AI Model Generation
Monetai's AI automatically builds and trains a custom purchase prediction model based on the user event data collected from your app's SDK.
Model Generation Process​
1. Data Collection
Once the SDK is integrated, it begins sending user event data to our system. This initial phase focuses on accumulating enough data to build a reliable model. We specifically track:
- General user actions (e.g., viewing a product page)
- Key purchase events (e.g., trial starts, paid conversions, renewals)
2. Automated Model Training
Model training starts automatically once a specific data threshold is met. No manual action is required.
- ✅ Training Condition: Training begins when at least 100 paying users and 100 non-paying users have been collected within a recent 3-month data window.
The entire model training process is fully automated. Our system checks your data daily and automatically begins training once the conditions are met, requiring no configuration or intervention from you.
Model Accuracy and Status​
How We Define Accuracy
We measure model accuracy using Recall, a standard machine learning metric. In Monetai, Recall means the percentage of actual paying users that the model correctly identifies. In other words, it's the key metric that shows how well the model find the users who were going to purchase, without missing any.
- Why Recall Matters: A high recall means that the model accurately identifies users who will buy anyway, which prevents from offering unnecessary discounts and protects your baseline revenue.
Accuracy Levels​
Your model's status and accuracy are categorized into the following levels:
| Accuracy Level | Recall Value | Description |
|---|---|---|
| High | 0.8 or higher | Highly accurate predictions |
| Medium | 0.6 ~ 0.8 | Solid, reliable prediction accuracy |
| Low | Below 0.6 | Accuracy can be improved |
| Optimizing | - | Collecting data, or initial model training is in progress |
Checking Your Model's Status
You can check your model's current training status and real-time accuracy in the right sidebar of the Monetai dashboard.
We recommend starting your campaigns when your model's accuracy reaches Medium (0.6) or higher. At this point, the model is reliable enough to run effective and profitable promotion campaigns.
Model Retraining​
Model retraining is currently not supported. If you believe your model needs to be retrained due to significant changes in your app or user base, please contact support@monetai.io for assistance.
Frequently Asked Questions​
- Q. What are the criteria for the AI model to classify users as 'purchasers' and 'non-purchasers'?
- A. Monetai's AI model doesn't judge users simply based on their purchase history. For more accurate predictions, a user is only included in the analysis if they meet the following activity criteria:
- Session Length: 12 minutes or more.
- First Event: The first logged event must be a custom event, not an automatic payment event from the SDK.
- Time Period: The data must be from within the last 90 days.
- Q. A purchase happened in my app. Why isn't it immediately reflected in the 'purchaser' count on the Monetai dashboard?
- A. Even if a purchase occurs, the user must meet all three criteria (session length, first event, and time period) before the AI model recognizes them as a valid 'purchaser' or 'non-purchaser' for analysis.
This process improves the prediction model's accuracy by excluding inactive users (e.g., those who log in only once to make a purchase)from the analysis.
Support
If you have any questions about the model generation process, don't hesitate to contact us at support@monetai.io.
Next Steps​
Once your model's accuracy is at a sufficient level, you're ready for the final step: