Collections that run themselves until a human is needed.
- Year
- 2023
- Role
- Backend services, workflow automation
- Timeline
- 2023 · Ippopay
- Stack
- Node.js, MongoDB
The system
Overview
The payments and collections backbone at Ippopay: QR payment processing, loan-collection workflows, and reminder automation orchestrated over $10M+ a month, built so operations scale with queues instead of headcount.
problem
The problem
Collection cycles ran on manual follow-up. Agents worked call sheets, payments reconciled slowly, and every increase in volume meant hiring more people rather than moving more money.
strategy
Strategy
Automate the predictable path. Reminders, escalations, and settlement flows become queued, retryable workflows; humans enter only on exception, where judgement actually matters.
development
Development
Idempotent webhook ingestion for payment events, SQS-backed reminder workflows with escalation ladders, and a merchant workflow state machine that keeps every account's position explicit and auditable.
- $10M+ in monthly volume orchestrated
- QR payment processing and reconciliation
- Escalation ladders with quiet-hours rules
- Idempotent, replay-safe webhook ingestion
results
Results
Collection time halved. The same operations team handled multiples of the original volume, and reconciliation moved from days behind to same-day.
Key takeaways
Queues are an org chart: they decide what humans never see.
Idempotency is the difference between automation and incident.
Money systems earn trust through auditability, not speed.