Home > Vendors >

---

For Trafficsources: My wish to improve auto optimizations (10)

---

I started running Smart CPA campaigns in 2017 when I switched to Pop traffic and in the last years I ran many Pop campaigns with some kind of auto optimization, let it be Smart CPA, CPA Goal, Target CPA, just excluding of non converting placements or whatever it´s called.

When I started running Push traffic I also used CPA options if possible and in all the years and hundreds or even thousands of campaigns I noticed few things that I would like to talk about.

In theory the concept of such campaigns sounds very good: Create a campaign, set a price per conversion that you are willing to pay, let the platform do all the testing and optimization -> PROFIT

Unluckily in reality it´s not that easy.

It´s true that running campaigns with auto optimization can be a huge timesaver and they can also bring in some nice profits but in no way they are guaranteed money makers.

In worst case you can lose there even more money than you would lose from traditonal CPC/CPM campaigns where everything is under your control.

The huge downside is when the algorithms don´t work reliable enough, then such campaigns can also result in high losses, especially when you run many of them.

Generally we can say that campaigns with auto optimization follow the same stages as manual campaigns: First testing, then optimizing and scaling.

The stage where you can lose most money is the first one, the testing.

Usually this stage will show already if a campaign is worth it to continue or not and if it´s not performing good enough it should be stopped.

When you have good campaigns the losses from the test stage are not that important because you can make good for it many times in during the lifetime of the campaigns but when you have bad campaigns then the losses should be kept as low as possible because these campaigns will never make good for that losses ever.

A good moment to stop campaigns that are not performing good enough is when you received enough traffic to make a data-based decision but where you didn´t overspend on that campaign yet.

And this is where many trafficsources could and should improve their algorithms to work more reliable to prevent unnecessary losses especially in the test stage.

From what I noticed there are mostly 4 factors that are resulting in too much spend and losses.

1) Starting the campaigns on RON
2) Keep non-converting campaigns running for too long
3) Sending too high volume right away
4) Don´t exclude non-converting placements fast enough

Before I go through the different factors let´s talk about the test stage of a campaign in general.

The test stage is not there to be profitable right from the start, the test stage is only there to test if a campaign has enough potential to keep it running or not.

If you want to find out if a campaign is worth it or not the best way is to run the test on high converting placements because then we already eliminate one element from the test: Non-converting placements.

And this is what the first factor - Starting the campaigns on RON - is about.

When I start a manual campaign I also use a WL when I have one, that way I can test a campaign on traffic where I know that it converts.
That´s faster, cheaper and more effective than starting on RON.

Trafficsources with auto optimization features receive postback infos from many many thousands of conversions each day so they should have whitelists for basically every country and every targeting.

The tests would be much more effective when they would use such WLs and in case that the campaigns don´t convert the tests could be stopped much faster with lower volume so that we don´t lose too much money when it´s not worth it.

It still happens more often that trafficsources start campaigns with auto optimization on RON.

There we come to the second factor to keep non-converting campaigns running for too long.

When we see that a campaign has no potential to be a winner we should stop it, there´s no need to spend more money on campaigns that will never have a chance to become profitable.

Unluckily there are trafficsources that don´t stop campaigns when they are clearly bad, they just keep them running.

Even when they "optimize" the campaigns then on a placement level it still doesn´t help to keep non-converting campaigns running, it only costs more money that could be spent better otherwise.

The auto algorithms should not only check if a placement is good or not, it should also check if the campaign in general is worth it or not and if it´s a dud it should be stopped.

Often factor 1) leads directly to factor 2) because when you start a campaign on RON it´s just normal that you have higher costs because you also receive traffic from more placements.

Nonetheless spending $100 on lots of mediocre or bad placements with few good ones in between where all placements just receive some traffic will mostly tell you way less about the real potential of a campaign than spending half of the money on few proven and high converting placements.

Factor 3 - Sending too high volume right away - is somewhat related, when a trafficsource sends high volume it can also be way too expensive, no matter if the placements are good or bad.

When you don´t know if a campaign is converting or not you don´t go all in, you first test the waters and if you see signs of success you increase the volume.

You don´t want high volume when the campaign sucks, you only want high volume when more volume makes more money, not less.

Sending the traffic in batches would help alot to keep the volume and consequently the adspend under control.

Send first batch traffic - wait for results - send next batch - wait for results - [...repeat the steps...] - when campaign is converting increase volume, when campaign is not converting stop campaign.

Factor 3 is especially important on Push traffic where you often get late clicks.
The higher the volume the higher the amount of late clicks the higher the spend.

Factor 4 - Don´t exclude non-converting placements fast enough - is pretty obvious, there´s no need to let placements run for too long when they just don´t convert.

This factor mostly happens when the trafficsources don´t check for performance often enough and it´s the only factor that is probably more important in the long run whereas the other factors are more important in the test stage.

Checking every 30-60 minutes can already result in high losses, especially when they send too high volume (from non-converting placements) right away and when you run many of such auto optimization campaigns at once it can easily end with few hundred Dollars per day in loss.

Factor 4 could easily be avoided for the most part when factor 3 wouldn´t happen

You see, all of these 4 factors on their own can already result in more losses than needed but often you will see a combination of two or more of these factors, some of them are often also pretty tight connected.

In the end it could all be avoided pretty easy and in the long run it would probably also mean more profits for the trafficsources because more affiliates would run auto optimization campaigns then.

Here is how a good auto optimization algorithm should start and run a campaign:

- Start traffic in batches of high quality placements instead of sending RON traffic on high volume right away.
The traffic can also be more expensive, to see if a campaign converts or not I definitely say quality is more important than quantity.
- Wait for results between the batches to have more control over the volume and spend
- If the campaign doesn´t convert on high quality placements stop it.
Non-converting campaigns don´t suddenly turn into winners just because we keep them running for longer.
- Control the traffic volume and/or increase the check frequency to exclude non-converting placements before they overspend too much
- If placements convert good enough increase volume there gradually and adjust bids over time

That system is also not perfect and when a campaign just sucks you can send whatever traffic you want and it still won´t convert - you can´t polish a turd

When you run enough of such non-converting campaigns it can also result in high losses because of the high bids in the beginning.
When the campaigns don´t convert and can´t make good for the initial losses it can add up to pretty high losses as well.
The difference is that then the reason is not that the algorithm isn´t working good enough, then the reason is that the campaigns are not good enough.

I really hope that some trafficsources are reading this post and use it as inspiration to improve the auto optimization algorithms.

Let me know what you all think of it, maybe you noticed the same things as well.

---

Digesting your post with all its information took me quite a while.

The points that you have mentioned made me hate several auto-algos. Worst was when I could not even see the usage of the defined test budget in real time.

I had hopes that limiting the impression per zone during test phase would help. However some sources need 20min and more until traffic is actually stopped (even on pops). So in the end you can only choose how much blood you want to lose. Problem would have been less bad if the source did not place all its shice* traffic in front of a fan pointing to me. That made me really angry. From pre-tests I knew that that particular source had way better traffic.

Sometimes I think that it would have been way better if I just could use my initial BL camp and let the source do the rest of the optimization + finding additional zones on higher bids slowly. In a way that would be like a cpa camp and you could prove with a prior smart cpm camp to the source that your offer is actually good enough for real cpa.

Coming back to overspend on single zone level...

I have seen it (on one source) several times that single zones had massive overspend during test phase. So it happened that one zone was 5-10 USD in the reds for a 5ct offer. However after several days of algo magic the zone was in deep green totally. Only kept it running because the camp as a package was improving steadily in healthy intervals.

One more thing...

For some offers (e.g. IVR) it can be really good to get a massive initial wave of excellent traffic. For most other offers a wider and slowlier spread would indeed be much better.

Probably the very best would be if you could define the test algo's parameters like:

- budget per hour
- share of premium zones
- max loss per zone
- variance of bid
- dayparting or even distribution of traffic on zone level (so that you only catch the first 2 hours of a day for a zone for your budget)


------------------------------
* shice is a pseudo english. Germans use it a lot. They just write it differently

---

Sure it would already help alot when you use your own BL or even WL but from my experience the result is mostly that you don´t receive enough volume to be worth it.

The "best" way to run CPA Goal/Smart CPA campaigns usually is to start them RON so that the trafficsource sends enough traffic.

I can imagine that it probably sounds a bit strange when I said in the previous post that such campaigns shouldn´t start on RON but it´s a huge different if you start the campaign on RON = no limitations from BL/WL and the trafficsource sends RON traffic or if you start the campaign on RON = no limitations from BL/WL but the trafficsource first sends high quality traffic before it opens up the gates to RON traffic.

Yup, that happens more often that you overspend (alot) on some placements during the test stage and later these placements are deep in green.

There it can make a huge difference why the placements are in loss during the test.

To make it clear what I mean here are two examples why placements can overspend alot during the test and how it can lead to completely different results.
Let´s assume the campaign itself is converting good so that it´s only about the placement level:

1) You have a very high quality placement with very high bids so that you receive enough traffic from that placement to spot fast if the campaign converts or not.
When the campaign keeps running and this high quality placement keeps bringing conversions then the bid gets adjusted and after some time it can be a very profitable placement although it was deep in red before.
In this example the reason for the overspend is that you paid high bid prices during the test but in turn you receive high quality traffic that brings conversions and that´s what we want for our tests.

2) You pay normal or low bid prices for a placement and receive high volume without any conversions.
The algorithm doesn´t stop the placement and just keeps it running what sooner or later results in the overspend.
In this case the reason for the overspend is that the placement just keeps running although it doesn´t convert.
Here it´s pretty unlikely that the placements suddenly becomes profitable no matter how long you keep it running, usually losers don´t turn magically into winners just because you give them more time.

You are absolutely right and when I was writing on the post I also thought about exactly these IVR offers.

For such offers where the converting timeframe sometimes is only few hours it can definitely help to get a traffic tsunami from the very beginning because it can already be too late when the scaling starts 1-2 days later.

In the end we we can´t consider every single offer or vertical with its quirks, this would just make everything too complicated.

There it´s much better to agree on a broader common ground that works for most campaigns.

Also there is still a big difference between "a massive initial wave of excellent traffic" and "a massive initial wave of RON traffic"

This would be fantastic if we could also set the parameters for such campaigns.

Let´s keep fingers crossed and maybe one happy day in the future it will be possible

---

Great post @twinaxe! @larsometer I'm loving you guys' discussion.

We should start a skype group with all the traffic source reps from all the major networks and have a discussion with them as well.

However - looking at this topic from the point of view of traffic sources - I'd imagine the dilemma they must be faced with.

On the one hand, by implementing the auto-optimization feature, they can attract more advertisers to join their network, and also increase loyalty in existing advertisers by helping them lose less money or make better ROI (potentially).

However, if they make auto-optimization TOO efficient, they could make less money from advertisers. (As advertiser lose less money to them.)

I'm guessing this may be why traffic sources aren't actively trying to improve their auto-optimization algo? (Not a rhetorical question - this is really just my guess.)



Amy

---

I already talked about it with one of my AMs few months ago and he was absolutely happy that I explained him in detail my opinion about such algorithms.

Of course the trafficsources will probably make less money per campaign when they work too good but it still doesn´t help when the advertisers then don´t run more auto optimization campaigns because they lose too much money there.

Longterm it´s much better for both sides when the algos would work more reliable.

It´s basically high ROI on few short-living campaigns vs lower ROI on many long-living campaigns

---

Saving advertisers money can potentially be a win-win. I completely agree with this.

It's HOW MUCH they want to perfect the algorithm in the advertiser's favor that is in question. And not every traffic source has the long-term vision that is required, in order to give up short-term maximizing of profits.

Another factor to consider here is that networks also want to please their publishers, as that's where they get their "merchandise". The better the algo is at saving money for advertisers, the less money publishers will get - especially the ones with lower traffic quality.

I can think of 2 solutions to this:

1)Reject pubs with low traffic quality.

2)Make the auto-optimization algo more efficient (like you guys suggested above), so the lowest-quality pubs would make so little money they'll quit the network.

I'm not sure how it's actually done at the networks - each network probably has its own way of dealing with low-quality pubs - or not.

The cons: The networks has less traffic to sell, which translates into less profits (short-term anyway) and lower volumes (which can make the network less attractive to advertisers).

The pro: The overall traffic quality of the network would increase.

Again - it takes a traffic source with long-term vision to give up short-term gains (no matter which of the 2 options they take).


While it's really interesting talking about this among ourselves, it would be great for reps from traffic sources to give their perspective. Although - it may be difficult for reps to be candid when discussing a sensitive topic such as this.

Let's try something: I'll start a subforum where traffic networks and affiliates can come together to figure out how to improve traffic source platforms to create win-wins. Maybe do the same for affiliate networks too. Let's see if we get any traction - but like I said, I really don't know how candid reps will be, but it's worth a try!



Amy

EDIT: Subsection set up - @twinaxe hope you don't mind me moving this thread into the new subforum!

---

@twinaxe

I like the ideas on better auto optimization. I faced this issue in my past and actually figured out a better way (at least I thought so).

I was sort of explaining this to you a little on the follow along about my past, but it got lost in a slew of other stuff and I didn't want to bog down the forum follow along. I built a desktop application (my skills actually started here anyway) that did this exactly with Propel (trafficvance). I built it standalone of a traffic source so that it could do the same management on ALL traffic sources I fed it. It handled everything from the tracker to the traffic source. Paused things at the cut point I preset, cut things when I said, started new campaign when it found winners, duplicated tracking, managed spreadsheets, offers, etc all from a desktop application. I just adjusted the settings for how often to check, when to cut, etc based on how I managed things. It was like having a full time employee that managed campaigns like a crackhead on steroids.

I am already algorithmizing out how to do it again with push traffic. I am getting ahead of myself as I am still figuring out push traffic but when I do, the first thing I am gonna do is build this tool and it will save a lot of time like it did with PPV. The advantage of building it standalone is that:

1. You set all the parameters based on the WAY YOU TRACK, NOT SOMEONE ELSES AUTO ALGORITHM.
2. My programs use BOTH API or frontend (uses the site like a human) so it has fail safes if one is down or if the source doesn't offer API access.
3. It can run all day everyday on a dedicated computer.

There are way more features it had for my PPV management but I won't go into it here cause it would take awhile.

Anyway, just my 2 cents. I am building another application just like the one I made for Propel many years ago. As my push traffic campaigns expand (launching another 3-4 more tonight, updates to come in the follow along) I will let you know eventually how it goes. But maybe you guys are looking the wrong direction. Don't wait for the traffic source to do it, take matters into your own hands and you do it first. I just have to get some winner campaigns in push damnit so it warrants me building another one from the ground up. Building a program of this magnitude usually takes me about 3 weeks so getting campaigns green is the major thing for me at this point so that I WANT to build the other side. Not conversions, I can get those, getting it TO GREEN.

At least that's my plan once again. Food for thought.

---

I totally agree with it, this also gives you the opportunity to have the whole automation stuff customized for your very own needs.

I started working on scripts to automate my campaigns for exactly that reasons myself but that´s not the point of this thread.

Fact is that there are trafficsources who offer to run campaigns with bid options where the platform takes care about the optimization and scaling and here I wanted to point out a few things how they could improve such features.

Again, the thread is about auto optimization features on trafficsources so it´s definitely not the wrong direction

Creating auto optimization solutions ourselves is great but it´s a completely different topic.

Not everyone is good in coding and not everyone wants to spend the time on creating such stuff.

The statement is same as saying

- Don´t wait for trafficsources to send good traffic. Take the matters in your own hands and build your own trafficsource.
- Don´t wait for CPA networks to provide you with good offers. Take the matters in your own hands and create your own offers.
- Don´t wait for trackers to implement all features that you want. Take the matters in your own hands and code your own tracker exactly the way you want it.

---

This is a good point. I also understand you meant it specifically for traffic sources and that THEY should do it better. I just always hated waiting on stuff like this cause it seems like it never materialized for me.

Even if you are good at coding is still doesn't mean you want to spend the time . Just saying that if you can, sometimes it's better to do it your way versus wait around. Anyway, all good points and I understand that what I was saying was probably better in a different thread. Trafficsources would be smart to listen though and get better at this internally. Would make many more affiliates want to use their service.

---

Exactly that´s what the thread is about, when they already have such a feature and it´s not working good enough then they shall do better

It´s not about saying "Hey trafficsources, implement some good auto algorithms that make my life easier", it´s rather like "Hey trafficsource, you already have an auto algorithm but it´s not working reliable so please improve it".

That´s 100% correct and I do it like this myself but as I said, that´s a different topic.

In the end the example above can be applied to so many other things as well.

For example I don´t like how bakers make bread so I bake it myself.

Nonetheless in general there´s nothing bad in buying a bread from the bakery.

---

Home > Vendors >