【Prediction】Mastering HUAWEI Prediction – Serve Users the Smart Way - Huawei Developers

Low retention and difficult conversion are huge headaches for operations personnel, who often find it difficult to pay for what they have lost. For example, following user churn, the recall rate may not be satisfactory, and costs are high.
Regardless of whether the users can be converted into paying users or eventually be churned, their choice is directly related to their attributes and behaviors in the early stage. Based on such data, HUAWEI Prediction uses AI algorithms to lock users who are likely to churn or pay fees in advance, and then implements targeted measures to extend the user lifecycle and boost the payment conversion rate.
HUAWEI Prediction can help you by: focusing on churn and payment conversion, offering two core operation scenarios, and performing multi-dimensional target audience predictions. It supports multi-touch operations for predicted audience through Push Kit and App Messaging.
This powerful service has drawn acclaim from developers, ever since it was launched. This article shares some key tips that can help you make the most out of HUAWEI Prediction during daily operations.
{
"lightbox_close": "Close",
"lightbox_next": "Next",
"lightbox_previous": "Previous",
"lightbox_error": "The requested content cannot be loaded. Please try again later.",
"lightbox_start_slideshow": "Start slideshow",
"lightbox_stop_slideshow": "Stop slideshow",
"lightbox_full_screen": "Full screen",
"lightbox_thumbnails": "Thumbnails",
"lightbox_download": "Download",
"lightbox_share": "Share",
"lightbox_zoom": "Zoom",
"lightbox_new_window": "New window",
"lightbox_toggle_sidebar": "Toggle sidebar"
}
Question 1: What can I do with predicted audience results?
Audience predictions work seamlessly with other growth services provided by AppGallery Connect, such as Push Kit, App Messaging, and Remote Configuration.
For example, you can use Push Kit to push promotional messages, such as new game versions or online gifts to users with a high churn probability, to keep users active and prevent churn. To implement this function, just select your preferred option under Prediction users on the Push Kit page of AppGallery Connect. Likewise, other services, such as Remote Configuration and App Messaging, can also use this method to filter audiences and reach target users.
Question 2: How can I assess the accuracy of prediction results?
The true positive rate and false positive rate displayed on the prediction details page show you an overall prediction result. The true positive rate is the ratio of the number of positive samples correctly predicted by the model to the actual number of positive samples, while the false positive rate is the ratio of the number of negative samples incorrectly predicted by the model to the actual number of negative samples.
For example, in payment prediction, the true positive rate indicates the ratio of paying users correctly predicted by the model to the total paying users. The false positive rate indicates the ratio of the non-paying users incorrectly predicted to be paying users by the model to the total number of non-paying users. The higher the true positive rate, the lower the false positive rate, and the more accurate a prediction result is.
Question 3: Why are prediction tasks unsuccessful after the prediction service is enabled?
This issue relates to the principle behind Prediction. The prerequisite for running a prediction task is that your app reports user attributes and behavior data through HUAWEI Analytics. Therefore, before using Prediction, you'll need to enable HUAWEI Analytics and integrate the HMS Core SDK to ensure that the corresponding user behavior data is reported.
For example, many developers report that payment and return predictions garner no results. This is because results are dependent on whether your app reports payment events. Prediction results can be generated only when sufficient payment events are reported to support prediction model training, for example, the INAPPPURCHASE event.
Question 4: How can I use the custom prediction task?
In addition to the preset churn, payment, and return scenarios, there's a custom prediction task that can help you out in other prediction scenarios. You can specify your prediction task for user behavior based on your specific product operations requirements.
For example, for a game app, operations personnel may focus on the probability that players complete a level. In this case, you can create a custom prediction task by using the game level completion as a target prediction event. For more details, please refer to Customizing a Prediction Task.
This article has provided some answers to frequently asked questions about Prediction. We'll continue to release more articles that delve into the Prediction service. In the meantime, you can also click here to learn more about the service.
To learn more about DTM, click here.
For more details, you can go to:
Reddit to join our developer discussion
GitHub to download demos and sample codes
Stack Overflow to solve any integration problems
Original Source