AI CSAT

AI CSAT is Lyro's own estimate of how satisfied the customer was with a conversation, scored from 1 to 5. After a conversation ends, Lyro reads the whole exchange and rates how well it went from the customer's point of view - even when the customer never filled in a rating.

It gives you a quality signal on every conversation, not just the small share where customers choose to leave a star rating.

AI CSAT vs Human CSAT

Lyro tracks two separate satisfaction signals. They answer different questions, so read them side by side rather than expecting them to match.

	Human CSAT	AI CSAT
Who scores it	The customer, by tapping a 1-to-5 rating in the chat	Lyro, by evaluating the conversation itself
Coverage	Only conversations where the customer answered the prompt	Every conversation that had at least one AI reply
What it measures	What the customer said they felt	How the conversation went, judged from the transcript
Good for	The ground-truth voice of the customer	A complete, unbiased-by-response-rate quality trend

Because customers who answer a rating prompt are a self-selected group (often the very happy or the very unhappy), Human CSAT can swing with a handful of responses. AI CSAT covers everything, so it is steadier and better for spotting trends. Use Human CSAT as the source of truth for how customers feel, and AI CSAT for a full-coverage read on quality.

How Lyro scores it

At the end of the conversation. Lyro waits until a conversation is closed, then scores it once over the full transcript - including any messages from a human teammate after a handoff. Scoring the finished conversation, rather than a half-finished one, keeps the number fair.
On a 1-to-5 scale, generous by default. If the customer got what they needed, the conversation scores a 5. Lyro only scores lower when something concrete went wrong.
Once per conversation. A conversation gets a single AI CSAT score. If it is reopened and continues, it is re-scored when it closes again.
With a short reason. Alongside the score, Lyro records a one-line explanation of why it landed on that number, so a score is never a black box.

A conversation is closed either when it naturally wraps up or automatically after a period of inactivity, so conversations do not sit open forever waiting to be scored.

Which conversations get a score

Any conversation where Lyro (or a human agent) sent at least one reply is scored. A conversation with no agent reply at all - for example, a visitor who opened the chat and left without an answer - does not receive an AI CSAT score.

Splitting AI vs human

Not every conversation is handled by the AI alone. When a chat is escalated to your team, a human becomes part of the outcome. Lyro labels each conversation by who resolved it:

AI - handled start to finish by Lyro, with no human involved.
Human-involved - escalated to a teammate at some point.

In the Analytics chart builder you can break AI CSAT down by this split to compare the two. It answers the question customers often ask: how good is the AI on its own, versus conversations a human stepped into? See Escalation for how conversations reach your team.

Note: AI CSAT scores the whole conversation, so a "human-involved" score reflects the combined AI-plus-human experience, not the teammate's solo performance.

Customizing the rubric

You control what each score means. In Settings, the AI CSAT section lets you edit the description for each tier from 1 to 5, so "what a 3 looks like" matches your own standards. Save your changes and Lyro scores new conversations against your rubric; reset at any time to return to the default. See Settings for details.

Reading AI CSAT well

Pair it with resolution rate. Resolution rate tells you the AI finished the conversation on its own; AI CSAT tells you it went well. High resolution with lower CSAT means the AI is closing conversations customers were not thrilled with. Remember the two cover different sets: AI CSAT scores every conversation (including human-involved ones), while resolution rate counts only AI-resolved ones.
Expect it to differ from Human CSAT, and treat the gap as information rather than an error - the two measure different things on different populations.
Watch the trend, not a single day. A one-day dip on low volume is noise; a multi-day slide is a signal worth digging into.

Where to see it

Analytics dashboard - the AI CSAT chart is part of the starter set. Break it down (for example by AI vs human, channel, or intent) and click a point to drill into the exact conversations behind the number. See Analytics.
Conversation export - include the AI CSAT column in the CSV export to get a per-conversation score for your date range.

Analytics - chart, break down, and drill into AI CSAT.
Resolution rate - the autonomy signal to read alongside quality.
Settings - edit the AI CSAT rubric.
Escalation - how conversations become "human-involved".
Measure and improve - turn these numbers into action.