We are beta testing Predictive Routing, and I’ve been analyzing the WebRTC connection model.
In our A/B test, agents using the persistent WebRTC connection (always connected to the media server) had 200ms faster answer times compared to agents using the per-call connection model (new WebRTC session per interaction). The persistent model consumes more bandwidth idle, but the ML model’s routing predictions are wasted if the agent takes 2 extra seconds to establish the media path.
On BYOC Premises Edges, the persistent connection model puts a continuous load on the Edge DSP.
Each persistent WebRTC session consumes a DSP channel even when idle. If you have 300 agents in persistent mode, you are permanently consuming 300 channels on the Edge hardware. For our remote site Edge appliances with only 500 channels, that leaves only 200 channels for actual calls.
Does the WebRTC connection model affect digital channels at all?
If an agent handles only chat and email, do they still establish a WebRTC connection, or does the platform skip the media negotiation entirely for non-voice interactions?
For callbacks, the per-call model is actually preferable.
The callback is scheduled for a future time. If the agent has a persistent connection, the system might route the callback to them based on their idle media state, even if they stepped away briefly. With per-call connections, the system only routes when the agent actively signals availability by establishing a fresh session.