Skip to content
All work
03Streaming / Location Analytics

Routing field agents by signal, not guesswork.

Year
2023
Role
Backend architecture, streaming pipeline
Timeline
2023 · Ippopay
Stack
Google Maps, Redis
100K+
events / day
35%
travel time cut
Stack
Google MapsRedisWebSocketsNode.js

The system

field eventsws ingestredis streamsgeo-fenceroutingagent app

Overview

A field intelligence platform for collections teams: live location events, geo-fenced zones, and behavioural data fused into routing that tells every agent where to go next, and gives managers a live map instead of an end-of-day spreadsheet.

01

problem

The problem

Field agents covered loan-collection beats on intuition. Routes overlapped, high-value visits were missed, and managers had no live view of coverage: the source of truth was a spreadsheet reconciled after the day was already over.

02

strategy

Strategy

Treat the field as a stream. Every location ping, geo-fence crossing, and visit outcome becomes an event that updates the routing picture, so the day's plan can adapt while the day is still happening.

03

development

Development

WebSocket ingestion feeds Redis streams; a geo-fence engine evaluates zone crossings in real time; a routing layer scores each candidate visit by distance, payment history, and time windows before it reaches the agent's app.

  • 100K+ location events a day, sustained
  • In-memory geo-fence evaluation on Redis
  • Visit scoring from behavioural data
  • Live coverage dashboard for beat managers
04

results

Results

Travel time per completed visit fell 35%. Managers moved from end-of-day reconciliation to live course-correction, and coverage stopped depending on who knew the neighbourhood best.

Key takeaways

01

Routing is a data problem before it is an algorithm problem.

02

Make the field visible first; optimisation is easy once you can see.

03

Streams beat batch when the decision window is measured in hours.