Cloud Music User Push Notification Optimization: Practices and Insights

This article introduces Cloud Music's practices and insights in optimizing user push notification reach and engagement.

Background

Current app external user reach methods include ads, SMS (phone calls), and notification bar pushes. The first two methods cost money, while notification push basic capabilities are mostly free. Previously, Cloud Music's push notifications were chaotic with multiple business entrances and access platforms, making maintenance cumbersome. There was poor attribution of push impact on user recall and frequency increase, and Android manufacturers are increasingly emphasizing notification management with stricter content and quota controls.

Entry Integration

The key optimization was channel splitting - separating business layer from sending channel layer. The business layer focuses on Push, SMS, and private message platform capabilities such as creative management, push plan management, position management, frequency control, version control, user segmentation, content classification, and risk control. The channel layer maintains unified manufacturer protocols and user device information. They used the Link product (Cloud Music's delivery reach platform) for online-offline integration, abstracting Push, SMS, and private messages as position concepts, enabling one delivery product to solve both online and offline delivery needs.

System Optimization

Platform Infrastructure: Using Cloud Music's core delivery reach platform Link as the offline traffic entry, combining external service providers and internal business management demands to build Push-specific basic capabilities.

Traffic Allocation: Android manufacturers categorize notifications according to Google Android O standards, dividing non-essential push messages into marketing categories with daily per-device limits (about 2 per day). They implemented: 1) Message strategy management by Link position channels (notification vs marketing Push); 2) Message content classification adapting to manufacturer standards; 3) Business-specific frequency control capabilities; 4) Non-immediate notification secondary touch; 5) Marketing Push frequency control by user segments.

Push Open Rate Improvement: They refined APP message reception settings to provide do-not-disturb time period switches and guided users to open system push permissions through in-app prompts, achieving a 2 percentage point improvement in overall open rate.

Push Channel Optimization: 1) Added Honor push channel - found that new Honor device users couldn't receive pushes because Honor separated from Huawei; 2) Identified device ROM types to initialize appropriate push channels (e.g., MIUI devices for Xiaomi push); 3) Connected third-party push service providers for non-mainstream manufacturers, improving reach rate from 0.09% to about 10%; 4) Implemented push failure attribution to identify issues like Xiaomi's daily frequency limits; 5) Utilized push service provider features like large image templates; 6) Maintained device validity by cleaning invalid device tokens based on manufacturer feedback and calculating inactive devices not exposed in one year, saving 40GB storage and improving delivery efficiency.

Strategy Enhancement

The personalized Push decision process for users involves: 1) Matching resources based on运营 rules (e.g., songs hearted in last 30 days); 2) Combining resources with configured creatives to generate final Push content; 3) Selecting the best content through combination optimization strategies (Random Pick, Resource Priority, Creative Priority, Resource+Creative Comprehensive Selection).

Phase 1 - Copy Familiarity Optimization: After analyzing creative click-through data, they found that adjusting copy to highlight user-familiar elements significantly improved CTR - some specific copy improved by over 100%.

Phase 2 - Creative Distribution Strategy: They discovered that low-quality creatives occupied most impressions while high-quality creatives had few exposures. After adjusting algorithm models to incorporate creative click exposure data and adapting resource algorithms for Push scenarios (prioritizing familiar resources over "fresh" ones), the creative priority bucket's click rate improved by nearly 80%.

Phase 3 - Improving High-Quality Creative Supply: They found significant head effects where top 3 creatives contributed 30% of clicks, causing data volatility when these creatives hit frequency limits. They used AIGC technology with small-sample prompts to generate hundreds of new creative candidates from three directions, which were then refined by operations, resulting in more stable daily exposure trends and creating "viral" creatives that drove total clicks to project highs.

Results

After nearly two months of optimization, personalized Push overall CTR doubled from the project's initial average. The final click user count nearly doubled. The optimization chain from entry integration to system optimization to strategy enhancement provided a solid foundation for each subsequent phase, achieving significant business results.