CSAT
In one sentence
CSAT, or customer satisfaction score, is the percentage of customers who report being satisfied with a specific interaction, measured by a single question asked immediately afterward and answered only by the self-selected minority who respond.
Not to be confused with Net promoter score, Customer effort score, or Service level.
Definition
CSAT, the customer satisfaction score, is the percentage of customers who say they were satisfied with a specific interaction.
It is the most direct measure of whether one experience worked. It is transactional and immediate, which is both its strength and its limitation.
How it is calculated
- Typically a single question asked immediately after an interaction, on a scale of three to five points.
- The score is usually the proportion selecting the top one or two options, expressed as a percentage.
- Which options count as satisfied is a decision that materially changes the number, and it is frequently undisclosed.
The response bias problem
- Response rates are typically low, and respondents skew toward the extremes.
- Very satisfied and very dissatisfied customers respond; indifferent ones do not.
- So CSAT measures the opinions of people motivated enough to answer, which is not the customer base.
What CSAT is good for
- Detecting sharp changes. A sudden drop is a reliable signal that something broke.
- Comparing interaction types within one organization.
- Identifying specific failure patterns when paired with transcript review.
What it is poor for
- Cross-organization comparison, since scale, wording and timing all differ.
- Predicting loyalty, which effort measures do better.
- Assessing an automated agent, because respondents frequently score the outcome of their problem rather than the interaction itself.
The automation-specific bias
- Some customers rate an automated interaction poorly on principle, regardless of whether it resolved their issue.
- Others rate it highly for speed even when it was left unresolved.
- Reading CSAT for automated channels requires pairing it with resolution data.
Common misconception
That CSAT measures overall customer sentiment. It measures the satisfaction of self-selected respondents about one interaction, immediately afterward. The quiet middle of the customer base is missing from it entirely.
Why it matters commercially
CSAT is the guardrail that stops efficiency metrics being optimized into damage. Containment and handle time both improve when contacts end sooner; CSAT is the number that notices when they ended badly. It is most valuable as a paired metric rather than as a standalone target.
In voice specifically
For an automated channel the bias sharpens. Some visitors mark a voice agent down on principle even when it solved the problem; others mark it up for speed even when it did not. So an automated-channel CSAT read without segmenting by whether the issue was actually resolved will mislead in both directions at once.
Where AsqVox fits
Sentiment analysis on transcripts is a proxy signal available for every conversation rather than only for the minority who answer a survey. That addresses the response bias problem directly, while introducing its own limitations, since inferred sentiment is not the same as a stated rating.
Visual
Who actually answers
CSAT hears the motivated ends of the base. The indifferent majority does not respond, so the score is not the customer base.
Statistics
Every figure carries its source and year. Vendor numbers are labelled as vendor numbers, and where no reliable figure exists this page says so rather than borrowing one.
There is no reliable current cross-industry CSAT benchmark that survives scrutiny, because scale design, question wording, timing and top-box definition all vary.
-no reliable figureInternal trend, measured the same way over time, is the only defensible use. A CSAT figure quoted for comparison against another company is comparing survey designs, not services.
Response rates for post-interaction surveys are generally low, which is well recognized in practice but not systematically quantified in current public research.
-no reliable figureThe direction is not in dispute and the mechanism, extreme responders answering while the indifferent middle does not, is well understood. A specific rate would not withstand scrutiny, so none is published here.
Automation containment benchmarks, for context: mature deployments run roughly 70 to 80 percent, average deployments 40 to 55 percent, and rule-based systems under 35 percent.
70 to 80% mature, 40 to 55% averageindustry rangeIndustry-reported ranges, 2025 - Industry-reported rather than audited. Quoted here because CSAT is the guardrail that should accompany any containment target: a high containment number with no CSAT beside it is an efficiency claim with the quality half missing.
Top-box and top-two-box conventions both exist and produce materially different percentages from identical responses.
-no reliable figureThere is no single correct CSAT percentage for a given set of answers, only a percentage under a stated convention. When the convention is undisclosed, the number cannot be compared with anything, including a prior period reported under a different rule.
Examples
In practice
An organization reports CSAT declining after deploying voice automation and considers rolling it back. Segmenting by resolution shows contained-and-resolved conversations scoring well, while contained-and-unresolved conversations score very poorly and are dragging the aggregate. The problem is the escalation threshold, not the automation itself, and only pairing CSAT with resolution data made that visible.
The everyday version
CSAT is asking customers whether they were happy with a specific interaction. The catch is that mostly the delighted and the furious bother to answer, so it tells you about the extremes rather than about everyone. It is excellent at catching a sudden drop and poor as a target you chase in isolation.
Usage
Who says it
- CX teams, universally, as the headline satisfaction number.
- It appears in operational dashboards and in service level agreements.
Where it turns up
- Next to quality guardrails, survey methodology, response rate and containment targets.
- The methodology detail, the scale, the top-box rule and the timing, matters more than the headline figure, because it is what makes the number comparable or not.
Common misuse
- Comparing CSAT across organizations, where scale, wording and timing all differ.
- Setting containment targets without a CSAT guardrail, which lets efficiency be optimized into damage.
- Reading automated-channel CSAT without segmenting by resolution, which mixes the rate-it-poorly-on-principle and rate-it-highly-for-speed biases into one misleading average.
Questions people ask
How is CSAT calculated?
A single question is asked immediately after an interaction, usually on a three-to-five-point scale, and the score is the proportion selecting the top one or two options, expressed as a percentage. Which options count as satisfied, the top-box or top-two-box choice, materially changes the number and is often undisclosed.
What is a good CSAT score?
There is no reliable cross-industry benchmark that survives scrutiny, because scale, wording, timing and the top-box definition all vary between organizations. Internal trend measured the same way over time is the only defensible use. A CSAT quoted for comparison against another company is comparing survey designs, not services.
What is the difference between CSAT and NPS?
CSAT rates satisfaction with one specific interaction, just after it happens. NPS asks about overall likelihood to recommend the brand. They are routinely swapped in dashboards, but they answer different questions, and effort measures predict loyalty better than either. Reading a transactional CSAT as if it were a loyalty signal is the usual error.
Why is CSAT unreliable for automated agents?
Because respondents frequently score the outcome of their problem rather than the interaction. Some rate an automated agent poorly on principle even when it resolved the issue; others rate it highly for speed even when it did not. Read without segmenting by whether the issue was actually resolved, automated-channel CSAT misleads in both directions at once.
Last reviewed 4 August 2026. Written and reviewed by Dhruv Dholakia, founder of AsqVox.