Deploying a voice AI agent without a performance dashboard is like driving blind. However, there is a strong temptation to look at the wrong metrics: the total volume of calls handled, the average duration of conversations, the monthly cost of the platform. These figures say nothing about the value created for your customers or your business. Here are the 5 KPIs that really matter — and how to interpret them.
KPI 1 — The autonomous resolution rate (TRA)
Autonomous Resolution Rate
% of conversations resolved by the agent without human transfer
This is the main KPI. It measures the percentage of conversations that your agent resolves entirely without human intervention. A TRA of 65 % means that 65 conversations out of 100 are managed from start to finish by the AI. The other 35 are transferred to human agents.
How to improve it: analyze the reasons for the transfers. If 40 % of the transfers come from the same uncovered intent (for example, "request for extension of payment deadline"), create a new flow to cover this case. The TRA should improve by 3 to 5 points per month during the first 3 months of ramp-up.
KPI 2 — The post-interaction CSAT (Customer Satisfaction)
AI agent CSAT
Satisfaction rating collected at the end of the conversation
Customer satisfaction measured specifically after an interaction with the AI agent (and not after an interaction with a human) is the only indicator that tells you if your customers are having a good experience with the automation. Collect it via SMS or post-call IVR (one question out of 5: "To what extent did you get what you were looking for?").
An AI agent CSAT > 4.2/5 is excellent. Between 3.5 and 4.2, there are areas for improvement to identify. Below 3.5, your setup has significant issues (inadequate scripts, too high latency, poorly managed transfers).
KPI 3 — The cost per resolution (CPR)
Cost per Resolution
Total platform cost ÷ number of monthly autonomous resolutions
This KPI allows you to compare the actual cost of your AI agent to that of a human agent (3 to 8 € per resolution). Calculate it by dividing the total monthly cost of your solution (subscription + cost per minute) by the number of conversations resolved autonomously. A CPR of 0.25 € versus 5 € for the human represents a 20x ROI on each automated resolution.
KPI 4 — The human opt-out rate (TOH)
Human Opt-out Rate
% of callers who explicitly request a human agent
This KPI measures the proportion of customers who explicitly ask to speak to a human, often within the first few seconds of the conversation. A high TOH (> 25 %) indicates that your customers have a strong resistance to voice AI — which may stem from the welcome message (too robotic, not transparent), poor past experiences, or a type of clientele that structurally prefers human contact.
To improve the TOH: work on the welcome message (the first 15 seconds are critical), ensure that the agent introduces themselves naturally without being intrusive, and facilitate the transfer to a human to avoid trapping resistant customers (which would worsen their aversion).
KPI 5 — The post-interaction re-engagement rate
Post-AI Re-engagement
% of customers who call back within 24 hours after an AI interaction
This KPI detects "ghost" resolutions: conversations that the agent considered resolved but were not actually resolved, forcing the customer to call back the next day. A high re-engagement rate (> 20 %) indicates that your agent is closing conversations without ensuring that the problem is truly resolved — a form of artificial resolution that inflates your TRA without creating real value.
The recommended weekly dashboard
Create a weekly performance review routine: TRA of the week vs previous week, top 5 most frequent intents, top 5 intents with the most transfers (potential for improvement), average CSAT of the week, calculated CPR. This 30-minute weekly review identifies improvement projects and drives your configuration roadmap.