CSAT stands for Customer Satisfaction Score. It is a one-question survey sent after a support interaction asking the customer how satisfied they were. Most teams use a five-point scale, some a thumbs up/down, some a one-to-ten. The metric is the percentage of responses in the top buckets.
It is the fastest read on whether your support is actually working. Other metrics tell you how busy you are. CSAT tells you how the customer felt about the result.
How it differs from NPS
NPS asks about the brand. CSAT asks about the interaction. A customer can give a low NPS because the product is missing a feature, and still give a high CSAT because the support reply was clear and quick. The two numbers track different things and should not be averaged.
CSAT is also more granular. You can attribute it to a specific agent, a specific reply, a specific topic. NPS is a slower, brand-level signal. Use both, but use them for different questions.
What good looks like
A healthy consumer-facing CSAT lives in the eighty to ninety percent range. B2B SaaS tends to run higher because the customer is more invested in getting an answer. Anything below seventy percent is a warning sign. Anything above ninety-five percent should be checked for sample bias or unsatisfied customers who chose not to respond.
What matters more than the number is the trend and the comments. A flat CSAT with rising volume of "neutral" or "dissatisfied" comments tells you something is off even before the percentage moves.
CSAT for AI replies
When an AI agent handles a conversation end-to-end, you can still send CSAT. Most teams do. A high CSAT on AI-only conversations is a strong signal that the deflection is real and not just the customer giving up. A low CSAT on AI-only conversations is a stronger signal than any internal accuracy metric. Customers know when they were helped.
In Keloa
In Keloa, CSAT runs on both human and AI agent conversations and is broken out separately, so you can see how each is doing. We pair it with deflection rate to give the full picture: how much the AI took off your plate, and whether the customer was happy with the result. See the product page for what the dashboard looks like.