Conversational Analysis and Delivery-Apparatus Signals for Verifiable Outcomes

A customer who needs bureaucratic support does not want a brilliant answer. They want their problem solved, to be told that it is solved, and for the conversation to end and never be spoken of again. That one fact is what separates voice agents for customer service from every other thing people are doing with language models.

Almost everyone building voice capability is working next to it rather than on it — on the engineering of delivery, and on making that delivery cheaper and faster. We work on what actually improves the way a model handles a conversation, and whether the customer's issue gets resolved.

An agent that answers a business's phone and is happy to hold an open-ended conversation, or complete tasks without verification, isn't a feature — it's exposure. We propose that labs build a variant of their models specifically for customer support, whose RL is built on reward signals derived from our conversational analysis techniques. These signals help determine whether the correct booking landed, whether the escalation was the right call, whether the caller will have to ring back the next morning — and by the same construction, they let us build a live failure monitor that flags conversations going wrong in real time.