How Data‑Driven Flywheels Power User Growth: Insights from Volcengine

Introduction

From a data professional’s viewpoint, this article shares methods and practices for user growth, emphasizing the need to navigate uncertainty and find stable paths for product expansion.

Five Key Themes

Garden of Forking Paths – exploring how choices and environments create uncertainty.

Data‑Driven Flywheel – connecting business and data to form a growth engine.

Sisyphus’ Stone – confronting relentless challenges.

Random Walk Fool – respecting randomness and avoiding forced explanations.

Q&A – addressing practical questions.

1. Information Management: From Chaos to Order

Before the internet, computers were used for scientific calculations. Early MIS (Management Information Systems) evolved from material management to enterprise resource planning, gradually turning chaotic information into orderly systems.

2. Information Distribution: From Low to High Efficiency

After managing information, the next step is distribution. Early directories listed information, but the explosion of web content made them inefficient, leading to search engines that rank pages based on quality and traffic. Users must craft keywords, while content creators invest in SEO. Recommendation systems further automate personalized delivery based on user‑item interactions.

3. Information Production: From Complexity to Simplicity

In recommendation and search systems, after retrieving information, it must be recognized, understood, and integrated, often requiring manual effort. With the rise of large models and AIGC, many downstream tasks can be automated, making information production faster and simpler.

4. Product as Entropy Reduction

Product development reduces entropy by turning disorder, inefficiency, and low performance into order and value. Iterative product improvements lower entropy, enhance user experience, and drive growth.

Data‑Driven Flywheel

Volcengine’s A/B testing leader outlines a data flywheel that centers on data consumption, linking business and data wheels to create a growth engine. Key practices include:

Using AB experiments to select high‑impact retention solutions (e.g., a 57% lift in retention).

ByteDance’s data‑driven culture: OKR management, dashboards, and a middle‑platform plus data‑BP architecture that provides tools and infrastructure for rapid data consumption.

Defining evaluation metrics (e.g., SLA = 0, demand satisfaction ≥ 90%, analyst coverage ≥ 80%, NPS ≥ 70%).

The flywheel consists of data consumption, business‑driven applications (AB experiments, agile actions, value enhancement), and data asset management (reporting, quality assurance).

Growth Case Study: Short‑Video Playback Boost

A vertical content platform aimed to increase short‑video plays to improve user activation. Using DataFinder for advanced analysis, two problems were identified: low playback rate in the homepage recommendation flow and low per‑user plays in the feed for new users. AB testing led to two optimizations:

Homepage redesign with short‑video cards, achieving >100% increase in playback.

Feed flow redesign with new‑user guidance, achieving >140% increase in per‑user plays.

Combined, these changes raised total playback by 300%.

Growth Stages and Retention

Growth evolves from acquisition and activation (Stage 1.0) to retention‑focused strategies (Stage 2.0), where retention becomes the north‑star metric. Effective retention tactics include:

Functional improvements and incentives that give users a sense of growth.

Content‑driven retention through diverse, engaging material and personalized recommendations.

Social interaction to build community connections.

User incentives such as gamified check‑ins or discounts.

Modern products operate in a 3.0 stage where growth is a continuous loop, requiring cross‑platform integration, resource consolidation, and innovative collaborations.

Q&A

Q1: How to evaluate an AB test metric and avoid Simpson’s paradox? A1: Simpson’s paradox arises when aggregated results hide subgroup performance differences. Proper attribution and subgroup analysis are essential.

Q2: What is DataFinder? A2: DataFinder is an interactive analysis tool that embeds common models (e.g., retention) and enables fast, click‑based analysis without writing SQL.

Q3: How should data science teams collaborate with business? A3: Teams should share metrics, form agile squads, and prioritize data‑driven decision making.

Conclusion

By continuously reducing entropy, leveraging data‑driven flywheels, and iterating based on measurable insights, products can achieve sustainable user growth and a healthy commercial model.