OPPO’s Unified Modeling Strategy for App Distribution: Balancing Cost Reduction and User Value

The interview begins by highlighting the industry‑wide pressure to reduce costs while maintaining user value, noting that sacrificing user experience for efficiency can jeopardize customer retention in a competitive market.

Lai Hongke, General Manager of OPPO’s Internet Application R&D Platform and Search Algorithm Department, describes OPPO’s app distribution ecosystem, which includes a software store and a game center serving vastly different user bases and business goals.

Because the distribution scenarios span finance, logistics, travel, e‑commerce, gaming, and social categories, the data is extremely sparse and fragmented, making traditional recommendation techniques ineffective.

To address this, OPPO built a “full‑scene unified modeling” system that shares global features and samples across all distribution channels, expanding its feature store from 2 TB to 30 TB and increasing the number of features from 100 million to over 10 billion.

The solution leverages a Multi‑Task Mixture‑of‑Experts (MMOE) architecture, allowing shallow and deep conversion goals to share embeddings, dramatically reducing the number of CVR models while improving conversion rates, especially for high‑value actions such as game payments.

OPPO also introduced the oCPX capability, an intelligent bidding system that lets advertisers set explicit optimization targets (views, downloads, registrations, payments) and automatically adjusts bids based on predicted conversion rates, thereby improving ROI and reducing user disturbance.

Through successive versions of its intelligent compute platform (V1.0 → V3.0), OPPO applied traffic throttling, user segmentation, and “traffic fraud” detection to prioritize high‑value users, achieving up to 20 % higher traffic capacity and a comparable increase in revenue without additional hardware.

Overall, the interview demonstrates how OPPO integrates algorithmic advances, data engineering, and operational tactics to achieve cost efficiency while continuously enhancing user‑centric value in its app distribution services.