Pop Campaign Optimization - Some Considerations (21)
This post was originally to be a response for member conduit's question here:
http://stmforum.com/forum/showthread...cess-look-like
He asked a really general question, but it's a very important one.
To conduit: The optimization process may not be as straightforward as you may think. It would be a lot easier if you could start a follow-along or otherwise provide a specific example of a campaign, because each case is different.
What I WILL do is list some of the factors and considerations to start the discussion and hope that others will join in. I'll also update this post will additional thoughts later on so I'll have something to refer newbies to.
The "Gist" of Optimization:
Basically there are 2 components:
1)Optimize your funnel: i.e. test offers and landers. And of course new angles will also drive lander optimization.
2)Optimize your targeting: i.e. target certain OSs/carriers/browsers/placements/time of day.
As long as you're testing new offers and landers, and cutting using stats tools, the funnel will have a chance to improve.
Optimizing the Funnel:
A)Test More Offers
For your chosen vertical+geo, test offers recommended by AMs first, then the remaining offers. For tips on offer selection:
http://stmforum.com/forum/showthread...-AM-FAQ-Thread! (look for section on offers)
If you need more offers, join more affiliate networks:
http://stmforum.com/forum/showthread...-First-Network
Of the traffic available on a given traffic source, there are different traffic segments you can either choose to target or not.
Usually, the major traffic segments to focus on are OSs (e.g. Android and IOS) and Mobile Carriers / Wifi if you're running mobile. Adult vs. mainstream if both are available on the traffic source. There are other major traffic segments depending on your traffic type/source (e.g. straight vs. gay for adult).
Your ultimate goal is to find the best offer for each traffic segment in order to maximize your profits. (e.g. Some offers convert best on Android+CarrierX, while others convert best for Windows Phones, while others do best on wifi.)
Suggestion: Set up separate camps for offers that accept 1)carrier traffic, 2)wifi traffic, 3)adult traffic, 4)mainstream traffic. Offer types and nature of traffic are different, and bids are usually different, so better to keep things separate. Some offers can be added to more than one camp (e.g. OfferX which accepts carrier+wifi, mainstream+adult).
Try to test multiple offers from the beginning to cast a wider net. Can set up paths in the tracker to group very similar offers together and run different landers in every group (e.g. path 1 for iphone 6 offers, path 2 for iphone 7 offers etc.). Can set up tracker rules to send to each offer the traffic types it allows (e.g. some offers only accept certain OSs/Carriers/devices). Alternatively, can set up multiple camps to target separate OSs each, or separate carriers, to test offers separately for each OS and carrier, but you may end up with a lot of camps to manage - for big geos with huge traffic this may be worthwhile.
When running pop traffic: One tactic is to run aggressive because if you don't your competition will (please do so at your own discretion because there will be risks of not getting paid by the advertiser, and/or getting banned by the traffic network). You can either cloak, or just run offers that allow aggressive promotion (ask AMs for offers that don't require lander approval and allow aggressive). Again - I'm just listing possibilities - you've been warned about the risks.
To cut offers in split-testing - use the tool described in this thread - it can compare offers with different payouts:
http://stmforum.com/forum/showthread...ou-pick-offers
For mobile, some offers will convert well for wifi and others will convert well for carrier, and some will convert well for both. A good strategy is to test offers separately for wifi and carrier. Please see "Carrier vs. Wifi" in the "Optimizing the Targeting" section below.
B)Test More Landers
You'll find lots of lander-related tips here:
http://stmforum.com/forum/forumdispl...-Landing-Pages
http://stmforum.com/forum/showthread...-AM-FAQ-Thread! (section on landing pages)
One strategy is to rip landers to establish some benchmark stats and to find a converting offer, and when things look promising, can devote more time towards testing custom angles (custom lander text and/or theme). This spy tool is recommended:
http://stmforum.com/forum/showthread...e-STM-Discount
Also - test more angles, as angles are what will drive your landers. There's an entire forum section on angles:
http://stmforum.com/forum/forumdispl...can-post-here)
Lander loading speed is especially important. Anything longer than 2s and visitors will click back in droves. See this thread for great tips on optimizing lander speed for pop:
http://stmforum.com/forum/showthread...ter-Cheatsheet
Please see the note in the "offers" section above regarding running aggressive on pop. The same applies to landers. Make sure your traffic source allows aggressive unless you decide to cloak (please note my disclaimer on risks of getting banned, above).
Post on how to split-test landers:
http://stmforum.com/forum/showthread...Banners-Part-1
Optimizing the Targeting:
-First thing to check is what kinds of targeting options are available for your traffic source. Examples:
Bid
Frequency
OS
Device
Browsers
Wifi vs. Carrier/Cellular/3G
Ad format (for pop: popunders, popups, redirects)
Placements
Time of day / day of week (dayparting)
-Bid: I usually use the "staggered bids" method - set up 3 or more of the same camp and assign different bids to them (low/avg/high), assign a small budget to each, and run them simultaneously to see which one yields the best ROI. Then I would use the winning camp to optimize the funnel. This way I test at the highest ROI to save money.
Know though that the bidding landscape is changing all the time with changing levels of competition. You may get outbid (characterized by loss in traffic volume and decrease in conversion rate) in which case increasing your bid may be appropriate.
Also keep in mind that if you identify specific traffic segments that convert well - e.g. OfferX converts well for Android, or OfferY converts well for CarrierX - that you'll want to test bids again for the new targeting options. Carrier traffic will cost more than wifi traffic for example.
Good posts on bidding:
http://stmforum.com/forum/showthread...-AM-FAQ-Thread! (look for "how high should I bid?")
-Frequency: I usually set it at 1/24 to start. The general trend is that the higher the frequency, the lower the conversion rate, but the higher the traffic volume.
So if your camp isn't green at 1/24, there'd be little point in setting a higher frequency.
However, if your camp's doing high ROI, consider testing higher frequencies to get more traffic - you may still end up with more profits in spite of the lower conversion rate. It would be like selling more goods at lower margins to end up with more overall profits.
-Ad formats for pop: Popunders usually have the most volume, so I'd just do my testing on popunders and scale to the other formats once I'm green (popups, redirects etc.)
-Placements: Tutorial on cutting placements:
http://stmforum.com/forum/showthread...Banners-Part-2
Here are some suggestions on different approaches you can use when cutting placements:
http://stmforum.com/forum/showthread...l=1#post248980
Also, identifying some of the placements that have a lot of bot traffic and blacklisting them first can save you money - see here:
http://stmforum.com/forum/showthread...ut-Bot-Traffic
-OS: Some offers will convert well only on certain OSs.
Example: If Offer1 converts better for Android and Offer2 converts better for IOS, there's no reason to just look at total conversions and eliminate the "loser" - simply direct (in your tracker) android traffic to Offer1 and IOS traffic to Offer2. Better yet, use separate camps for the 2 OSs and test different bids for each.
You can just target the best-converting OS, or just cut underperforming OSs as you collect data. More details on strategy in the last section of this post.
-Devices & Browsers: Offers and landers can perform differently on various devices and browsers.
If you see a lander perform well for some browser but not for others, it may be worthwhile to check how that lander looks on the "bad" converting browsers - it may be because it's not coded properly for displaying on said browsers, or perhaps certain scripts are not coded to execute properly on those browsers. The same goes for devices. The tool to use for checking how a lander will display for various browsers and devices is https://www.browserstack.com/.
-Carrier vs. Wifi: To run volume, test offers that accept wifi traffic to identify ones that convert well. Alternatively, test one-click carrier-billing offers on carrier/3g traffic - these can be very nice pockets of profits with high conversion rates and margins. Please see post on mass-testing offers for carrier traffic:
http://stmforum.com/forum/showthread...esting!-Part-1
Better yet, do both: Test and find offers that convert well for wifi, and test separate offers for carrier to find the best. When you do this you'll have the option of running on traffic sources that don't have carrier targeting, by setting tracker rules to direct different carriers and wifi traffic to different offers (whereas you can't really do that if you only have an offer that does well for carrierX, but no offer that does well on wifi).
-Dayparting: Usually the worst-converting hours will be when the local people are sleeping or most busy working, such that a)more of the traffic will be bot traffic (my own speculation) and b)visitors will be mostly focused on being sleepy / working and less on surfing the net and least of all reacting to ads.
Some Considerations on Strategy/Approach:
-Drilling down into different stats is important. By drilling down into your stats to identify the good and bad traffic segments for your best offer, you can target the good segments separately and bid higher. Alternatively, just cut the bad segments and leave the promising segments. (By traffic segments I mean stuff like OSs, carriers, browsers, placements, etc.)
-Example: If Offer1 converts better for Android and Offer2 converts better for IOS, there's no reason to just look at total conversions and eliminate the "loser" - simply direct (in your tracker) android traffic to Offer1 and IOS traffic to Offer2. Better yet, use separate camps for the 2 OSs and test different bids for each. The same applies to some of the other targeting, e.g. mobile carriers.
A 3rd approach would be to cut all unpromising segments first, optimize your funnel on the best-converting traffic, and then retest the "unpromising" segments again because then they may be profitable. You can do this if you're working in a geo and traffic source that have lots of traffic (otherwise you won't have enough traffic left for testing offers and landers after cutting so much stuff). Doing so will allow you to optimize your funnel for cheaper. This will work for placements and time of day / dayparting (because an offer that converts well for some placements will probably also convert well on some of the other placements); but the same approach may or may not work for OSs and devices and mobile carriers vs. wifi (because an offer that converts well for Android may not convert so well for IOS, and an offer that converts well on carrier traffic may not convert well on wifi).
-Do remember that many campaign variables are inter-related. For example, when you cut placements, ROI of OSs / carriers etc. will increase. In other words, cutting placements may make previously unprofitable OSs/carriers profitable. For this reason, try not to cut for multiple variables at the same time unless you know what you're doing.
-Bird's eye view on optimization process: Try to get the bigger variables right before tweaking the smaller variables. Think of optimizing a camp as optimizing a machine with big dials for making larger changes and then smaller dials for fine tuning:
What this means is, if your campaign ROI is at -60%, then don't waste your time testing lander CTA variations because that probably won't give you the jump in ROI that you need - you'd just be wasting money testing lander variations for a small improvement in ROI.
Instead, test more offers, or completely different lander angles - because those are the kinds of testing that DO have the potential of achieving large jumps in ROI.
A note about offer and lander testing: A good offer will convert with an OK lander, but not even the best lander will convert a shitty offer. So a good funnel-optimization strategy for pop, would be to rip a bunch of landers and test those first, then use the winning lander to mass-test offers. When/if you find a promising offer, then go back to testing custom landers you created yourself, or landers you've ripped and heavily modified with new angles etc., and/or test lander variations of the original winning lander(s).
Also, by finding a great offer+lander combo first, you'll be able to make more traffic sources profitable when you scale. If you're trying to scale a mediocre offer+lander combo and relying on cutting placements etc. to reach green, you'll spend a lot more money and end up with a lot less traffic volume - and STILL not able to make too many traffic sources profitable.
Based on the same reasoning, try not to cut too many placements until you've found a good offer + at-least-OK-lander combo. Same goes for dayparting. Exception: You want to only target the best placements and hours to test offers and landers for cheap, and then retest the other placements and hours after you have a good funnel.
Your ultimate goal is to reach green AND end up with enough daily profits to be worth your while in maintaining the camp (i.e. a 100% ROI camp won't mean anything if you're spending $1 to make $2 everyday). Each thing you cut will have the potential of saving you ad spend, but will also decrease your traffic volume. Another consideration is that a currently-unprofitable segment may become profitable after you optimize your funnel and/or cut other segments/variables. So, try to base each optimization decision on how it will affect your campaign overall - if your decision is aligned with the ultimate goals above, then it's the right decision.
Other Great Posts on Optimization (that haven't been mentioned already):
http://stmforum.com/forum/showthread...nificance-quot
http://stmforum.com/forum/showthread...gn-Or-An-Offer
http://stmforum.com/forum/showthread...Your-Campaigns
Hopefully this post will help some of you that are new to pop traffic! I don't consider myself an expert so will welcome any feedback/comments from the veterans. 
Amy
[Reserved just in case]
*Rigorously Takes Notes
Thanks Amy!
Nice post Amy, i have a question.
Currently i am on a landing page cutting process, and i am having the inconsistent data issue.
1 day i have tremendously highly converting LP, but the next day it performs like crap and other one performs better.
Moreover, it gives different type of performance in different set of bidding range, for e.g, some LP performs great in this bid range but it performs crap and another bid range and vice versa.
Usually how do you optimize in this situation?
great write up AMY!
Originally Posted by peter ngo
For the bigger picture we need to ask the question: What type of landers do MOST visitors seem to be responding best to?
Behind tracker stats are actual people, and different people will react differently to different landers.
My guess would be that on different days, you were getting traffic from different categories of placements (therefore different types of audience), thus the different reactions to your landers. I'm also guessing that your lander angles may not be very general/broad-appeal?
Same reasoning for the different bids: At different bids you were getting traffic from different placements, thus the different lander performance.
What I would suggest, is to start testing more landers until you find one that can convert the audience more consistently, i.e. one that will appeal to a vast majority of the general audience.
Statistical tools can only do so much, because they are just that - tools. When your traffic doesn't behave consistently, the results will be skewed when you use stats tools - because the underlying assumption built into these tools is that all impressions/trials are the same.
If you like, you can post actual stats so we can see how "bad" the problem actually is, i.e. what degree of difference you're talking about when you say "tremendously highly converting" vs. "performs like crap". It may very well just be normal.
Amy
Especially for pops and referring to frequency cap, there's another solution to increase volume:
Let's assume you're running 1/24 and profitable. Now you're looking to try to increase the volume by setting the frequency cap to 2/24. In theory a good idea, but there's a bottleneck:
The higher the bid, the 'fresher' the impression you get from a visitor. Depending on the traffic source, cloning a campaign with a lower bid can save you significant ROI percentage to running 2/24 at the same bid.
Imagine the following scenario:
Top bid: you $10
2nd bid: competitor $6
Instead of paying $10 for 2/24, you can clone the campaign to $8 CPM and have (in most cases, this does depend on the traffic source algorithms and the way they distribute traffic) the same effect as running 2/24 for a lower bid with the cloned campaign. Additionally (although this is a very small side effect), you might scare people off to compete directly with you, since 2 campaigns running $10/$8 CPM look more competitive than 1 campaing $10 at 2/24.
Tremendous tip Franz! Thanks!
EDIT: I just reread your post above again - and realize that I've been doing something similar for a while.
I talk about testing staggered bids all the time, i.e. replicate the same camp and set them to multiple bids to see which one will give the best ROI. After a campaign reaches green, I would re-test different bids again, and would sometimes leave running all the camps that are green. Although I know I'd be competing with myself a bit, I also know I'd be getting traffic from different placements as well as different "session depths". And by having multiple instances of the same camp running (at multiple bids) I find that there seems to be less fluctuation in performance.
Amy
What if you just duplicated your camp a bunch of times without changing the bid? Like 10-15 same bid camps. Is that even allowed?
Originally Posted by eddiezhan
Amy
Haven't been here in a longtime. This is an awesome post vortex! Appreciate it!
Hi Amy,
Would like to ask you some questions regarding optimization:
1. For finding the best lander, do you consider how consistent it can generate the conversions for us? or just use the A/B tool and do the split test calculation will do? I have a lander that I think is the best. But lately it does not generate too much of sales. 1 day will be 1 conversion and another day will be 4 conversions. I afraid this is also because of the offers.
2. If I take it as my best lander, as you always say, I should test more offers. DO I need to do anything at this stage other than waiting for the "best" to come out? Do I need to cut and only left 1 offer at the end? or Start building the Lander+Offer+Carrier+placement combo? What is the goal for this stage?
Harrison
Originally Posted by onleads
Sure enough, if you generate more conversions first, over several days, before cutting landers, that will make your decision more accurate. However, that would also require more test budget and time - making you less efficient.
Even by using the split-test calculator, you can never be 100% sure that the "winning" lander will be the best performer in the long run. But really, with so many split-tests you'll no doubt be doing over your AM career, you only need to be right more often than you're wrong, in order to make money.
There will always be hour-to-hour and day-to-day fluctuations in campaign performance. I simply make the assumption that all landers will be subject to the same fluctuations and just trust the split-test calculator when it tells me a lander is ready for cutting. But if you want to be more accurate, by all means collect more data before cutting a lander.
If in doubt, you can always rotate the runner-up lander back in for another test.
But if you asked me, I would suggest to test more offers first before retesting landers I've cut previously.When I mass-test offers using a single lander, I would pay special attention to OSs. I would gather quite a few conversions in order to see a trend - OfferA may convert better for Android and OfferB may convert better for IOS, for example, in which case I would run OfferA for Android and OfferB for IOS. Another example: If IOS is nowhere near breakeven but Android looks promising, then I would cut offers based on android conversions alone.
Same goes for mobile carriers.
I would leave cutting placements last - except the real budget drainers. OR, if the geo has a ton of traffic, I may just pick the best-converting placements and do all my lander and offer testing on those, then once I have winners, open up the rest of the placements.
This type of discussion is rarely satisfying because I obviously can't list all possible scenarios and make suggestions for each.
This is why follow-alongs work so well - we'd only be looking at one specific scenario.Thanks for your questions! See you on your follow-along thread!
Amy
Hi Amy, need your advice here.
What will be the next steps after finding the best lander+offer combo? (Already mass tested offers)
Harrison
Edit: I found the answer here
Originally Posted by onleads
Drill down into OS and carrier to see if anything is green. Also drill down into OS and carrier for the winning lander+offer combo to see if there's any green. If you don't see anything at least breaking even, I wouldn't even bother continuing with the camp. (And of course, check placements to see whether there are big bad performers you can cut to make some of the major traffic segments above profitable, before making that final decision.)
If you do see green/breakeven segments, and if those segments are giving you enough traffic for the camp to be worth your time managing, then you could just run those segments.
For a final boost, cut placements and implement dayparting.
Again, it's difficult to give specific advice without knowing specific details.
Amy
This is something I really needed at this point in my learning curve. One question on this quote:
" Can set up paths in the tracker to group very similar offers together and run different landers in every group (e.g. path 1 for iphone 6 offers, path 2 for iphone 7 offers etc.). "
For
Path 1:
Lander 1 -> Offer 1
Lander 1 - > Offer 2
Path 2:
Lander 2 -> Offer 1
Lander 2 - > Offer 2
Path 3:
Lander 3 -> Offer 1
Lander 3 - > Offer 2
What to make sure this is the right setup and that I'm not hurting a campaign in any way!
Originally Posted by tpm767
In this case though you'd need to cut landers for each offer type separately. It would actually be better to test one offer type at a time, for example JUST iphone 6s offers. I was doing some mass-testing on a new traffic source. It makes the data more confusing when you're testing so many different things all at once. But each to his own!
Amy
This is awesome!
Thank you so much for the information.
"Do remember that many campaign variables are inter-related. For example, when you cut placements, ROI of OSs / carriers etc. will increase. In other words, cutting placements may make previously unprofitable OSs/carriers profitable. For this reason, try not to cut for multiple variables at the same time unless you know what you're doing."
Hi Amy,
Great Post. Just a quick question about how u deal with optimizing multiple variables. it sounds like you would only cut placements first..then wait to see if bad OS or browser improves. but this seems like a very long winded affair. let me know how you deal with this situation. Thank you!
Originally Posted by vtredskins
There are so many different scenarios we could come across when analyzing actual stats, but this is my basic reasoning:
-When cutting anything that is responsible for only a small fraction of the total traffic (e.g. one of the smaller placements), I would ignore the impact this cut will have on the other variables (e.g. OS), because the impact would be small. Cutting some small placements won't be enough to make an unprofitable major OS green.
-When cutting something that is responsible for larger portions of the total traffic (e.g. large placements, a major OS or carrier), I would definitely consider the impact this can have on other variables. For example, if Android OS is almost at breakeven, then cutting a couple of the largest and worst placements may just push it into enough green. This is actually less of a "long winded affair" than you may think, because you only need to assess the performance of the largest placements and cut the worst, then see how the other variables perform without those big bad placements. You don't need to completely finish cutting all bad placements before cutting other variables.
On the other hand, if you have a major traffic segment (e.g. Android) that's doing very negative ROI, and you look at placement stats but don't see some big bad placements zapping a significant portion of your budget, then you know you probably won't make Android green by cutting placements. So you may decide to just cut out Android, without having to cut placements first and then collect more data to see how android will do.
I may be making this more complicated than it really is. Basically I'd cut the stuff with hopelessly negative ROI first (knowing that cutting other stuff would probably not make them green), then collect more data before cutting stuff that's closer to breaking even.
Will stop here before I complicate things unnecessarily.
Amy
Epic post Amy.
Home > General > Affiliate Marketing Forum