CSAT is typically collected as a single question right after an interaction: "How satisfied were you with this support experience?", answered on a scale (often 1-5 or a thumbs up/down). The score is calculated as the percentage of respondents who gave a positive rating.

For AI agents specifically, CSAT matters more than resolution rate alone: an agent can technically close a ticket while leaving the customer frustrated, so tracking satisfaction separately from completion catches quality problems that a pure throughput metric would miss.

Frequently Asked Questions

How is CSAT different from NPS?

CSAT measures satisfaction with one specific interaction. NPS measures overall likelihood to recommend the company as a whole. A customer can be satisfied with a single support chat (high CSAT) while still being a detractor of the company overall (low NPS), or vice versa.

What's a good CSAT score for an AI agent?

It varies by industry and ticket complexity, so compare an agent's CSAT against your own historical human-agent baseline rather than an external benchmark, since "good" for a simple order-status question looks different than "good" for a complex billing dispute.

Can a low response rate skew CSAT results?

Yes, customers with strong feelings (very happy or very frustrated) are more likely to respond than neutral ones, so a low response rate can make the score look more extreme in either direction than the true average experience.

Why track CSAT separately from whether the AI agent resolved the ticket?

Resolution rate measures whether the agent closed the case; CSAT measures how the customer felt about how it got closed. An agent that resolves everything but frustrates customers along the way needs a different fix than one that simply can't finish the job.