How to Split Test Offers Properly? (5)
Hi Guys!
I have a question regarding how to properly split test offers. Let's say I want to split test 10 offers, do I need to use only one winning lander against the 10 offers or can I test those 10 offers using say 5 of my best landers? Which would be better and would I get accurate results if I test 10 offers on 5 landers? Can I hit two birds with one stone? Like statistical significance on both the offers and landers at the same time?
Split-testing 10 offers and 5 landers at once will require quite a large budget for you to decide which offers or landers to keep running. It may result that some offers will have a higher conversion rate with some landers compared others. That being said, testing new offer/lander combinations in smaller groups would be best, that way you can avoid cutting good offers based on a "casual" conversion.
Originally Posted by wes888
There are a number of different ways you can approach this.
I want to point out that it's not good practice to cut landers based on conversions from different offers - stats calculators inherently assume all conversions to be worth the same and be the same in nature, which is just not the case when you're using conversions from different offers.
So you should use a single offer to test and cut landers. But if you don't have a proven offer in the first place, what do you do?
You could either:
1)Throw a bunch of landers and offers in the a camp, run until one of the offers is showing promise (I like to wait until one of the offers has made 2 conversions), then use JUST that offer to cut landers. Once down to a winning lander, you'd need to retest all the other offers as they hadn't reached statistical significance before.
2)Just use 2-3 semi-proven offers to test landers, cut down to a winner, then use that to mass-test offers. You can get these by asking AMs for offers that they know have converted decently for at least a few other affiliates, or find them based on intel from a spy tool like Adplexity - look for offers that are being run with volume that exhibit an uptrend.
For your reference: I've provided a rough testing outline at the end of this post:
https://stmforum.com/forum/showthrea...l=1#post322026
Hope that helps!
Amy
Thanks Amy!
I am thinking in the same line that the data will be screwed if I test 10 offers and 5 landers simultaneously. Is there a way to make the testing process shorter? i.e. Instead of 2 weeks to test offers and then another 2 weeks to test landers, is there a way to shorten the testing process?
Is there a way to achieve this in two weeks time instead of testing for a month? Thank you again.
Also just curious, are you female? 
Originally Posted by wes888
It's a catch-22 situation in the beginning - basically you need a good offer to test landers, or a good lander to test offers. In the beginning we have neither.
Since there are usually just a few popular landers floating around, I would suggest to cut landers down to a winner first, and THEN mass-test offers.
But which offer would you use to test landers? If you use a random offer and it turns out to have low conversion rate, it would waste a lot of money before giving enough conversions to help you cut landers. In the post above I suggested to either 1)ask AMs for offers that have converted semi-decent or better for at least a few affiliates, or 2)use promising offers identified through spy tools like Adplexity, or 3)throw a bunch of offers in with the landers and wait until one of the offers shows promise and use just that one offer to cut landers (2 conversions would be good; 1 conversion could be a "lottery" conversion). And of course you can use a combination of methods 1-3.
If you can come up with a shorter process than that then I would love to hear about it! I will write that into my lander testing tutorial as well (which is overdue already - aiming to get that out ASAP).
Amy
Home > General > Affiliate Marketing Forum