CSAT Score: How to Measure Customer Satisfaction After a Chat

A CSAT score (customer satisfaction score) tells you how customers felt about a specific interaction — for live chat, usually the conversation they just finished. It's the simplest quality metric a support team can track, and one of the easiest to misread. This guide explains when to ask for a rating, how to calculate CSAT, what counts as a good score, and how to turn low ratings into actual improvements.

What CSAT measures (and what it doesn't)

CSAT answers one question: "How satisfied were you with this conversation?" It's transactional — it measures a single interaction, right after it happened.

What it captures well:

• whether the customer felt helped;
• the tone and speed of the conversation;
• problems with a specific agent, bot answer or process.

What it doesn't capture:

• long-term loyalty or likelihood to recommend (that's NPS territory);
• the experience of customers who never contacted you;
• the reasons behind the score, unless you read the conversations.

A customer may rate a chat low because the policy was bad, not the agent. That's useful to know, but it means CSAT should be read alongside the transcripts, not instead of them.

When to ask for a rating after a chat

Timing shapes the result more than the wording.

• Ask when the conversation is resolved, not in the middle. Ratings mid-conversation measure impatience.
• Ask once per resolution. If a conversation is reopened and resolved again, that's a new cycle and can get a new rating.
• Keep it inside the chat. A rating prompt in the widget gets far more responses than a survey email sent hours later.

In ChatDeck, when a conversation is resolved — by an operator or by the AI assistant — the visitor is asked for a 1–5 rating right in the widget. Each rating belongs to that resolution cycle and is attributed to whoever closed it, the operator or the bot.

How to calculate CSAT: formula and examples

The standard formula counts the share of satisfied responses. On a 1–5 scale, "satisfied" usually means 4 or 5.

CSAT (%) = (number of 4 and 5 ratings / total ratings) × 100

Example. In a month you received 80 ratings: 45 fives, 20 fours, 7 threes, 5 twos and 3 ones.

1. Satisfied responses: 45 + 20 = 65.
2. Total responses: 80.
3. CSAT = 65 / 80 × 100 = 81%.

Note that CSAT is calculated from responses, not from all conversations. If you had 400 conversations and 80 ratings, your response rate is 20%, and that number belongs next to your CSAT in every report.

Average rating vs percent satisfied

There are two common ways to express the same data:

• Percent satisfied (the formula above) — easy to communicate, insensitive to the difference between a 1 and a 3.
• Average rating — in the example, (45×5 + 20×4 + 7×3 + 5×2 + 3×1) / 80 = 4.24. More sensitive to very low scores.

Both are valid. The important thing is to pick one and use it consistently over time. ChatDeck's analytics show the average rating (1–5) for the period, a separate average for conversations the bot closed on its own, a comparison with the previous period, and a rating trend chart with a line per operator and one for the bot.

What is a good CSAT score

There is no single benchmark that applies to every business, and published figures vary by industry and by how the survey is run. As a rough rule of thumb:

• many chat teams consider roughly 80% satisfied (or an average around 4.2–4.5) a healthy level;
• a sudden drop matters more than the absolute number;
• your own trend over time is the most reliable benchmark.

A low response count makes any score noisy. Ten ratings can swing ten points from one week to the next. Look at monthly or rolling numbers until you have enough volume.

Rating bots and humans separately

If an AI assistant handles part of your chats, mixing its ratings with your team's hides what's really happening. A bot that resolves simple questions well can inflate the overall number while human-handled complex cases get worse — or the reverse.

Track at least:

1. Overall CSAT — the customer's experience of your support as a whole.
2. Bot CSAT — conversations the assistant resolved without a person.
3. Per-operator CSAT — for coaching, not ranking.

Low bot ratings usually point to gaps in its instruction or knowledge, or to a handoff that comes too late. Our article on live chat metrics shows how CSAT fits with response time and handoff wait time.

Turning low ratings into improvements

A rating is a pointer, not a diagnosis. Build a simple weekly routine:

1. Pull every 1 and 2 rating from the past week.
2. Read the full conversation for each one.
3. Tag the cause: slow reply, wrong answer, policy, product bug, tone, bot error, unresolved.
4. Count the tags. The most frequent cause is your next fix.
5. Fix the system, not just the conversation: update the saved reply, the bot instruction, the help article or the process.

Common patterns and fixes:

• Slow first reply → adjust schedules and notifications; see first response time.
• Customer had to repeat themselves → improve transfers and context.
• Bot couldn't answer → refine the assistant's instruction or hand off sooner.
• Policy complaints → pass them to whoever owns the policy, with examples.

Our live chat best practices cover the habits that prevent the most common low scores.

Raising response rates without biasing scores

More ratings make CSAT more reliable, but some tactics distort it.

Good ways to get more responses:

• ask right after resolution, inside the chat;
• keep it to one tap;
• close conversations properly instead of letting them drift.

Tactics that bias the score:

• asking only customers who seemed happy;
• agents asking "please give me a 5";
• offering rewards for ratings;
• hiding the rating prompt after difficult conversations.

If agents are measured on CSAT, make sure they can't choose who gets asked.

Frequently asked questions

How do you calculate CSAT score?

Divide the number of satisfied responses (4 and 5 on a 1–5 scale) by the total number of responses and multiply by 100. For example, 65 satisfied out of 80 responses is 81%.

What is a good CSAT score for live chat?

As a rule of thumb, many teams aim for around 80% satisfied or an average rating above about 4.2, but benchmarks vary. Your own trend over time is more meaningful than any industry figure.

What is the difference between CSAT and NPS?

CSAT measures satisfaction with a specific interaction, right after it happens. NPS asks how likely someone is to recommend your company and measures overall loyalty. They answer different questions and work well together.

How many responses do I need for CSAT to be meaningful?

There's no strict cutoff, but with only a handful of ratings the score swings widely. Aim for dozens of responses per period before drawing conclusions, and use monthly or rolling numbers when volume is low.

Get started

ChatDeck asks for a 1–5 rating after every resolved conversation and shows average ratings for operators and the bot, trends and comparison with the previous period, with no per-operator fees. Create a free workspace and see which other numbers to watch in live chat metrics. The Read your analytics tutorial shows where each rating chart lives.