Attention: This post has been updated. Original conclusions should be treated as invalid. Affiliate Network names have been removed.
This may be surprising to the full ten subscribers that read this blog, but this will be my second post this month!
Okay, if you’re not split-testing your offers with multiple affiliate networks, you are a fool — simple as that. So, recently I just ran a quick test over a month (of October) for both Affiliate Network X and Affiliate Network Y. I picked one offer that they both had with an equal payout. I then waited until I had enough data to be statistically accurate.
Here are the results:
Network Y: EPC: $4.00
Network X: EPC: $4.62
Results had an approximate 95% confidence–which would be valid in the view of most marketers. My traffic did not go directly to these networks. I had a system of two landing pages that each user had to go through to eventually reach the affiliate network.
To calculate the percent confidence, I took the traffic that hit my first page (and put this number in the “impressions” boxes), then the number of conversions (I put in the “actions” boxes) for each corresponding affiliate network.
I’ve had some great success with Network Y in the past. So, this could just be because of inaccurate tracking or advertiser issues.
Update: During the months of October and November, Network X converted exceptionally well. However, this month (January 2009) conversion rates have went down significantly. So, in conclusion, you need to split-test more than just two networks, and make sure you consistently track changes in conversion rates.
Update 2: I’ve gone direct with the advertiser, and conversions are now better than with Network X. 🙂
Update 3: I recently had a discussion with my Network Y AM. I doubt Network Y is scrubbing. I believe it was more of an issue with the advertiser. I had an issue with one advertiser (state insurance submit) that was converting around 60%, but then gradually reduced to a 10% conversion rate.
Hopefully, final conclusion. Test always. 95% confidence is not enough to judge an entire network.
Leave a Reply