The problem we set out to solve
Our clients ran separate phone lines in every branch, with no call recording and no way to see who was waiting in a queue. Hardware PBX systems were expensive and hard to change.
Architecture at a glance
Our call center runs Asterisk clusters behind a SIP proxy on Kubernetes. Calls reach agents through WebRTC in the browser or through desk phones, and every call event flows into a live dashboard.
- SIP proxy and Asterisk media servers
- WebRTC softphone in the browser
- Event stream to dashboards and CRM
Designing for unreliable networks
Many agents work on mobile data. We tuned codecs for low bandwidth, added jitter buffers and made the softphone reconnect a call within seconds after a network drop.
What we learned in beta
Queues and call recording mattered more to customers than any advanced feature. We moved both into the Starter plan and delayed features fewer than one in ten teams used.
Key takeaways
- Reliability on bad networks beats feature count
- Queues and recording are table stakes
- Ship to a small beta group before adding features