Optimizing online gaming with player personalization

Limited data visibility slows decision-making

Known for delivering immersive gaming experiences worldwide, FunPlus faced increasing complexity in managing their vast data ecosystem. As the company scaled, they needed a more robust data infrastructure to support business intelligence, personalization and predictive analytics. However, their legacy data warehouse, Snowflake, couldn’t meet the company’s growing requirements. Data sources were diverse — spanning PostgreSQL, S3, MySQL, MongoDB and external APIs – creating integration issues that further complicated efficiency. 

For FunPlus, BI dashboards represented the most crucial use case but required timely and reliable data. Unfortunately, Snowflake demanded significant manual intervention to ensure accuracy and consistency. FunPlus data practitioners had to create multiple access credentials, add partitions and write custom code to manage disparate data sources. Chen Chao, Data Platform Lead at FunPlus, explained, “Our data was everywhere, and each source had unique access points. These inefficiencies made data access and analysis time-consuming, which limited our ability to support rapid data-driven decisions.”

Another challenge was the need for player insights using ML. FunPlus wanted to leverage AI to generate predictive player tags, but the legacy system made it challenging to deploy these models effectively. Data science workflows needed a more flexible environment to handle complex ML tasks and support Python scripting to avoid duplicate SQL maintenance. Previous workflows on SageMaker involved multistep processes that were inefficient and labor-intensive. The limitations in ETL processes and lack of automated error handling further complicated FunPlus’ ability to perform advanced analytics at scale.

Data accuracy, speed and reliability with Databricks

FunPlus needed to streamline data access, automate ETL and support ML workflows in a unified environment. With the help of Koantek, they migrated to the Databricks Data Intelligence Platform to centralize data management, perform federated queries and enhance team collaboration. 

FunPlus uses Delta Lake to connect scattered data sources within a single platform for reliable storage and enhanced data management. Delta Lake’s data versioning and schema enforcement features enabled FunPlus to maintain clean, accurate data that can easily be shared across teams and used to fuel real-time dashboards. 

As a fully serverless-enabled platform, FunPlus has benefited from instant and elastic compute to power 80% of their workloads. As a result, they have benefited from superior performance without having to wait for infrastructure resources to become available to run jobs or overprovisioning resources to handle spikes in usage. The team is also using SQL Serverless to empower their data analysts and operations team to run complex queries and quickly generate critical insights. By consolidating their SQL queries within Databricks, FunPlus reduced the need for additional tools and provided a consistent experience across teams. Zhang Zhizhi, Tech Lead at FunPlus, said, “Databricks’ serverless capabilities allow us to manage and analyze large volumes of data with improved speed and accuracy, without worrying about resource allocation. Analysts can perform deep dives into data and create detailed reports without relying on multiple systems.”

Unity Catalog provided FunPlus with centralized metadata management and data lineage tracking, making it easy for teams to understand the source and structure of data tables. This transparency facilitates better collaboration, as team members can easily access data and track dependencies without extensive documentation. Unity Catalog also supports FunPlus’ security needs by enabling role-based access controls that allow them to scale democratization confidently. 

For ML applications, FunPlus uses MLflow to empower data scientists with simple experimentation among different models. They can track performance metrics and deploy models into production, all within the Databricks environment. This allows FunPlus to iterate quickly and deploy insights directly, impacting player engagement and retention.

Additionally, FunPlus recently started using generative AI (GenAI) tools within Databricks such as AI/BI Genie and Streamlit to accelerate data-driven decision-making. AI/BI Genie is integrated into their BI visualization platform via an API, providing an interface for business users to easily query operational metrics, props production and order and sales data using natural language, greatly improving the efficiency of accessing data across the organization. FunPlus’ data scientists are also experimenting with Streamlit to quickly and flexibly create interactive data applications using Python and share them with business users. In the future, FunPlus plans to explore using large language models (LLMs) to generate Streamlit code, paving the way for a “text-to-data app” workflow.

Improving efficiency and lowering costs

Since moving to the Databricks Data Intelligence Platform, FunPlus has seen measurable improvements across several key areas. Core to their ability to deliver value to the business is how Databricks has been instrumental in enhancing data engineering productivity — to the tune of a 48% improvement. With increased engineering velocity, they’ve seen the most impactful changes in FunPlus’ ability to deliver reliable, real-time data to their BI dashboards. Chen explained, “By unifying data management, FunPlus now has a consistent, accurate view of operational metrics that teams rely on to guide strategic decisions. This newfound data reliability has reduced the need for manual data checks, improving overall decision-making speed and accuracy.” FunPlus estimates that BI-related processes are now 20% more efficient, and operational costs have been cut by 40%.

Personalization has benefited from Databricks’ scalable data management and stable data APIs. FunPlus can deliver personalized player experiences more effectively during special in-game events. With Databricks, FunPlus has achieved an API stability of 99.9%, enabling a seamless and engaging player experience. This reliability is critical in driving player retention and engagement, as users receive consistently personalized interactions, from exclusive offers to tailored content. That stability translates directly into player satisfaction and increased revenue from in-game purchases.

Using the ML capabilities on Databricks, FunPlus has built predictive models that offer valuable player insights. With efficient model management, training and deployment in a streamlined workflow, FunPlus implements models for player behavior predictions, such as churn likelihood and segmentation. These insights allow FunPlus to take preemptive actions, targeting at-risk players with personalized content that encourages continued engagement.

With a strong foundation on the Databricks Data Intelligence Platform, Zhang said, “FunPlus will continue learning player behavior and preferences, providing personalized growth advice and consumption recommendations within games, helping improve player experience, product retention rates and revenue.”