If you're building real-time communication into a healthcare product, you'll hit a wall. Not in week one, not in month six, but somewhere around the point where concurrent sessions go from hundreds to thousands and the CPaaS invoice starts looking like a salary.
This post is a practical breakdown of when to stay on CPaaS, when to migrate to custom RTC infrastructure, and what the architecture actually looks like under load.
TL;DR
- CPaaS is the right starting point for most teams. Don't over-engineer day one.
- The transition from CPaaS to custom RTC is a phase, not a switch. Hybrid is normal.
- Capacity planning in healthcare must size for peak burst, not average load.
- The triggers for migration are usually cost curves, control limits, or compliance — in that order of frequency.
The communication stack you're actually running
Most healthcare comm stacks aren't a single channel. They're a sequence:
- Video consultations — WebRTC, the visible layer
- Voice (VoIP + PSTN fallback) — when video fails or isn't viable
- SMS / WhatsApp — high-volume reminders and nudges
- Secure in-app chat — async between visits
- IVR — first-touch routing and triage
- Store-and-forward async — patient updates, images, queries
Each channel works in isolation. The hard part is the layer between them — signaling, orchestration, and context persistence.
If your stack treats these as separate pipes instead of one continuous conversation, you'll feel it the moment a video call drops to voice and the agent has no idea what was just discussed.
When CPaaS makes sense
CPaaS is genuinely the correct answer when:
- Concurrent sessions are in the hundreds, not thousands
- Per-minute pricing is still in rounding-error territory
- You don't have an infra team that wants to run SFUs and TURN servers
- Speed-to-launch matters more than per-session cost optimization
This is most early-stage healthcare products. There's no shame in it. Building media infrastructure to save money you're not yet spending is a classic mistake.
Inside The Architecture Crisis: https://www.ecosmob.com/blog/scale-patient-communication-custom-rtc/
When custom RTC starts making sense
The transition usually shows up in three signals:
1. The cost curve
CPaaS pricing is linear. Every minute, every message, every relayed media stream is metered. At low volume the meter is invisible. At scale it dominates the infra budget — and the line item grows in lockstep with usage, not efficiency.
A useful rule of thumb: if your CPaaS bill is growing faster than your engineering headcount and you can predict next year's volume within ±25%, custom RTC is probably cheaper at steady state.
2. Control limits
When you can't tune what you need to tune, you have a ceiling problem:
- Can't tune SFU behavior for your specific session profile
- Can't relocate TURN servers regionally
- Can't enforce custom media routing logic
- Can't directly observe SFU-internal metrics
- Can't run isolated tenancy at the infrastructure layer
Most of these are fine — until they aren't. Once one of them is blocking a reliability or compliance requirement, you have a forcing function.
3. Compliance pressure
HIPAA + HITECH (US), PHIPA / PIPEDA (Canada), GDPR (EU patients). BAAs that constrain where data flows and who touches it. State-level laws that add regional rules. None of these are fundamentally incompatible with CPaaS, but the chain of custody is harder to audit when you're not operating the infrastructure yourself.
Capacity planning, in concrete terms
Healthcare RTC traffic is bursty. Plan for peak, not average.
Top comments (0)