Arka Rewards
F&B
Client
Arka
to Explore
Building the data foundation for faster decisions and personalized rewards.
- Data Engineering
- Data Warehouse
- Analytics
- AI/ML
Preparing Arka Rewards’ data infrastructure for continued growth.
Arka Rewards is a loyalty program running across Arka’s growing portfolio of brands. As the program expanded and supported more members and transactions, data volumes grew with it.
Analytics became increasingly important for monitoring engagement, evaluating performance, and informing reward decisions. Arka also planned to use this data to deliver more personalized reward recommendations.
However, reporting still ran on the same databases processing live platform activity. As analytics and operational workloads grew, they began competing for the same resources. Arka partnered with us to build an analytics infrastructure that could scale with its growing data needs.
Separating analytics from production without disrupting live operations.
Dashboards that once loaded in seconds began taking up to three minutes as they compiled multiple charts and queries. Individual queries behind each chart could take up to a minute and a half. These delays slowed access to the insights teams needed for business decisions.
To address this, we needed to move reporting workloads off the operational databases without interrupting live transactions. The new setup had to preserve historical data, maintain accuracy and freshness, and support both ongoing analytics and AI-powered reward personalization.
Organizing reliable data for analytics and AI.
We implemented a modern data warehouse on Google BigQuery as a dedicated environment for reporting and analysis. Separate from the operational databases, it consolidated current and historical data without placing analytics workloads on the live platform.
We organized the warehouse using a three-layer structure known as medallion architecture. The Bronze layer preserved raw production data as a complete historical record. The Silver layer cleaned and standardized the data by removing duplicates, aligning inconsistent formats, and adding calculated fields such as customer lifetime value. The Gold layer organized it into business-ready datasets for reporting, analytics, and AI.
To improve query performance, the Gold datasets were structured around dates and fields frequently used in reports. This brought individual query execution down from up to a minute and a half to roughly 500 milliseconds, allowing dashboards that had taken three minutes to load in under one.
Together, these layers gave Arka a consistent source of reliable data while allowing each use case to access the level of detail and performance it required.
Keeping data accurate and up to date.
To keep the data warehouse synchronized with production, we built automated pipelines using change data capture, a method that transfers only new or modified records instead of repeatedly copying entire tables. This helped reduce processing time and data transfer costs.
Google Cloud Composer orchestrated the workflows, while Dataform handled transformations across the Bronze, Silver, and Gold layers. Hourly and daily update schedules were matched to how frequently each data source changed, with automated alerts for pipeline failures and data quality issues.
Using member data to improve reward recommendations.
With validated, business-ready data available in the Gold layer, we built a Python-based recommendation engine using model-ready inputs such as transaction frequency, redemption preferences, engagement levels, loyalty tier movement, and churn risk signals.
The engine combined two recommendation methods: content-based and collaborative filtering. Content-based filtering identified rewards similar to those a member had previously engaged with. Collaborative filtering used patterns from members with comparable behavior to identify other relevant options.
Together, these methods helped determine which rewards and merchants to recommend based on each member’s preferences and behavior.
Connecting the engine to a clean and standardized data warehouse allowed each recommendation to draw from more reliable member signals.
Faster analytics and stronger member engagement.
The new data platform separated analytics from production, significantly reducing dashboard load and query execution times without adding load to the customer-facing platform. It also provided the reliable data needed for AI personalization, contributing to stronger member activity after the recommendation engine launched.
+180x
faster query execution, from up to 1.5 minutes to roughly 500 milliseconds
+21%
increase in member transactions after launching AI personalization
38,1%
personalized reward redemption rate, up 6.3 percentage points from 31.8%