Announcing the official beta launch of PredictSK
After months in the development lab, we’re excited to announce that PredictSK is officially going into beta this week.
AppsFlyer’s original predictive analytics solution was ready to go into beta about a year ago. After months of development, the original solution was meant to provide AppsFlyer’s customers with the means to better manage their user acquisition campaigns, by introducing predictive insights early into their campaign’s launch.
This solution was meant to solve the uncertainty caused by the period of time UA managers had to wait between the campaign’s launch time and the point where sufficient LTV insights were gathered. All in an effort to make educated and data-driven optimization decisions.
Our market research showed that the average advertiser waited around 30 days or more (from the moment a campaign launched, prior to any optimization decisions). This time period was considered necessary for gathering LTV data. As well, UA managers did not optimize the campaigns during this period, for the sake of gathering “clean” data, despite that the campaign continued to incur.
Some advertisers who applied advanced BI and data science tools (whether in-house or third party) could potentially shorten this waiting period to as little as 10 days using insights delivered by these tools. Some of the leading enterprise-level companies are even able to cut this period to mere days through the science of predictive analytics, but usually at a significant cost.
All of the above would involve a form of cost-in-return for time-to-decision. Waiting of course requires no direct cost, but it results in campaign budgets bleeding until clear insights are available. It was left for advertisers to decide what “price” they’re willing to pay and a balancing act of cost vs. time-to-insights.
Predict’s initial mission was to lower the cost factor for the predictive analytics portion of the equation, by making this advanced solution affordable for mobile marketers of all sizes.
And then came SKAdNetwork
Apple’s announcement of the (then) upcoming changes to its user-privacy regulations also introduced a new attribution mechanism that would quickly become synonymous with obfuscation. IDFAs will now become obsolete, but SKAdNetwork would become the main discussion point. The use of one conversion value measured across 24 hours to measure and determine user value meant measurement data is now almost unavailable. The use of a six-bit mechanism to communicate with SKAdNetwork had marketers running calculations of what value could or should be used, and which bit tactic they adopt.
However, AppsFlyer has always been set on privacy-centric advertising. The elimination of user identifiers simply encouraged us to change the question from: who the user is? to: what can their behavior tell us?
A quick review of our predictive analytics MVP made it clear what had to be done. A predictive analytics solution that can use LTV measurements across a limited time frame to produce fast actionable insights was now not a nice solution to have, but a must have for anyone looking to operate in this new reality.
Several modifications and improvements had to be made for accommodating the new SKAdNetwork technology. Our predictive benefit score associated with each user would now become the conversion value; but unlike regular conversion values, it would encapsulate the entirety of measurable events across 24 hours. Our measurement time frame would have to be modified from 72 hours to 24 hours, while maintaining its accuracy level. And additional machine learning algorithms would be added to translate hundreds of possible scoring combinations into 64 possible values.
These were just a few of the challenges we faced since embarking on this journey, but the result is a better solution for the entire industry.
Boots on the ground
Today officially marks the beginning of PredictSK’s beta, with the first AppsFlyer customers onboarding the solution, initiating the solution’s optimization and perfection phase.
Generating a unique predictive model per app requires a necessary onboarding period where PredictSK’s AI engine is trained on each developer’s unique LTV logic and maps correlations between early signals and end results (more on this in future posts).
In a recent survey conducted across AppsFlyer’s customers, nearly 50% of respondents rated predictive analytics as one of the two most anticipated releases for iOS measurements.
As more AppsFlyer customers are set to gradually onboard the solution in the coming months we encourage you to study and understand the benefits of predictive analytics in mobile marketing to make the most of your campaigns.