What is Customer Satisfaction Score (CSAT)?
Customer Satisfaction Score (CSAT) is the percentage of survey respondents who say they were satisfied with a specific interaction, product, or purchase. Teams collect it with a short survey fired at a defined trigger, count the responses at the positive end of the scale, and divide by all responses received.
The score is bounded by who answers. A survey that reaches every closed ticket still hears back from a minority of customers, and that minority skews toward the two extremes: people who were delighted, and people who were annoyed enough to type.
How Customer Satisfaction Score is calculated
CSAT runs as a four-step loop: trigger, scale, calculation, routing.
The trigger fires the survey at a defined event, usually ticket resolution, chat close, or call end. Which conversations are eligible is set upstream by contact rate, and questions closed by self-service, counted in deflection rate, never enter the denominator at all.
The scale decides what satisfied means. A 1-5 rating is the common form, with thumbs up/down and 1-7 variants in use. The seven-point agreement scale behind Customer Effort Score measures a different construct, so the two sets of responses cannot be pooled into one number.
The calculation is a share: satisfied responses divided by total responses, times 100. Most teams treat only the top two boxes as satisfied. If 240 people respond and 186 rate the interaction 4 or 5, CSAT is 186 divided by 240, times 100, or 77.5%.
Routing is the step teams skip. A score with no named owner and no review cadence records history without changing anything downstream of it.
What counts and what does not
Top-box ratings: Ratings of 4 and 5 on a five-point scale count as satisfied, and every team should publish which boxes it counts.
The neutral middle: A 3 pulls the score down under top-two-box scoring even though the customer voiced no complaint, which quietly flatters teams using shorter scales.
Non-responses: Customers who ignore the survey sit outside both the numerator and the denominator, so the score describes respondents only.
Question wording: One question about one interaction is the standard form, and customer satisfaction survey questions shift the response distribution more than most teams expect.
Trigger timing: A survey sent days after resolution measures a memory of the interaction that has already blurred into everything else that happened.
CSAT vs NPS vs CES vs customer health score
Support teams run several satisfaction numbers at once and then argue about which one is real, usually in the same meeting. Net Promoter Score asks about the relationship and predicts advocacy for the brand as a whole. Customer Effort Score asks how hard the customer had to work and predicts the loyalty damage friction causes. Customer health score combines product usage, support history, and engagement into a renewal signal for an account. Customer Satisfaction Score answers something narrower than all three: how did this one conversation land, minutes after it ended, for the person who sat through it.
What it counts | What it misses | Typical benchmark | |
|---|---|---|---|
CSAT | Satisfaction with one interaction, captured at or near the moment it closed | Everyone who ignored the survey, plus every contact resolved without one | Cross-industry satisfaction indexes are published annually |
Net Promoter Score | Stated likelihood that a customer would recommend the brand to someone else | Why the number moved, and the quality of any single conversation | Widely reported by companies, with no neutral publisher |
Customer Effort Score | Perceived effort the customer spent getting the issue resolved | Emotional response, and whether the outcome was actually correct | No public index; read against your own baseline |
Composite of usage, support, and engagement signals per account | How the last conversation felt to the person having it | Thresholds are defined internally per account model |
If you need to know whether a specific conversation worked, CSAT is the one to instrument first. If you need to know whether the account will renew, the health score carries more signal, and effort tends to expose churn risk earlier than either.
Why Customer Satisfaction Score matters for customer experience
A support organization without a post-interaction score learns about dissatisfaction through escalations, cancellations, and public reviews, all of which arrive weeks after the conversation that caused them. CSAT compresses that lag to minutes and attaches the signal to a specific ticket, agent, channel, and contact reason, which is what makes it usable at the case level.
The failure mode when it is missing is quieter than an angry review. Teams optimize the numbers they do have, usually handle time and backlog age, and a queue can look healthy while customers leave every conversation feeling processed and unhelped.
The tradeoff is real. Every survey spends a little of the customer's patience, and surveying more moments lowers response rates across all of them. Teams that survey every touchpoint often end up with more data points and less confidence in any of them.
What is a good Customer Satisfaction Score?
No single number qualifies as good across industries, and the two honest reference points are your own trend and published cross-industry data. The American Customer Satisfaction Index is the benchmark worth quoting: it has scored customer satisfaction on a 0-100 index since 1994, its national reading has sat in the mid-70s in recent years, and industry-level scores spread roughly from the mid-60s to the mid-80s around that centre.
Borrow it carefully. ACSI measures overall satisfaction with companies and their products, so it does not convert directly into a support team's top-box percentage, which is collected from a different population at a different moment.
Read your own number with three things attached: the response rate, the sample size, and the contact reason mix behind it. A high score from a handful of responses tells you less than a moderate score from a large sample, and a score that moves whenever the ticket mix moves has said nothing about service quality.
How AI agents change Customer Satisfaction Score
AI agents change the score before they change the experience, by changing the mix of conversations a human ever sees. Automation resolves the short, well-documented questions first, and those are the conversations that historically scored highest. The human queue then inherits complex, multi-touch, emotionally loaded cases, so human CSAT can fall while every individual agent performs exactly as before.
Two consequences follow. Measure automated resolutions and human resolutions separately before reading the blended number, because one average hides the composition shift completely. And tie escalation to a confidence score so the agent hands off before it guesses, since a confident wrong answer costs far more satisfaction than an early transfer.
Automation also makes surveying continuous and nearly free, which is exactly why blended dashboards mislead, a pattern covered in these trust metrics for AI support.
Implementing Customer Satisfaction Score measurement
Judge a CSAT program on the axes that decide whether the score ever changes a decision.
Coverage comes first: voice, email, chat, and non-English conversations all need a survey path, or the score describes your chat queue and gets read as the whole company. Integration surface is next: the survey should fire from the help desk's own close event and write the rating back onto the ticket, so score, transcript, agent, and contact reason live in one record.
Governance decides who owns verbatim comments, who may view agent-level scores, and whether those scores enter performance reviews, a choice that changes how agents ask for ratings.
Verbatims are free text, and customers paste order numbers and account details into them, so regulated buyers ask how deletion requests reach the survey store and whether the vendor's SOC 2 Type II report covers it. The constraint teams underestimate is suppression: without a per-customer survey window, your highest-contact customers answer most often and quietly set the score.
Customer Satisfaction Score and quality assurance
CSAT and customer service quality assurance score the same conversation from opposite ends: the customer rates how it felt, the scorecard rates what the agent did. Reading them together is where the value sits, because a conversation the customer rates 5 that fails policy review is a compliance risk, and one that passes QA while the customer rates it low usually points at a policy the customer found unreasonable.
Neither metric acts on its own. A customer feedback loop is the machinery that turns a low rating and a QA finding into a rewritten macro, a corrected help article, or a changed refund rule, and then tells the customer it changed.
What does CSAT mean in plain terms?
CSAT stands for Customer Satisfaction Score, and the full form describes the job precisely: satisfaction, scored. Think of it as a tip jar at the end of a conversation. Most people walk past it, some leave something because they were genuinely pleased, and some leave something because they want you to know they were not.
Without it you would still know how many tickets closed and how quickly, and you would know nothing about whether closing them actually helped anyone. A ticket can be resolved correctly by policy and still end with a customer who quietly stops buying.
The tradeoff is that the tip jar only hears from people willing to stop and use it. Strong satisfaction and strong irritation are both over-represented, so the weekly number moves for reasons that have little to do with the average customer's experience, and reading one week as truth is how teams end up chasing noise.
Common Customer Satisfaction Score mistakes
Comparing scores built on different scales is the first. A five-point top-two-box percentage, a thumbs up/down ratio, and a seven-point average are three different measurements, and averaging them across channels produces a number that describes no population at all.
Scoring individual agents on thin samples is the second. A handful of responses in a month has enormous variance, so ranking agents on it rewards luck, and attaching pay to it teaches agents to ask satisfied customers for ratings and stay quiet with everyone else.
Surveying only cleanly closed tickets is the third. Abandoned chats, calls that dropped, and customers who gave up before reaching anyone leave no rating behind, which removes the worst experiences from the sample and lifts the score.
Treating verbatim comments as a word cloud is the fourth. The rating tells you something moved; the sentence underneath tells you which policy, article, or handoff caused it, and that text needs an owner and a routing rule.
What is the CSAT formula?
The CSAT formula divides satisfied responses by total responses and multiplies by 100. Satisfied usually means the top two boxes on a five-point scale, so 186 top-box ratings out of 240 responses returns 77.5%. Publish which boxes you count, because a two-box definition and a three-box definition produce different scores from identical data.
What is the difference between CSAT and NPS?
CSAT and NPS measure different objects. CSAT scores one interaction shortly after it ends and tells you whether that specific conversation worked. NPS asks how likely someone is to recommend the brand and tracks the overall relationship. Support teams commonly run CSAT per resolved ticket and NPS quarterly at the account level.
Is CSAT the same as Customer Effort Score?
CSAT and Customer Effort Score are separate metrics collected in similar ways. CSAT captures how satisfied the customer felt; CES captures how much work the resolution demanded, usually on a seven-point agreement scale. A conversation can produce high satisfaction and high effort together, which is why teams tracking both catch friction that satisfaction alone hides.
What is a good CSAT response rate?
CSAT response rates vary widely by channel, by timing, and by how many surveys a customer has already received, and no neutral published norm exists to aim at. Track your own rate as a trend, compare it across channels, and treat a falling rate as a warning that your sample is narrowing toward the extremes.
Why does CSAT drop after deploying an AI agent?
CSAT often drops for the human queue after automation because the composition of that queue changed. Routine questions get resolved automatically, and what reaches people is harder, older, and more frustrated. Measure automated and human-handled conversations separately, then read the blended score against both components before concluding that service quality fell.
How often should CSAT surveys be sent?
CSAT surveys should fire on a resolution event, with a suppression window so one customer is never surveyed repeatedly in a short period. Many teams survey a continuous sample of closed conversations rather than every single one, which protects response quality. Timing matters more than frequency: a fresh survey earns more responses and more specific comments.