bid accordingly.
When you set the CPA goal, we automatically suggest the leaning budget per source we think is best for this traffic and targeting type.
Specifically, for RON style campaigns (push, pop, and domain), we recommend not higher than 25% of the CPA GOAL for learning budget per source.
NICK: Why is only 25% of the CPA GOAL recommended for the learning budget per source?RON traffic is high, and there are many sources. If you sample them each with more than 25%, you will spend a lot of money on the learning phase we want to spend wisely.
In any case, you can make sure your learning budget per source matches your goals by calculating the number of clicks you could get with this budget, to see if it's enough.
Learning budget/min learning bid (CPV) = number of clicks for learning budget
For KW campaigns, we recommend the CPA GOAL as the learning budget since traffic is more targeted, and we want to give it more chance to convert.
NICK: Can you explain a bit about how the optimizer works? Do sources get paused automatically, or is it more just optimizing the bids?The Optimizer works in 2 phases:
- Learning phase
- Optimizing phase
optimizing is always being done on a source level optimizing is being done all day long in real-timeEach source goes through a cycle. The learning phase will "end" if a source reaches its "learning budget" or has conversions.
If a source is not performing well, the bid will gradually decrease to almost 0 (you will see a lot of 0.000001$ bids) - that is equal to pausing since it won’t be able to buy traffic with such a low bid.
Note we never fully pause a source, just lowering the bid to a minimum as possible - giving it the option to reawaken if a late conversion appears.
If a source has conversions, we calculate the optimal bid according to CPA GOAL, number of conversions, etc...
So the optimizer pauses underperforming sources and optimizes bids for the ones that had any conversions at all in a way that keeps them within the CPA GOAL.
The range of min-max bid gives the optimizer the chance to raise the bid in case a source is performing well, and we should increase the bid to gain more traffic to withstand the competition. In many cases, most sources won’t reach the max bid if not needed.
So our recommendation is always to keep at least three times between min and max bid to have enough range to play.
In many cases, you will see bid being lower than the min bid in sources on optimizing phase if needed to stand the CPA GOAL. But never higher than the max bid
NICK: At what point would a source have its bid reduced to almost 0? At the end of the learning phase, with no conversions, or later on in the optimization phase over time?At the end of the learning phase, with no conversions or if the cost per conversion is way too high, it will gradually reduce the bid to try and remain in the CPA goal, and in some cases, bids can go down to a very low bid (0.000001$).
NICK: Is the bid optimization only considering the source as a whole, or does it consider other things within the source (i.e., browser, OS, etc.)?Yes, all targeting settings. Device, OS, Browser, etc.
The only thing not considered is the ad creative for push campaigns (see the next question).
You will see an average bid on the platform, but backside, there are many different bids per each of the various targetings.
NICK: Are bids also adjusted for each creative in each source? For example, If I had three creatives and only two of them convert and the third one doesn't, would the 3rd creative's bid get dropped so low it gets no traffic?No, not yet. It is up to you to disable creatives yourself.
NICK: Do bids get optimized per segment within the source then?Yes, that is the best way to optimize with SelfAdvertiser. And it happens all day long, in real-time.
NICK:
I know! That’s why we love it so much
. NICK: I'm really curious about the 25% of the CPA goal for the source's learning budget recommendation... Why is this what you suggest if the source will essentially be paused at the end of the learning phase without conversions? What if, after 50% of the CPA goal, there could be a conversion? Won't a 25% source learning budget cause me to miss out on a lot of potential volumes?You are right, and that is why on a KW campaign, we recommend on 1*CPA GOAL and, in some cases, higher.
But on a RON campaign, there could be many sources - thousands in many cases.
Let’s say your cap goal is $4. if you spent $2 on each source, or even more before pausing (which in some GEOs is a whole lot of clicks), you will spend a lot of money on learning.
Since RON campaigns have many sources, we believe it’s better to eliminate the non-performing ones quicker and then going through more potential traffic so you will find the best ones for your offer.
NICK: Is this "magic" auto-optimization effective on multi-geo campaigns? No, the magic is only on a single GEO campaign since it considers all other settings, and the performance per source per geo could be very different, it could damage performance.
JIMMY: Do you have any recommendations for the CPA goal ratio - e.g. 80% of payout, etc.?I always recommend that the overall CPA GOAL should be the payout- % Profit margin you would like to keep.
Let’s say your payout is $10 and you would like to maintain a 20% profit margin.
CPA GOAL = $10*(1-0.2) = $8
CPA GOAL is only for optimized sources, meaning sources that are converting. So, setting the CPA goal too low to try and avoid losses, will cause the optimizer to lower bids in performing sources, sometimes unnecessarily.
JIMMY: What kind of spend is the algorithm going to allow before pausing a source (e.g., 2x/3x payout, etc. with 0 conversions)? What about sources with conversions (i.e., what kind of CPA will it allow it to get to before pausing)? I’m curious to know how aggressive it is (e.g. with Propeller CPA Goal, I’ve found I’m much more aggressive than the algorithm, so I still have to do some optimization)...We need to divide it into 2 parts- the learning phase and optimizing phase of a source cycle:
Learning phase - the system will pause a source (reduce bid to almost $0 once it reached the learning budget and didn't have any conversions). So, in case the learning budget is $2, it will be paused at around $2 spent.
Optimizing phase - sources get optimized all day long, through all targeting settings, taking into account a long period of statistics.
If a source is on average doing good in comparison to the CPA GOAL, the optimizer will try to buy more of it.
If a source is underperforming, the optimizer constantly calculates the CPA for all targeting settings, and you will see it will lower the bid and stop buying traffic.
We don't pause it straight away if the CPA GOAL is let's say $8 and the source is at $9 CPA. we try to optimize and reach the $8 CPA with a lower bid, and if we “don't succeed” we again lower the bid and try.
This happens all day long and if a source does not reach the goals it will be paused.
SERVANDO: If we make changes in the campaign during or after the testing phase, does it reset the algorithm and start from scratch? What if we do some optimization (e.g. blocking a browser version, ISP, or OS) on our end after the campaign has been optimized, is it recommended?No, making changes to the campaign settings during the testing phase does not reset the algorithm and does not start from scratch.
A couple of examples:
- If you change the learning budget from $4 to $2, sources that already reached the $2 learning budget will be paused immediately (rather than reaching the old $4 learning budget) and sources yet to reach the $2 learning budget will be paused when they reach it.
- If you change the CPA GOAL the optimizer will optimize from that moment on according to the new CPA GOAL. let’s say you change it from $10 to $8 since you want to be more profitable- you will see bids being lowered for sources above $8 CPA. It could also work the other way around if you want to grow and buy more traffic from the performing sources - rasing the CPA GOAL from $8 to $10 will help.
SERVANDO: Are there any recommended settings for traffic distribution and frequency capping while using Auto Optimizer?Yes, the default settings when opening a new campaign is our recommended one, at least to start with.
You can see that for RON campaigns we recommend 1-12 hours (frequency) and evenly distribution of the budget.
** Important to note that if you set up a campaign with a really low budget- $10 for example, the $10 budget will be divided throughout the day to less than 50 cents per hour.
That is why we recommend at least a $20 - $50 daily budget.
For KW campaigns we recommend 1 - 30 minutes (frequency) and as soon as possible, and that is because on a very targeted campaign, we want to try and buy all available traffic for the bid ranges you set. If you would set it on evenly, for example, you could “lose” potential traffic because of the hourly budget.
----------------------------------
If you have ANY questions about SelfAdvertiser's Auto Optimizer feature feel free to ask them here in this thread. @SelfAdvertiser would be more than happy to answer them all for you.
⬇⬇⬇⬇⬇⬇⬇⬇⬇⬇⬇
