The contents of this post teach you the fundamentals of optimizing Cost Per Action (CPA) campaigns. Publishers and affiliate marketers can use this information to generate more revenue from their advertising inventory.
There are two main parts of this post. The description and understanding the analysis metrics, and the spreadsheet.
If you would like to start off by learning the basics of CPA ad units and where they can fit on your website or mobile app, click here to view the original adding incremental revenue with CPA ad units blog post.
Access my hand-built CPA optimization spreadsheet, built for A/B testing your performance ad sales inventory.
To download the spreadsheet, all I ask is that you share this essay on your favorite social network.
This financial model was developed based on years of experience developing sophisticated monetization models for digital publishers.
The main objective of this spreadsheet is to understand & synthesize data from a variety of sources.
Click here to download the spreadsheet, share
with twitter, facebook, linkedin, google+, vkontakte, etc
- Weighted average cpm, cpc, ecpm
- Efficiency metrics
- Clear a/b testing on ad sales allocation
Column A – Allocation is grouped by Country Champ = Champion campaigns, test campaigns go up against already proven campaigns with strong eCPM, EPC and Efficiency (column o)
Column B – Campaign title For the suffix, I usually add 1 word description of the thumbnail / creative. This way you can further optimize high performing campaigns by testing creative differences in an effort to increase CTR.
Column C – % of country The % of impressions allocated within the counrty
Column D – % of total The % of impressions allocated compared to the total amount of impressions across all geographies.
Column E – Impressions This column is pre-conditioned by dividing by 1,000 (CPM formulas use Impressions / 1000 frequently).
Column F – Clicks The number of clicks generated on the individual campaign. This figure can be taken from either your ad server or an ad network / advertiser dashboard.
Column G – CTR Click through rate. Determined by the number of clicks divided by impressions
Column H – Conversions Data taken from an ad network or advertiser.
Column I – Conversion % number of conversions divided by the number of clicks.
Column J – Revenue data taken from an ad network of advertiser.
Column K – % of country revenue revenue from the campaign divided by total revenue of the country
Column L – % of total revenue Campaign revenue compared to the total revenue from all campaigns
CPA Analysis Metrics & Overview
Column M – EPC. Earnings per click. Earnings divided by the number of clicks.
Column N – ECPM. Earnings per 1,000 impressions.
Column O – (+/-) Eff. Efficiency. Comparing the % of impressions allocated total, to the % of revenue total.
Why are these analysis metrics important?
Earnings per click. Understanding earnings per click helps us identify poor converting campaigns. If EPC is low, it means that you’re sending too many clicks without converting users.
Action point: look into the creative, text, description and make sure the landing page matches the ad perfectly.
Also, some campaigns run on a CPC basis, where optimization is focused on improving the CTR of the creative. It does not require any conversions, and with this type of campaign you can easily price out direct CPC ad sales by looking up your current campaigns EPC.
ECPM is the biggest indicator of a high performance CPA or CPC campaign. It’s also the metric monetization pros understand as being the holy grail. Since we directly influence the amount of impressions allocated through our ad server, optimizing on a CPM basis is easy.
For example, if we know campaign #1 has an earnings per thousand impression (eCPM) of $3 and it’s filling 50k impressions per day, while campaign #2 has an eCPM of $6 filling 10k impressions per day… we would prioritize campaign #2 on our ad server, maximizing the higher earning potential as identified by eCPM metric.
eCPM is a universal metric, but requires you to make sure you’re comparing apples to apples on a campaign by campaign basis.
All analysis metrics require controlled tests.
For example, comparing US Campaign 1 vs US Campaign 2.
(+/-) Efficiency compares two percentages to understand individual campaign performance relative to one another. More specifically, this column analyzes the % of impressions of the campaign in relation to the total number of impressions…compared to the % of revenue of the campaign, compared to the total earnings. The result is a + / – from 0, which identifies successful campaigns as receiving low volume but yields high revenue.
If that doesn’t quite make sense, take a look at the formula’s on the spreadsheet and conceptualize the initial percentages.
At the bottom of the spreadsheet, there’s a macro view which shows the ad unit we’re optimizing on a macro geotargeted level. By organizing the spreadsheet into both micro and macro views we can understand the details and how they impact the ad unit earnings as a whole.
By reviewing the macro view, monetizing experts can quickly identify undervalued aspects of the inventory, prompting a new test.
Setting up your CPA ad inventory
Read our previous blog to learn how to pick a location for your CPA ad unit, and setting up your inventory.
Customize your ad with HTML / CSS
Configure your campaigns with an ad server
Get live campaigns from recommended Ad Networks
Go live, and test!
Assuming that you have already setup your first round of campaigns, and have 3 geotargeted campaigns setup already for US or your largest traffic source… then we’re ready to move forward designing our first A / B test.
Let your current configuration run for at least 3 days initially to get data. Update your spreadsheet after 3 days with the number of impressions, clicks, conversions and revenue (Column E – Column J).
Having this initial data set will update the analysis figures and provide deeper insights into the performance of each campaign.
Update all geotargeted campaigns, check formulas and review the macro performance
Use the following questions to stimulate your thinking:
What countries campaigns are performing best on an eCPM basis?
What’s the most efficiency country?
What’s the lowest Earnings Per Click (EPC)?
When you’re finished with this, look at your micro geotargeted view above.
Of the country with the highest earnings:
What’s your top earning campaign?
What’s the highest eCPM campaign?
What’s the lowest EPC?
What’s the most efficient campaign?
Action point, choose your champ. What’s the absolute best campaign? Move that to the top of the country grouping.
Find the 2nd champ, or the lowest performing campaign.
Note: In general, I’m a firm believer in having at least 2 campaigns per geo within your top 5 geo sources. This way you can see which campaign is converting, deactivate the lowest performer, and then run another campaign against the champ.
Another train of thought. If you’re reviewing the campaigns, to further optimize a campaign that’s working — change up the thumbnail, title, description and further test CTR & Conversion rate.
The objective is to relentlessly test every option, every angle, to drive up yields.
Running your first A/B test
“In marketing and business intelligence, A/B testing is jargon for a randomized experiment with two variants, A and B, which are the control and treatment in the controlled experiment” Wikipedia
In practical terminology, we’ll structure our first A/B test to challenge one of our champion campaigns that we’ve identified above.
Practical example, make sure your spreadsheet is opened.
Let’s look at the micro view of the US allocation.
In this scenario, I have 2 different campaigns running but Campaign 1 has 4 different creativies: Character, battle, walking and fire.
I’ve marked the spreadsheet to show the action points and test.
Row 10, the “Lose” Campaign 1 – Fire campaign was identified as the lowest earnings per click, eCPM, and efficiency. The action taken is to deactivate the campaign. In it’s place, I’ll be testing a new Campaign to fill the 11% total impressions that were being delivered to [Fire].
I’ve marked a row in this US grouping as “Test”.
Ad server action needed. In order to run a new test you’ll need to get dirty in the ad server, and setup a new campaign based on the initial HTML / CSS that was deployed.
By setting up a new campaign you’ll need to take a new thumbnail, add a description & title, get new links and setup subids to track the details of the campaign.
Once the banners has been setup, then you can link the live ad zone or ad unit and test to make sure it’s working.
The above description really ran through the process quickly, if you’re confused at all reference the CPA setup document, located here:
The test campaign should run for a minimum of 48hrs before running the CPA optimization document. This allows for more data to accumulate, which you can safely make assumptions from.
The US example above is not exactly an A / B test where 50% inventory is allocated to 1 campaign and the remaining 50% is dedicated to another campaign, which is yet another reason why the analysis metrics are important. As mentioned before, these metrics provide a way to compare campaigns on a relative basis.
So you’ve added your first test, updated the data for the first 48hrs, what’s next?
Back to asking your spreadsheet questions.
How did the new campaign perform on a EPC, eCPM and efficiency basis?
Is it performing better than the champion campaigns?
If you’re test campaign is performing better, bump up it’s priority in your allocation and run the test again for 48hrs. Try to bring it to an even 50/50 distribution. Then you can further determine if you want to prioritize the campaign over the champion, and downgrade your original allocation.
The moral of the story is, testing and keep testing. Optimizing ad units is a bit tricky to get the hang of, but once you understand the fundamentals and testing methods, it’ll become easy to zero in on various components of your inventory and explore different ways to challenge and take advantage of campaign performance.
Next week I’ll be updating my document to analyze week to week performance and drilling into just how well you can optimize your inventory over a long term basis.
Through this same process, I was able to generate over 800% growth on my initial week setup. Get after it you data science guru!
If you found value in this post, please share it!
Disclosure: I have yet to edit this post and will be refining the strategies and organizing the article better. – Kyle 8/27/14