How the agent's answers are evaluated
This article explains how we evaluate the AI phone agent. You learn how we measure every conversation through direct questions, models and automated checks.
The calls are followed up by evaluating every case; we ask the caller and a model compares the transcript with the agent's instructions. All data is processed in EU data centres.
Direct questions to the caller
We ask the caller if they got enough help. They reply during the call or via text message. You receive their actual response instead of internal ratings. Real Talk delivers CSAT for all interactions and surfaces trends along with areas to improve.
Reviewing every case
A model reads the transcript and checks the given instructions. It decides if the conversation meets your defined quality standards. We evaluate all individual interactions instead of random samples. The transfer rate shows how often the agent hands over to a human. Containment rate and First Contact Resolution track issues the agent resolves without help. You access your call data directly through our MCP. You analyse the conversations by asking questions. You own the insights without depending on us.
Continuous tests to find errors
We build automated checks for every task the AI phone agent must handle. These run constantly to identify errors when instructions change. This catches potential mistakes before callers notice a thing.
Feedback from customer service
We can follow the conversation after the AI phone agent transfers it to your human team. This reveals what the agent needed to know from the start. You can turn this extended tracking feature on or off anytime. You select the specific phrasing you want to mandate, and we implement it for your agent. Locked in expressions prevent concepts from drifting and stop answers from becoming unclear.