- Ad mediationUnity LevelPlay
- In-app ad networkironSource Ads
- On-device solutionsAura
- ironSource ExchangeironSource Exchange
- App marketingLuna
- Ad mediation
- Log in
- Sign up
Building a predictive model
Let’s put everything you learned into practice. Learn from Michel Hayat, Senior Product Marketing Manager at AppsFlyer, how to actually build a predictive model for your app.
Hi there, you’ve made it to our third and final course on predictive analytics. I’m Michel Hayet, a senior product marketing manager at AppsFlyer, and in the past two episodes, we’ve covered what predictive analytics is and the challenges that SKAN presents for app marketers. Here, we’re completing the loop and talking specifically about how building a predictive model can solve the concerns brought on by SKAN.
The key challenge here is the measurement issue, specifically long-term measurements.
The question to answer is:
What value will a specific user have after using the app for thirty days?
To answer this question, break it down into three sub-questions that address the pillars of LTV: monetization, retention, and behavior.
The first question is how much monetary value will the user create at this stage (that’s the monetization pillar)?
The second, how long do you think the user will remain active in the app (here’s retention)?
And third, do you believe this user will convert into a paying user (the behavior question)?
Before attempting to answer these questions, you first need to construct the predictive model by analyzing and mapping as much of the app’s first party historical data as possible. A few rules to keep in mind here: The more data you have available, the better. The more this app relies on post install event measurement, the better. And finally, the more events you can measure over the first twenty-four hours, the better.
This model can take a few weeks to construct, test, modify, and perfect.
You then need to validate its results to make sure the predictive model goes hand in hand with the app business’ core logic. What we mean here is that different apps have different ways of calculating LTV and defining which events are most important. One game app may decide that their key event is users finishing the tutorial within four hours of opening the app. Another game business could say it’s crucial that their users reach level five within three days of installing the app. It’s important to define what your app business’ unique logic is so your predictive model complements it.
Measuring campaign performance
Let’s put this to an example using the AppsFlyer predictive analytics product logic.
The AppsFlyer SDK lets you measure all of the user’s measurable events in the twenty-four hour window. Ad views, tutorial completion time, in-app store visits – pretty much everything you need.
These events, along with other elements like timing and interaction type get analyzed by the predictive analytics AI engine. The result? Answers to the three questions I mentioned earlier about revenue, retention, and behavior.
Keeping all that in mind, let’s talk about the indications we need to keep our eyes open for.
The number of ads we were able to present to the user. Whether they interacted with these ads and how. Did they visit the in-app store? Did they add items to their cart?
This means that we are actually using the science of predictive analytics as a measurement solution, one that can provide indication into likely future performance.
Interested in learning more about this solution? Head over to AppsFlyer.com to learn more.
Thanks for watching another episode of the LevelUp Academy. We hope you now have a good grasp of predictive analytics and how to use it for your business. Tune in for more content soon!