Skip to main content

Conclusion

Recommendations to BFB​

Based on the architectural analysis, the following strategic recommendations are provided to Big Friendly Bank:

  1. Prioritize Flink over Legacy ETL: Do not attempt to adapt existing batch ETL tools for this use case. The latency requirements of POS financing are incompatible with batch windows. Adopt Flink as the standard for all real-time decisioning.
  2. Implement Caching Strategies: External API calls cost money. BFB should implement a caching layer within the Flink lookup (e.g., using lookup.cache options). If an applicant clicks "submit" twice, or applies at two different stores within an hour, the system should use the cached credit score rather than paying for a second API call.
  3. Enforce Data Contracts Early: Use Schema Registry to strictly define the LoanApplication schema. Publish this schema to the POS partners. This pushes data validation to the edge, ensuring that the BFB decision engine never wastes cycles processing malformed applications.
  4. Security First: Roll out CSFLE immediately. In the event of a cloud security breach, encrypted data without the keys is useless to attackers. This protects BFB’s reputation and customer trust.

Conclusion​

The proposed architecture transforms Big Friendly Bank from a legacy institution into a real-time digital lender. By leveraging Confluent Cloud for durable event storage and Apache Flink for asynchronous, stateful processing, BFB can effectively solve the "Unknown Applicant" challenge. This solution eliminates the trade-off between speed and accuracy, allowing BFB to assess credit risk in real-time without stalling the customer at the Point of Sale. The inclusion of Cluster Linking for disaster recovery and CSFLE for security ensures that this speed does not come at the cost of reliability or compliance. This architecture provides a scalable, future-proof foundation for BFB’s growth in the competitive BNPL market.