Skip to content

Announcing AdLibertas Machine Learning Prediction: fast, accurate, user pLTV and pROAS

Historical Recursive Regression Modeling (HRRM) – Our proprietary Machine Learning Prediction (MLP) models learn from historical performance to provide accessible, fast, accurate predictions on your acquisition campaign performance.

The most common question we get from app marketers is “can you help me predict campaign performance sooner?” Over the years we’ve built predictive tools for app marketers, but one shared difficulty was the ability to make the data actionable. Boiled down, the problems with predictions are: which prediction model should I use, and how confident should I be with the prediction?

In aiming to solve these challenges we’ve developed a machine learning model that will automatically build and deploy the predictive model on your campaign while providing historical measures of accuracy for reference.

Now app marketers can easily see modeled performance and make campaign adjustments with confidence.

How it works

Gathering data:  For those unfamiliar with the AdLibertas platform, we first will need to collect all app data: mmp, in-app analytics, in-app ad impressions, purchases, and subscriptions where applicable. Since we rely on your existing technology stack, there’s no technical integration, we simply connect using APIs.

Historical training: Our model intelligently samples historical campaign performance and applies supervised learning models to build the most applicable projection model for your campaign. In general, we can get started with only 30 days of historical performance.

Future projections: We then use learnings and historical samples to build projection models for future performance on your in-progress campaigns.

Your projections can be applied as predictive LTVs (pLTVs) on campaign status or predicted return on ad spend (pROAS) at the target dates of your choice.


Accuracy & Weights

Prediction accuracy will depend on each app, demand source, and even campaign. We’ve found the accuracy of our ml-models will largely depend on three factors:

Performance of your users – Two factors of user performance: when your users monetize and the density of monetization events. Generally, accuracy will be increased the sooner and the more uniformly your users monetize. As you can imagine 1K users seeing an ad impression on day 1 will be easier to predict than 1 user who subscribes on day 30. We give you the ability to customize the projection timeline to get the accuracy you desire.

Size of your audience – Larger groups of users generally create more predictable behavior, for new campaigns with limited datasets our supervised learnings increase the sample size to increase accuracy.

Maturity of your cohort – By and large, the more mature your cohort, the more accurate your end measurement. Since getting information sooner is always the goal, we offer the ability to get multiple maturity outcomes for a campaign to ensure you can balance fast and accurate results.

If you’d like to see if the AdLibertas MLP is right for your app, please get in touch with us!