Deploying Code to Offline Stores Part 1: Building a Data Pipeline for Paperless Stores

Manually managing millions of price tags scattered across over 1,300 stores nationwide had reached its limits. Paper labels were costly to replace, inventory and price information was reflected late, and every promotion caused a surge in physical workload for store staff.

To solve this problem, we built an event-driven message pipeline centered on ESL (Electronic Shelf Label). The tech stack consisted of Kotlin, Spring Boot 3.x, MSK, SQS, Oracle, ElastiCache(Redis), and S3, and instead of simple API calls, we placed a message queue in the middle, designed to absorb large-scale traffic and unstable networks at individual stores.

The core was a structure that centralized source data in the cloud in a form suited to ESL, and delivered it consistently to devices at every store. This ensured that every store always saw the same data at the same point in time, and reduced the burden of querying the legacy DB directly.

On the operations side, devices were managed remotely through central monitoring and automatic retries. Even when a failure occurred at a store in Jeju Island, the system detected the issue and attempted recovery without an on-site visit, absorbing physical distance through technology.

Functionally, three things were important.

• Dynamic View rendering: Dynamically changing the ESL UI depending on the nature of the promotion, increasing the information delivery power of the limited screen
• Near real-time out-of-stock reflection: Detecting Stock Events to immediately reflect out-of-stock information on the screen, and automatically synchronizing online/offline data consistency
• Batch-based price changes: Reliably processing tens of thousands of price changes as scheduled batches during large-scale events, structurally blocking human error

The results were clear. Time spent on label management and inventory response was reduced by an average of 2 hours per day per store, and payment claims due to price entry errors were eliminated. In the end, this wasn't just about turning paper labels digital—it was about transforming stores into operating systems controllable through data.