What's Your Favourite AM "Rule Of Thumb"? (10)
I'm just in the process of writing a rigorous guide to determining how much to spend on your campaigns (a pretty large task involving Actual Math). However, whilst doing it, I've been struck by how many different approaches there are to AM rules of thumb - and how reliable many of them are.
Currently, AM rule of thumb advice is scattered all over the place, so I thought I'd start a single thread to gather it all!
So, come on - what's your favourite short-line-or-two rule that you use to run your campaigns?
I'll start! I still mostly go by the old "4x payout for Web, 7x payout for Mobile" rule for cutting offers that haven't converted yet. I've actually done some statistical analysis on that rule and it's pretty sound - and it's also fast and easy!
What's your favourite Rule Of Thumb?
"100 clicks and no conversion yet? I'll get upset"
Placement optimization rule of thumb is my favourite, since it's the one I use the most by far:
- 2-3x payout spent and -100% ROI
- 4x payout spent and <-67% ROI
- 6x payout spent and <-15%ROI
- 8x payout and <1% ROI
I mostly use the above mentioned, but here is one for killing bots/underperforming placements that are eating up budget:
> 50 to 100 visits & CTR < AVG/2 to 3 & 0 conversions
ahaha a bit complicated, but works for me.
My favorite: Cut placements that are in loss by >= 2x payout. Even 1x or 0.5x if I'm wanting to cut very aggressively, depending on the particular situation.
Amy
It's interesting the numbers are so different for cutting. Caurmen can you speak more on why 7x is statistically better? I run apps and I can't imagine spending 7x the pay out on a 4 dollar install.
@fightingfffreedom - I'm working on a detailed article about all this right now - although it may be a couple of weeks arriving as it's a BIG topic.
First thing: I use the 4x rule for cutting OFFERS that haven't converted at all. I have different cutoff numbers for placements, generally pretty close to the ones listed above.
However, in brief: studying the 4x rule, it turns out if you examine the bracket of results you're testing for statistical significance in, it works pretty well for Web. Essentially, the longer you test the lower ROI potential you'll be able to reliably detect, because your margin of error is narrow. As your spend goes up, the bracket of ROIs that are still statistically likely diminishes, and the lower boundary is always -100 if you're looking at a dataset with no conversions.
However, if you cut too soon, you end up effectively looking for very unlikely conversion rates. If you test something really low, like 0.5x payout, for example, you can only really say "I've tested, and I'm pretty confident this offer won't convert at 500% ROI. However, it could easily convert at 350% ROI". Not too useful (unless that's what you're searching for).
4x, for me, produces a dataset I'm pretty happy with for a reasonable cost - it confirms that the offer will, at best, produce a small loss on the current funnel. At the end of the day, any offer I find that is working has a good chance of generating a lot of money - and even if I have to test 20 offers to find one, the single good offer will easily pay out more than the amount I invested to find it.
It's fairly easy to figure out a rule of thumb to go by yourself - just spend a bit of time plugging plausible costs per click and overall spends for your usual payout range into a spreadsheet, then run the results through a statistical significance calculator to find out what ranges you're testing for.
However, on mobile you've got a lot more confounding factors than on Web. You've got devices and carriers to consider. More combinations means more testing required. At that point, the maths gets super-hard and extremely fuzzy, but a "fudge factor" that seems to work in practise - and I got from talking to a bunch of huuuuuuuuge affiliates way back when - is to simply apply a 7x rule instead of a 4x rule.
(This is further complicated by bad placements. My usual rule there is to trust no data whatsoever on offers until I've run a bot test, and subsequently to trust the spread and pause placements early to get a broad spread of results. If it's a generally good traffic source, I've eliminated all bot placements, and I've got 10 placements that all spent around the same amount, and none of them have yielded a conversion, logic says the problem is probably either the offer or the angle. If I've tried a couple of angles, the odds are the problem is the offer.)
Obviously all rules of thumb are trumped by more detailed personal experience. If you've run a lot of a specific type of app on a specific traffic source or a few of them and you know that any decent offer will convert far before the 7x rule - or if you're running a more narrowly targeted testing campaign with only carriers and devices that you know convert well on this class of app - then your rule of thumb for that class of app will be better for that.
But in general most people kill tests too early rather than too late, so if in doubt it's usually worth spending a few dollars more and/or checking the specific confidence interval to see what your data's actually saying.
Hope that makes sense!
This is what i use for cutting placements based on CTR:
Basically the more it deviates from the green CTR range the less visits it'll take to cut it. If it deviates on the left side, i.e closer to 0% CTR it'll need more time to reach significance but if it is for example 85% CTR and converting range is [2%, 6%] then i will cut at 85% after 10 visits or so. Maybe somewhat inaccurate and i do re-run on some of these sites later sometimes.
@caurmen Did you write up a article on this up? I can't find it if you did...
Thanks
Rich
I don't think that one ever made it out of testing, I'm afraid - the math got too hairy to the point where it felt like it was of very niche interest.
If folk would still like to see the math basis behind cut-off rules, though, I shall stick it back in the pile of to-dos - possibly with less equations than last time I tried it 
Home > General > Affiliate Marketing Forum