CSAT (customer satisfaction score) and what’s good

To work it out, divide the satisfied responses by all the responses to your survey, then multiply by 100. If 40 customers answered last month and 34 of them chose a 4 or a 5, your CSAT was 85%.

If you answer support yourself, you already know how most conversations went, because you wrote the replies. A score built from the few customers who answer a survey adds little to that. The questions they sent in the first place tell you far more.

Deacon is an AI customer support agent that answers your customers from your own docs in seconds. It shows you what they keep asking and where they get stuck, and it has a free plan.

How to calculate CSAT

On a five-point scale, satisfied usually means a 4 or a 5. Here’s a month at a small software company that answered 140 support conversations and got 35 ratings back.

The 4s and 5s come to 27, so CSAT is 27 ÷ 35 × 100, or 77%. A tool that shows the average rating instead would put the same month at 4.1 out of 5, and counting only the 5s gives 51%. All three describe the same 35 customers, and none of them says anything about the 105 who didn’t answer.

When the survey is sent and who answers it

A CSAT survey asks about one conversation, soon after it ends. Intercom asks for a rating with five emoji when a teammate closes a conversation, as long as a teammate replied and the conversation ran past 250 characters. Zendesk emails its survey 24 hours after a ticket is solved unless you change it. Help Scout puts rating links under every email reply once you switch ratings on, and customers have 10 days to click one.

Most customers don’t answer. Customer feedback surveys sent through SurveyMonkey in 2025 had an average response rate of 31.81%, counting everyone who started one. SurveyMonkey doesn’t say how many surveys that covers. So around two in three customers say nothing, and the score can’t tell you whether they were happier or angrier than the ones who did.

Some customers are never asked. Zendesk surveys solved tickets, Intercom skips short conversations, and a customer who gave up before writing in never gets a survey at all.

What is a good CSAT score?

Around 88% for tickets such as email, and around 96% for chat and messaging. Those are the medians across more than 32,000 companies on Freshworks’ products in 2024, and companies of up to 250 employees scored about the same. These are the published figures we found with a named sample, read on 27 September 2026.

The two sources count differently. LiveChat asks for a good or bad rating and averages every rated chat. Freshworks counts the customers who give a positive score after their problem is solved, then takes the middle company, with outliers left out. So compare your CSAT with your own last quarter, counted the same way, rather than with anyone else’s.

Four ways CSAT misleads

A handful of responses swing it. With 20 responses, each one is worth 5 points, so one customer changing their mind moves you from 85% to 80%.

Customers rate the product as well as the reply. A correct, friendly reply saying the feature they need doesn’t exist can still get a 1. Intercom’s CSAT report even has a chart of the topics driving dissatisfaction, and the example it gives is product bugs.

The scales don’t line up. Each tool picks its own scale and its own idea of satisfied, so the same customers can produce different scores.

Satisfied doesn’t mean staying. In a study of more than 75,000 customers that CEB published in Harvard Business Review in 2010, 20% of the customers who said they were satisfied meant to leave anyway, and 28% of the dissatisfied meant to stay. The same study found that how much effort customers had to put in predicted loyalty better, which is what customer effort score measures.

CSAT vs NPS vs customer effort score

All three are one-question surveys, and each asks about something different.

CSAT and customer effort score ask about one conversation, while NPS asks about the company as a whole. In CEB’s study, CSAT was a poor predictor of whether customers would stay, spend more and speak well of the company. NPS did better, and customer effort score did best after a service conversation.

What CSAT tells a founder who answers support alone

Not much you don’t already know. You wrote every reply, so you know which conversations went badly, and at your volume a single response moves the score by several points.

The numbers you can count without asking anyone say more. First response time is how long customers waited for you. First contact resolution is the share who got their answer without writing back, and the resolution rate is the share whose problem ended solved.

The questions themselves say the most. A dozen people asking how to cancel in one month means the setting is hard to find, and a dozen asking for the same integration tells you what to build next. Customer insights goes through what each kind of question points to, and voice of the customer covers the other ways to hear what customers think.

If you want a score anyway, ask one question at the end of each conversation, keep the scale the same from month to month and read every comment. With a few dozen responses, the comments are the part worth having.

How Deacon answers customers and shows what they need

Deacon answers your customers from your own docs in seconds, at any hour and in the language they write in. When your docs don’t cover a question, it says it doesn’t know instead of guessing, and asks for the visitor’s email so you can reply.

Deacon doesn’t run surveys. It shows you what customers asked and whether they got an answer, from the first conversation. On Analytics › Widget, Answered is the share of messages Deacon attempted that got an answer, beside counts of messages and people. Waiting for a reply counts everyone who left their email and hasn’t heard back from you yet. Once one page’s questions go unanswered more often than anywhere else, Analytics › Widget names that page, so you can see where people get stuck.

Topics group the questions one written answer would cover, count the people who asked and label most groups as a bug report, a feature ask, a how-to or another type. That’s what customers are stuck on and what they want next, in their own words, from everyone who asked rather than the few who fill in a survey.

Every question your docs didn’t cover has an Answer this button. Write the answer once and press Save & check, and Deacon shows you the reply it now gives, so the next person who asks can get it in seconds.

Chatbot analytics goes through everything the dashboard counts, and all of it is on the free plan, which needs no card.