This week we’re featuring everything you need to know about using CPI (Conversion per Impression) as your testing metric. Please, consider reading our article on scientific Google AdWords split-testing if you would like to know more about how to set it up and evaluate results.
This is a metric that shows the ratio between impressions and conversions.
When you consider ad testing, which combination is better?
It’s impossible to say which is better since that information relies on two different metrics: CTR and Conversion rate.
What CPI does is combine these two different metrics to form one single metric that will show you which ad will receive the most conversions from the impression.
Every time your ad is displayed; you have a chance of a conversion. You picked a keyword. Someone searched for your keyword. At this point in time there’s a chance of a conversion. The user must both click on your ad and then convert to receive the actual conversion; but measuring from the impression shows you the total conversions possible.
CPI is calculated by dividing total impressions by the total conversions.
CPI = conversions / impressions
It is generally displayed as a percentage. Here are some examples:
| Ad | Impressions | Conversions | CPI |
| 1 | 10,000 | 12 | 0.12% |
| 2 | 10,000 | 5 | 0.05% |
| 3 | 10,000 | 15 | 0.15% |
| 4 | 10,000 | 14 | 0.14% |
| 5 | 10,000 | 13 | 0.13% |
| 6 | 10,000 | 10 | 0.10% |
In this example, ad 3 has the highest CPI and ad 2 has the lowest Conversion per Impression.
The main reason to use CPI is when you want the most conversions possible. As this metric takes into account both CTR and conversion rate, it’s a simple metric that will show you which ad will lead to the most total conversions possible.
There are times that when you examine the full metrics behind various ads, you might not always pick the highest CPI winners. This usually happens for a few reasons.
Let’s take a look at a full chart of data and then examine how we’d pick the various winners (click the chart to see a larger version).
If we just want the most conversions possible; then ad 3 is our clear winner. It receives 15 conversions for every 10,000 impressions. This is why CPI is such as great metric. Ad 3 is not the winner in CTR (that’s ad 4) or the winner in conversion rate (that’s ad 6). In comparison to our other ads; ad 3 has both a good CTR and a good conversion rate, but it is not a winner or loser in either metric. However, when you use both CTR and Conversion rate to calculate the most conversions possible for and ad; then ad 3 is our clear winner.
If we are struggling with Quality Score; then we might pick ad 4 as it has a much higher CTR than the other ads, which is also why it has a lower CPC than the other ads, and its CPI is not too far behind our winner.
If our goal is a $30 CPA and we want the most conversions under $30; then we’d eliminate all the ads with a CPA higher than $30 and pick the highest CPI ad from the ones that are left as our winner; which is ad 5.
If we only wanted 10 leads a month and therefore, pay as little as possible for those 10 leads; then ad 6 would be our winner as it has the lowest CPA of all the ads and it can reach 10 conversions per month (assuming this was monthly ad data).
Assuming we want the most conversions possible, or even the most possible under a $32 CPA target, then ad 3 is our clear winner. However, we should learn from the other ads in our next set of ad tests. This is where examining ads that lost in your overall metric but won in a single metric is useful.
For example, ad 4 is a clear CTR winner. Why? Was there a line in the ad that was very attractive to users? We might want to use that line for our next ad test. We could duplicate ad 3, add that line, and then run another test.
Why did ad 6 have such a great conversion rate? It did have a very low CTR. Odds are, this ad was ‘pre-qualifying’ users with ad text that was meant to filter out users. We could look at what that qualification is, use it in a new ad 3 duplicate, and test that combination.
Finally, ad 2 was a conversion rate failure. Why? We might want to add a note about a line to avoid as this ad clearly attracted the wrong types of clicks.
The one downfall of CPI is that it doesn’t care about revenue or total order amounts. If you have an ecommerce site and want to base your conversions off of ROAS or revenue targets, CPI is not a great number to use.
CPI is great for lead generation, or the most conversions possible. However, since revenue and ROAS are not numbers used in its calculation, there are other metrics, such as ROAS or RPI/PPI (revenue/profit per impression) metrics that are better to use for ecommerce companies. We’ll cover these additional metrics in the coming weeks.
When testing ads, you often want to create minimum data requirements for some of your metrics before you even examine if a test has achieved statistical significance.
These metrics will vary by your testing metric.
When considering CPI as your metric, there are three considerations to use:
With minimum data requirements; usually you want every ad in the test to reach the minimum data numbers before you examine the tests.
When you consider CPI as a testing metric, you usually care about two metrics: Impressions & conversions from a minimum data standpoint.
Some people are tempted to set a high number of conversions, and only use conversions, as their minimum data when testing by CPI. However, the issue is that if you have an ad with a very low conversion rate, then it might take a long time for that test to hit minimum data since the ad rarely converts. Thus you usually don’t want to set a very high minimum data number on the number of conversions every ad must have before a test is valid.
Now, you always have to temper this by how your account acts. For instance, if you are doing 500 conversions a day in your ad group and you are testing 3-5 ads, then you might set a threshold of 100 conversions per ad. Just remember, if one ad doesn’t achieve that number, then the test will never be complete.
As for minimum clicks, this is usually a number you can ignore. You don’t want very low CTR ads to stop your test. As CPI is based upon only impressions and conversions; usually CPI testing completely ignores minimum clicks since clicks aren’t even used in calculating that metric.
The primary number you want to set as your minimum is impressions. When an ad is displayed, then you have a chance at a click, cost, conversion, and a way of determining your CPI for each ad. If you have a very high volume ad group, you might set a very high minimum impressions (such as 1000 or even 10,000). If you have medium volume ad groups; then you might only use 500-1000 impressions. If you have low volume ad groups, then it is recommended that you use multi-ad group testing instead of A/B testing and can once again set higher thresholds.
Just remember, your ad rotation settings will also determine how often an ad is shown. So if you are using optimize for conversions or clicks, it will take longer for each ad to achieve minimum data.
Conversion per impression is one of the best ad testing metrics you can use since it is a simple number that lets you see which ad will lead to the absolute most conversions.
The largest downside of CPI testing is that it does not take into account revenue per conversion. So while it’s a great metric to use for lead generation companies, it might not be suitable to ecommerce companies.
If you have hard CPA targets, then you can use CPA as a filter to remove ads that are above your target CPAs and then pick your highest CPI ad as your winner; so many companies should use both CPA and CPI testing metrics to choose their winners.
If you are not taking into account revenue for your testing metrics, then you should always evaluate CPI when determining your winners – it’s that good of a metric to use for your testing needs.
If you would like to easily see your Conversion per Impression metrics along with other testing metrics (such as CPA), take a look at what Adalysis has to offer. You can easily see your CPI along with many other metrics to make quick determinations as to what ad really is the best one to use for your account to ensure that you are hitting your advertising goals.
Pingback:How to Test Landing Pages & Ads Together - Adalysis
How do you determine statistical significance when testing by profit per impressions?
You will still use the same SS formulas that you would for testing by clicks, conversions, etc. It’s just that the data will use impressions as opportunity and conversions as the action.
Can you use CPI for more than ad testing? For example, determining winning Ad Groups, etc.
Hi Nick,
Of course, we use it for a lot of things; however, its primary usage is probably in ad testing.
Are the statistics in this article realistic, or just examples?
These are all examples from actual accounts, so they are realistic examples.
Check Your Account For Free!
