Ad network comparisons – it shouldn’t take a quant
As the first topic on the official AppLovin blog, it would seem odd to focus on ad network comparisons. However, with GDC in full force this week and developers being top of my mind, I thought it’d be a good way to start out.
To start let’s point out the several layers to look at when assessing an ad network. These include:
- Global Fill Rate: At what level does the ad network fill and how does it fill by country? That number may range anywhere from 99% to 0% and may certainly vary by country.
- Ad Units: Interstitials & Banners being the most prevalent. Fill rate may vary based on ad unit size.
- eCPM: What kind of eCPM can the ad network provide? This should be assessed on a country-by-country basis, although looking at it globally certainly simplifies things.
When comparing ad networks, it’s truly a combination of the above and there are often significant mistakes made when comparing. AppLovin is particularly sensitive to this and I’ll explain why in a moment.
- Split testing ad networks must be done randomly and not in successive order (i.e. the same placement in both location in the app as well as time wise with the user). Meaning if AppLovin were to see the second ad refresh for each user, performance would be reduced. Rough statistics show that the value of each refresh is as follows. Slot 1: 100%, Slot 2: 80%, Slot 3: 50%. To test this properly, split test either randomly with a percentage to each or assign one network to odd device IDs of users that see ads and the other to even IDs.
- Split testing must be done at the same time and not during different time periods. That’s due to the new user vs. returning user matrix for each individual app differing over time. Performance on the advertising side will often move in tandem with the new user curve. This is not well known, but new users will often vastly outperform returning users; the longer users are active within your app, the more committed they are to your app and that makes them much less likely to click on an ad and convert on someone else’s offering.
- Split testing by looking at eCPM comparisons should take into account fill on a global basis as well as eCPM comparisons on a country-by-country basis. Here are some ways that a network can inflate eCPMs at the cost of maximizing revenues: 1) They won’t fill in lower value countries. Automatically, eCPM increases. 2) They won’t fill the 2nd, 3rd, or 4th impressions for each unique user. Again, given my point above that means they’re maximizing the value of each impression @ 100%. Some networks will combine 1 and 2 and really show “great” performance. Assuming you have a good test environment, always keep an eye on revenues that networks generate for you as the key metric. End of the day, we’re all here to make money!
At AppLovin, we strive to generate industry-leading CPMs, but we care more about generating you the highest earnings you can make from having ads in your app. To keep with the motto of not wasting potential ad impressions we fill 99%+ across our network. eCPMs should be compared via the accounting of fill rate, country data, ad unit size, and placement. I’m always happy to discuss this type of stuff in more detail, so shoot us a note or comment here.