Skip to main content
CUSTOMER STORY

Leveling up fraud detection in online gaming

Quago uses Databricks to deliver more authentic and fair gameplay

4x

Faster SQL queries compared to a vanilla Parquet lake

25%

Estimated reduction in gaming company server costs

10s of millions

Potential dollar savings by reducing in-app purchase losses due to fraud

PLATFORM USE CASE: Delta Lake,Databricks SQL
CLOUD: AWS

Quago is leveling the playing field within the online game industry with an in-game behavioral analytics platform designed to mitigate cheating and fraudulent activity to ensure the best gaming experience possible. Through the analysis of players’ device sensor data, Quago can quickly identify patterns and provide game companies actionable insights around cheating, churn and player behavior. Seeking cost-effective scaling and infrastructure management, Quago migrated to the Databricks Data Intelligence Platform. As a result, they experienced significant improvements in performance and scalability, achieving 4x faster SQL queries compared to their previous lake infrastructure. This resulted in reduced server costs for their customers and potential savings of tens of millions of dollars by lessening in-app purchase losses due to fraud. Additionally, Quago’s ability to address fraud quickly and accurately enhances player trust and loyalty, ultimately improving customer lifetime value for gaming companies.

Inability to scale data ingestion and accuracy hinders gaming fraud detection

Through advanced ML technology, Quago analyzes players’ device sensor data to provide gaming companies with real-time insights on cheating, churn prediction, user acquisition (UA) fraud and other behavioral patterns. Of course, doing so requires the ingestion and management of vast amounts of data to ensure high accuracy in detecting cheating and maintaining player community integrity. “We gather all of your touches, accelerometers, gyroscopes, battery sensors and more to then generate a data layer that helps us understand human behavior in a very granular way,” said Dan Blechner, Co-founder and Chief Product Officer at Quago. “Each swipe on the screen generates more than 200 data points.”

Cheating detection is a significant offering for Quago’s customers. Every successful game is a target for fraudsters because they can make money from selling services to gamers to cheat and win tournaments. This hurts the experience for other players and company revenue, and it can ultimately become an existential threat to the game through damage to the core gaming experience.

Quago faced challenges managing their vast amounts of data throughout the pipeline. They had to ensure high accuracy in detecting cheating and achieving comprehensive recall to differentiate between various types of players for appropriate actions — such as outright banning or simply educating them — to best preserve the player community. “Let’s say a successful game is around 5 to 10 terabytes per day. That’s just for one game. We’re ingesting data from multiple games at the same time,” explained Blechner. “Server cost becomes a big challenge when managing all this data. It’s crucial to be able to process all of this data within a budget that’s relevant for the magnitude of the problem.”

Quago originally used a vanilla Parquet lake, which was costly to scale due to performance, storage efficiency and query optimization challenges. Struggling with cost-effective scaling and the resource-intensive nature of managing infrastructure for both performance and availability, Quago’s engineering team looked for a solution that could solve these problems for their small team.

Maintaining a cost-effective, scalable data lake with Databricks

Quago must provide real-time results — in less than a second in some cases — and unparalleled accuracy and availability. Their platform also has to clean analytics quickly. If Quago can’t meet these requirements, then gaming companies will have skewed data. For example, when an attack occurs, it creates a spike of new users that typically disappear without making any purchases. “Our customers are trying to make their decisions based on data. Once you can’t trust the data, such as basic metrics like one-day or seven-day retention, you can’t make good decisions,” offered Ran Arieli, Co-founder and Chief Technology Officer at Quago.

Quago needed the best mechanism possible for monitoring, alerts and managing the complexity of Apache Spark™. “As I see it, it’s impossible without the Databricks Data Intelligence Platform. Our technical team has vast experience with complex data systems, pipelines, high-speed production, managing schemas and schema changes, and handling the dependencies between the different components for data processing. Even so, doing that without a platform like Databricks is nearly impossible — especially if you want to maximize ROI without exploding your cloud budget,” said Roy Green, Head of Engineering at Quago.

In the early stage, Quago’s engineering team migrated everything into Databricks Delta Lake. “This turned out to be an amazing decision,” Green laughed. “We launched with all of our customers for all of their use cases. We’ve seen a big, positive impact around performances and scammer management.”

And as far as needing a cost-effective solution? “We’ve been able to keep a very small and efficient team. Without Databricks, we would need a lot more people,” added Arieli. “Plus, the cost would be much higher on the cloud side.”

Better performance that saves gaming companies tens of millions

The ability to run Spark in both interactive and automative modes is a critical requirement for Quago’s production, and Green’s engineering team achieves this seamlessly. “Delta Lake is an essential component in our system. All of the data, including raw data, biometric sensor data, metadata and the game data itself, is managed there,” said Green. “On top of Delta Lake, we have a Spark logic in order to build the models, retrain them, do research and launch more models.” What would managing Spark with this amount of data be like without a managed platform like Databricks? “A nightmare,” answered Green.

Green believes that his team has achieved 4x faster SQL queries compared to using a vanilla Parquet lake. He noted, “The development time and time to market, especially in the areas of managing the jobs and changing schemas, took much longer with vanilla Parquet. In general, our ability to use Databricks to update our models is amazing. We update our models per customer, per app, sometimes as much as three times a week.” Quago customers benefit a lot from that because all of the retraining and refreshes help make the models even more accurate.

When Quago wins, their customers win. According to Blechner, “We’ve seen a reduction in our customers’ server costs. Typically, fraudulent activities create way more activities than the usual player, and this requires more from the gaming server.” The Quago team has seen a reduction of up to 25% in their customers’ server costs.

Once the Databricks Platform is fully deployed, Blechner believes that Quago’s ability to mitigate fraud will reduce customer in-app purchase losses by tens of millions of dollars. Most importantly, Quago supports gaming companies in their mission to provide an excellent player experience. With an in-game analytics platform built using the Databricks Data Intelligence Platform, Quago can help companies with immediate scammer resolution. Addressing problems quickly and accurately creates trust and confidence with players, boosting customer loyalty and lifetime value.

Learn more about the Quago platform.