# eCPM Variance in Mobile Ad Networks

If you have been placing ads in your apps, you are not alone. More and more apps are doing so and even the top grossing games have reached 50% ad penetration. Figuring out how your payout is calculated, however, proves to be a very difficult task. The ad-networks report an eCPM level that along side the impression volume makes your payout as shown below:

Basic formula (that is actually wrong):

Payout = Impression volume x eCPM

In reality, the situation is a bit different. eCPM is only calculated in retrospect as the average CPM across all impressions. So the basic formula is actually

The real formula:

Payout = sum(Price for impression1, Price for impression2, ….)

eCPM = (Payout / Impression volume)x1,000

### Impressions can have drastically different prices

So what the formula above means is that eCPM is an averaged number. Now, you might remember from your statistics class that average can be misleading when variance is high. Don’t remember? Here is an article to brush up on this topic. Impression payout certainly falls into the high variance category. Let’s see how high the variance is.

### eCPM Variance in CPI Networks

What is a CPI network? The following networks pay mostly based on rev-share of their CPI campaigns: Unity Ads, Vungle, Chartboost, Supersonic/Ironsource, Applovin, Adcolony, Tapjoy. In these networks, the CPI variance is especially high. This is a typical distribution of the actual prices:

Price Proportion of impressions that pay this price
\$0 99%
\$500-\$1,000 0.2%
\$1,000-\$2,000 0.4%
\$2,000-\$3,000 0.2%
\$3,000-\$4,000 0.2%

If this seems a bit high it’s ok. We were also surprised but the simple reason is that eCPM is 1,000x than the price of the impression. \$4,000 eCPM really means that there is an impresison that paid \$4. This is simply the impression that led to the install. \$4 is not a very high price for an install after all, there are much higher prices.

### Variance in bid levels in RTB

If you are using an SDK from Mopub, Inneractive or Smaato, it means that each impression is getting a differnet price based on an auction. The bid levels here also vary drastically. In some cases you can see bids of over \$100 when a user is being retargeted and in some cases the bid levels are at \$0.05 since no one is bidding on that user except for a scavanger that bids the minimum on everything and catches the leftovers.

Facebook and Admob both operate an entire eco-system. Each of them have both CPI campaigns that are paying per installs or at least optimizing per installs along side retargeting campaigns and other forms of demand. The result here is also the same – big variance in eCPMs when you look at specific impressions and one single avarage eCPM reported for all the impressions.

### Why should you care about variance in eCPM

If you are basing any business decision on ROI calculation that uses the avarage eCPM, you should care about eCPM variance. Trying different App Icons and App names, new features, traffic sources, monetization setups. These are all decisions where you try to measure the ROI. If your ROI calculations uses the average eCPM to estimate ad revenue as part of that calculaitons – it’s likely that you are making the wrong decision.

If you want to know the eCPM of every single impression and make better ROI calculations you should check out SOOMLA TRACEBACK – Ad LTV as a Service