Intelligent Growth Algorithms and Their Applications in the Smartphone Industry – OPPO Andes Smart Cloud

Industry Background The smartphone market has reached saturation with over 1.065 billion online users, a declining growth of new users, and longer device replacement cycles. Manufacturers face challenges such as limited incremental demand, intense competition, and the need to explore high‑value markets, increase ARPU, expand to wearables or automotive, and improve channel efficiency.

Algorithm Architecture OPPO’s growth algorithm consists of five layers: basic data, data construction, feature profiling, model building, and application scenarios. Basic data include phone status, product attributes, order data, marketing interventions, real‑time behavior, and creative assets. Standardized processing creates a user flow graph linking phones to natural persons.

Model Construction The team employs various models: uplift (causal) for marginal marketing gain, PU‑learning for precise audience selection, multimodal understanding (ViLT) for text‑image tasks, PID for automatic control, AIGC (Stable Diffusion) for creative generation, and CTR/CVR predictors for click‑through and conversion estimation.

Application Scenarios

1. AIGC‑Driven Content Supply By integrating CLIP‑based image and VILT‑based text encoders, the team generates high‑quality marketing creatives, improving click‑through rates by about 15%.

2. Multi‑Scene, Multi‑Goal Multimodal Recommendation Using a ViLT‑based multimodal backbone and a multi‑objective ranking framework, the system optimizes CTR, CVR, and other business metrics, achieving over 20% CVR lift.

3. Causal Inference for Precise Audience Uplift modeling (Two‑Model, Single‑Model, Direct‑Model) combined with PU‑learning and graph learning identifies high‑value users, improving ROI by 11.3%.

4. Precise Advertising (RTB) Real‑time API (RTA) and Real‑time Bidding (RTB) pipelines use multi‑task MMOE‑style models with CTR, conversion, and device‑upgrade probability features, calibrated via tree‑based binning, resulting in ~25% ROI improvement.

Summary and Outlook Growth in the smartphone sector relies on defining a north‑star metric (e.g., new‑device activation, retention, ROI) and aligning downstream KPIs such as click‑through, win‑rate, and conversion. Future work will focus on refining causal models, handling feature distribution drift, and enhancing cross‑channel budget allocation.

Q&A The session addressed hyper‑parameter optimization (PSO vs. reinforcement learning), multi‑objective integration in ranking, handling feature drift, data collection for uplift models, and the trade‑offs of exposure bias correction.