The challenge
Our client grappled with poor address data quality, resulting in higher operational costs due to delivery delays. The existing system, reliant on a dropdown address selection, struggled with vast address volumes for specific areas, forcing the customer to scroll. This led 4000 daily users to manually enter addresses, increasing the likelihood of errors.
Faced with the increasing reliance of stakeholders on accurate data, our client understood the importance of a new solution. The implementation of the new solution should avoid disruption of ongoing operations. That is when the retailer asked VirtusLab for assistance.
The solution
VirtusLab simplified the address entry process by introducing a user-friendly single-search field, eliminating the complex two-step postcode-address selection. This upgrade enabled a dynamic search-as-you-type feature, providing a fuzzy search for queries including typos.
We used AWS ElasticSearch to optimize system performance in an iterative approach. This was the foundation for the next steps, which included rewriting our client’s services in Java and launching them on Kubernetes to establish a unified platform. We further improved the platform's service architecture to ensure better integration.
The next stage centered on strengthening the retailer’s team and processes. We cross-trained our client's staff to broaden their skill set and implemented consistent coding practices. With a capable team and streamlined processes, we integrated data from diverse sources across various countries.
At last, we prioritized data security and governance. We enacted procedures to protect data and established governance protocols to guarantee its availability and accessibility.
The results
VirtusLab’s solution reduced the amount of manually typed addresses by 75%, leading to enhanced accuracy and reduced delivery costs. We also designed the solution to be applicable beyond addressing issues, such as address comparison for fraud detection and database migration.
As ElasticSearch requires adjustments, we iteratively refined the solution to meet the client’s expectations. This led to:
- Indexed Documents: Decreased by 70%
- Infrastructure Costs: Reduced by 15%
- Response Time Latency (P99): Improved by 50 times
- Search Result Accuracy: Increased from 60% to 90%
The tech-stack
Languages: Java, C#
Database technologies: Elasticsearch, CosmosDB, MongoDB
Infrastructure: Kubernetes, Microsoft Azure