1 Geo, 1 Carrier, 1 TS, High XX/Day Profits - Part 3 (1)
1 Geo, 1 Carrier, 1 TS, High XX/Day Profits - Part 3
[Update Aug/2018: It has come to our attention that this traffic network may have closed down - please hold off on joining until further notice.]
(This is Part 3 of this case study. Part 1 can be found here.)
28-Aug-2016:
I find that on wwwPromoter, testing staggered bids simultaenously will not give good results.
Just to illustrate this, I've done a quick test to show some stats. In the original camp I was bidding $1.33CPM. I set up bids for $2.33, $3.33, and $4.33 and let run for around 1-2 hours:
A couple of observations here:
1)The high-bid camps basically yanked all the traffic from the original camp. This means testing staggered bids simultaneously is not the best way. Instead, I would increase the bid once every couple of days and compare the amount of daily profits (which is what we're wanting to maximize). Day-to-day fluctuations would affect results for sure, but that's the best approach I could think of.
2)The higher bids resulted in lower ROIs than the original bid. I won't bore you with the actual calculations and stats - will just list the ROIs associated with each bid for your reference:
Bid=1.33 - ROI=192%
Bid=2.33 - ROI=176%
Bid=3.33 - ROI=20%
Bid=4.33 - ROI=55%
These numbers are telling us that bidding higher may get us lower ROI, but remember that we're trying to maximize DAILY PROFITS and not ROI. Even if a higher bid does result in lower ROI, the extra traffic volume brought on by the higher bid may still result in more daily profits. We shall find out for sure later when we actually increase the bid.
Decided to test bids by gradually increasing the bid in the original camp instead later on (as described above).
So I killed the higher-bid camps. Would continue to run the original camp and cut placements.
1-Sept-16:
So the Dialog camp had been running the best offer for several days without any changes. Time to cut placements! This was what I did:
Placement-Cutting Step 1: Cut Unpromising Channels
In
As you can see, Channel 22748 was just bad news period.
Channel 20578 only had a couple of placements in profit and most of the placements had 0% CTR, so I cut the entire channel.
Channel 20557 only had several placements, and one was doing profits, so it was a borderline case - could cut or not cut the entire channel.
Channel 9989 only had 1 major placement and was in loss - so another borderline case - could decide to keep and collect more stats for the smaller placements.
So to summarize: I decided to cut channels 20578, 20557, 22748, and 9989. To cut channels or subids in wwwPromoter, you'll need to go into "Edit" mode for the campaign, click on the "Traffic Sources" page, and paste a list of channel and/or subid names into the "Exclude:" field:
MAJOR NOTE: Don't go blacklisting these channels for all your camps now! A channel and subid can perform differently for different geos and targeting. I've seen channel 9989 perform very well for some geos and targeting. So do your own testing!
Placement-Cutting Step 2: Cut SubIDs with 30 Impressions But Zero Clicks
Next, I drilled down to just SubIDs in
Basically with the average CTR being as high as it was, if a placement sent 30 impressions without a single click, I could be sure something was wrong with that placement.
Why 30? Because I understood from this post that 30-50 impressions should be large enough of a sample size when trying to determine CTR.
So I opened the exported SubID stats in excel, sorted by number of clicks in ascending order, then highlighted all the placements with 0 clicks and sorted by number of impressions, then cut all placements that had over 30 impressions.
(BTW - I've since made a tool to recommend the best CTR cut-off values for cutting placements. You can find it here.)
Placement-Cutting Step 3: Cut SubIDs in Loss by More Than 2 Times the Payout
Cutting placements in loss by more than 2x the payout has become one of my favorite rules of thumb. Why 2 times? It's just a number that has worked well for me personally. You can set it to a larger value to be more certain you're not cutting potentially-profitable placements, but of course by getting that extra accuracy you'd have to spend more money to gather data. Again, this tool will help you to cut placements that are in loss by X amount of payouts.
In order to apply this rule of thumb, you'll need both placement revenue stats from Voluum, and placement cost stats from wwwPromoter. I've already exported SubID stats from Voluum as a CSV. To get cost stats from wwwPromoter, log into wwwPromoter, click on the "Reporting" tab and select "Channel Sub IDs", then specify filtering options and export data as below:
Rules of thumb will not give you statistically significant results. I would not use them for cutting important campaign elements like offers or landers, or even big placements. If you want to cut placements accurately using statistics, this would be the tool to use.
The drawback of using this spreadsheet is you would need to spend more money to collect more data before statistical significance can be reached. To me, using stats to cut thousands of small placements is overkill. Given how short-lived many pop camps are, we don't have the luxury of spending weeks and months to wait for statistically-significant data when testing thousands of placements.
Summary and Next Steps:
Using the criteria described above, I blacklisted a total of 4 channels, plus 219 placements.
I wanted to wait a few days first to observe the effects of cutting all those placements. After that I would start testing bids.
Please continue to Part 4.
Amy
Home > General > Affiliate Marketing Forum