For the first time, Google generated the highest number of installs in the app marketing space, globally. Facebook, however, retained the #1 power ranking position in the Index thanks to its high quality and greater client adoption.
AppLovin has established itself as the #3 player in Gaming, but ironSource is growing faster, increasing its share in the Gaming app install pie by 32%, compared to AppLovin’s 10%.
TikTok Ads is experiencing massive growth worldwide and takes the #1 spot in the Growth Index. The Chinese giant is poised to become a significant global player.
Fraud continues to impact the rankings, but the good news is that the install share of fraudulent networks dropped by 60%, while clean sources grew by 25%.
Apple Search Ads is rapidly growing, increasing its share in the global app install pie by 82%. It has plenty of untapped growth potential, as fewer than 20% of iOS apps run ads in the store.
We only included media sources that met our strict conditions on two fronts:
Volume Ranking: A ranking of media sources based on the total number of non-fraudulent installs each was attributed for.
Power Ranking: We normalized and combined the number of non-fraudulent installs, the number of apps running with each media source, and the weighted retention score (see detailed explanation below). We then factored an additional fraud penalty based on the network’s overall fraud rate for the region in question.
We compared the performance of the top 350 media sources in H2 2018 vs. H1 2019. The comparison was calculated by combining a number of factors: install growth, number of apps growth, average installs per app growth, and growth in share of the app install pie.
We factored and normalized the number of attributed retargeting conversions (a conversion occurs when an existing user that has the app installed engages with the retargeting campaign, excluding re-attributions), and the revenue generated from these conversions (based on all events reported after the retargeting attribution occurs and within its attribution window).
STEP 1: We calculated the non-organic retention rate of each app per media source and per region. We did this separately for each day of a 30-day period, dividing the number of users who were active on the day in question by the total number of users who first launched the app in the selected timeframe. We added two longer term signals — week 8 and week 12 post install — dividing the number of users who were active on the week in question by the total number of users who first launched the app in the selected weekly timeframe.
STEP 2: We calculated the organic retention rate of each app on a regional level, separately for each day over 30 days, and for week 8 and week 12.
STEP 3: We then compared the non-organic and organic retention rates for each timeframe. Using organic retention as a benchmark significantly reduces the impact of a given app’s quality, and therefore offers a far stronger indication of a media source’s performance.
STEP 4: We calculated a weighted average using a retention-based logic; the longer a user is retained, the higher the assigned weight. As such, the day 1 non-organic to organic ratio had the least weight, and week 12 the most weight. This weighted average serves as our retention score.
STEP 5: We calculated a network’s overall weighted retention score per region and category group in question by taking the retention score of each app separately and factoring the number of installs it delivered.
Install fraud rate: We divided the number of a network’s fraudulent installs coming from Device Farms and Bots by its total number of attributed installs.
Poaching fraud rate: We divided the number of a network’s fraudulent installs coming from click flooding and install hijacking by its total number of attributed installs.
Overall fraud rate: We divided a network’s poaching and install fraud by its total number of attributed installs.
Clean installs calculation: We reduced the number of fraudulent installs from each network’s overall install count according to its install and poaching fraud rates (the latter is based on stealing organic or non-organic users of other networks and therefore impacts the install count).
Clean retention score calculation: We reduced a network’s retention score according to its poaching fraud rate (most of this fraud is based on stealing organic users, thereby elevating a network’s retention and engagement levels).
Fraud per region and category: Because the level of fraud differs by region and category for different media sources, we used the specific fraud rate for each in the region and category in question.
Exclusion: Networks that did not meet our overall fraud rate threshold by region were excluded from the Index in question.