Pre-Authorized and Asynchronous Payment in Urban Taxi Mobility Apps: Reducing End-of-Trip Friction with an Event-Driven Microservices Architecture and Impact Analysis Using Production Data

This study presents an “pre-authorized and asynchronous payment” approach developed to reduce the waiting time, uncertainty, and operational congestion caused by having to wait for payment confirmation at the end of a trip in urban taxi mobility applications. In the existing synchronous flow, delays caused by third-party payment services (including timeouts), network instability, and insufficient card limit/balance both slow down the passenger's exit from the vehicle and delay the driver's transition to the next trip. In the proposed solution, trip closure is decoupled from the payment process: the TripEnded event is published via Kafka, the Payment Service performs collection in the background, and the client flow is not blocked. On the authorization side, a Redis-based TTL cache reduces unnecessary repeated authorizations; and a settlement decision logic manages differences between the authorized amount and the final charged amount. In risky cases, the flow is redirected to 3D Secure; for failed collections, staged backup-card attempts, debt recording, and scheduled automatic collection (retry) are applied. In the evaluation using production data, the average client blocking time decreased from 46.33 seconds to 2.31 seconds, and P95/P99 values improved significantly. In field measurements, the vehicle exit (drop-off) time decreased from $\text{4. 6 7}$ seconds to $\text{1. 4 3}$ seconds; and the collection rate within the first hour increased from 38.56 % to 79.88 %. On the operations side, the number of calls decreased by 20.24 % as critical complaints dropped. The results show that an asynchronous “non-blocking collection” design strengthens both user experience and collection continuity. During gradual rollout, the pre-authorized payment infrastructure was used in about 25 % of card trips, and as this rate increased, friction at the moment of trip closure systematically decreased. Since reducing approval steps can increase malicious-use risk, device/GSM-based limits, anomaly detection, and fraud-flagging rules were introduced, and stricter verification policies were applied for suspicious transactions. The architecture was supported with observability (logs/metrics/alerts) and CI/CD processes to achieve a scalable trip-closure payment model.