How We Automated Driver Withdrawals: Architecture Evolution & Risk Controls

1 Background

As the volume of Huolala’s business grows, the withdrawal flow, the main channel for fund outflow, has expanded in scale and amount. Initially it relied on manual approval for safety, but rapid growth made manual checks unable to keep pace, prompting a need for automation.

2 Withdrawal Process

The process consists of four stages:

Wallet credit: Drivers receive income via transactions, rewards, claims, etc., credited to the internal wallet.

Withdrawal request: Drivers initiate a withdrawal.

Financial approval and payment: Finance manually samples approvals.

Bank payout: System submits the request to the bank for final disbursement.

Risks exist at each step, such as fraud, system exceptions, limited sampling, and bank failures.

3 Architecture Evolution

3.1 PHP era – “silo” architecture

Early stage focused on rapid feature delivery; each endpoint built its own withdrawal logic, leading to duplicated code, tight coupling, and maintenance difficulty.

3.2 Phase 1.0 – Platform architecture

Business growth required higher performance; the stack migrated from PHP to Java and a unified platform provided common services, improving stability, scalability, and development efficiency. Withdrawal approval still relied on manual sampling.

3.3 Phase 2.0 – Automated payouts

To replace manual approval and enable weekend/holiday payouts, two key problems were addressed:

Fund security: Add a final interception mechanism to block abnormal funds caused by business or technical issues.

Stability: Ensure reliable withdrawal flow to avoid driver dissatisfaction and complaints.

3.3.1 Fund security

Risk control is divided into fund‑risk and internal‑control. Fund‑risk automates the manual sampling rules into system rules for full‑order checks. Internal‑control adds technical safeguards such as idempotency checks and data consistency verification.

3.3.2 Stability

Monitoring dashboard covers five modules: main process metrics, exception points, risk rules, business indicators, and application performance.

Alerting includes fund‑control hits, timeout without payout, and risk‑downgrade alerts.

Tools such as black‑white‑gray lists, payment revocation, risk downgrade, abnormal alerts, and manual fallback ensure comprehensive protection.

4 Benefits

Improved fund safety: Automation covers 100% of withdrawals, intercepting risks and recovering losses worth tens of thousands.

Enhanced system capability: Supports automatic payouts on weekends/holidays.

Better driver experience: Satisfaction up 25%, complaint rate down 29%.

Cost reduction: Lower manual finance effort, higher operational efficiency.