What is CSAT? Definition, formula, and score guide

CSAT (customer satisfaction score, pronounced "see-sat") is a metric that measures how satisfied customers are with a specific interaction, product, or experience. It is usually collected with a one-question survey, such as "How satisfied were you with your experience today?", and reported as the percentage of customers who answered 4 or 5 on a 5-point scale.

The basic calculation is simple. Interpreting the result is not. Two teams can survey the same customers and report different CSAT scores because they use different scales, cutoffs, eligibility rules, or reporting methods.

So the first thing to ask about any CSAT number is how it was calculated.

What is CSAT?

CSAT measures one experience at a time, such as a support conversation, a delivery, or a purchase. The survey usually asks a single question right after it, such as:

How satisfied were you with the support you received today?

Customers answer on a scale, commonly one to five. The team then reports either the percentage of responses considered satisfied or the average rating.

CSAT is useful for day-to-day decisions because it can be tied to a specific journey, channel, queue, policy, or change. For example, it can show that billing conversations are getting worse while order-status conversations stay steady.

It is not a complete measure of customer loyalty or the overall relationship. A customer can be satisfied with one support interaction and still prefer a competitor. CSAT is best read as one part of a wider measurement system.

How to calculate CSAT

The standard percentage formula is:

CSAT = satisfied responses / all valid completed survey responses x 100

On a five-point scale, teams commonly count ratings of four and five as satisfied. This is called the top-two-box method.

If 150 of 200 respondents select four or five:

CSAT = 150 / 200 x 100 = 75%

Some organizations count only the highest rating. Others report the average of all ratings, such as 4.6 out of 5. Any method can support an internal trend if it is appropriate and applied consistently. The results cannot be compared as if they were the same measure.

CSAT calculation methods compared

Consider 200 responses to a 5-point CSAT survey:

The same 200 responses produce three different CSAT scores:

This is why a CSAT number should always include its method and scale. “CSAT is 80” is not enough information to interpret or compare.

What belongs in the denominator?

The denominator is where many CSAT reports go wrong.

In the standard survey result, the denominator is the number of valid completed survey responses, not the number of surveys sent or total customer interactions. Response rate is a separate metric:

Response rate = completed survey responses / eligible survey invitations x 100

Track both. A high CSAT based on a small or changing slice of customers may not represent the whole experience.

Define survey eligibility before collecting data. Common decisions include:

• Which journeys and channels receive a survey?
• Is the survey sent after every interaction or only after a case closes?
• Are transfers, reopened cases, spam, tests, and duplicate contacts included?
• How often can the same customer be surveyed?
• What happens when one issue spans several contacts?
• Are human- and AI-handled interactions sampled in the same way?

There's no single right answer to these questions, but your team should agree on one answer for each and write it down.

How to design a useful CSAT survey

Ask about a defined experience

“How satisfied are you?” is broad. “How satisfied were you with the support you received today?” ties the answer to an event the team can investigate.

If you need feedback about a product, delivery, or onboarding journey, name that experience instead. Do not blend unlike experiences into one question and expect the score to identify what changed.

Keep the scale stable

Changing from stars to numbers, from five points to ten, or from top-two-box to top-box breaks the trend. If a change is necessary, run the old and new methods in parallel long enough to understand the difference, then mark the break in reporting.

Use a consistent trigger

Survey timing should match the experience. Support interactions may be surveyed at a defined closure event. A product or delivery experience may require enough time for the customer to use or receive it.

Consistency matters more than an arbitrary rule such as “send immediately.” Document the trigger by journey and keep it stable.

Protect customers from survey fatigue

Frequent users can receive many invitations and become less likely to respond. Set contact limits, coordinate surveys across channels, and preserve a control group if the team is testing a new survey treatment.

Pair the score with a diagnostic question

An optional open-text prompt such as “What is the main reason for your rating?” helps explain the number. Categorize the feedback consistently, then connect themes to the underlying conversations and operational data.

CSAT survey question examples

Common CSAT questions include:

• How satisfied were you with the support you received today?
• How satisfied are you with your recent purchase?
• How satisfied were you with your onboarding experience?

The most common scale runs from 1 (very dissatisfied) to 5 (very satisfied). Teams also use 1-to-10 scales, star ratings, emoji faces, and thumbs up or down. Pick one scale per journey and keep it.

What is a good CSAT score?

For outside context, the American Customer Satisfaction Index (ACSI), an independent benchmark of US consumer satisfaction, reports industry scores on a 0-to-100 scale. In its 2026 telecommunications study, internet service providers scored 73 and wireless providers reached a record high of 77. ACSI builds its scores from several survey questions, so they are not the same as a top-two-box CSAT percentage. Use them to see how satisfaction varies by industry, not as a target for your own score.

For your own program, a good CSAT score is one that is measured consistently, represents the intended customer population, and improves without harming another important outcome.

External benchmarks can give context, but they're easy to misuse. A benchmark built from a different question, scale, sampling method, or customer group can't be compared directly with your post-interaction top-two-box CSAT, and comparing the headline numbers can be misleading.

Use this order of comparison:

1. Your own baseline: Is the same measure improving for the same journey and customer population?
2. Internal peers: How do comparable queues, channels, regions, or products perform under the same method?
3. External references: Are the instrument, scale, sample, period, and case mix similar enough to make the comparison useful?

Set targets by journey rather than forcing one goal across the company. An account-access question and a complex claims process have different starting conditions. A lower score can reflect a harder case mix, not worse execution.

How much data do you need?

There is no universal minimum number of CSAT responses. Reliability depends on response volume, variability, the size of the difference you want to detect, and how precisely you need to estimate the result.

Practical safeguards include:

• Show the response count and response rate next to the score.
• Use a rolling period for low-volume journeys instead of reacting to weekly swings.
• Do not rank small teams or queues on differences that could come from a few responses.
• Add uncertainty intervals when the score supports material decisions.
• Check whether the respondent mix changed before attributing a movement to service performance.

The more you segment, the smaller each sample becomes. Segment only to answer a decision-relevant question.

CSAT vs. NPS vs. CES

These metrics answer different questions.

Net Promoter Score (NPS) uses a separate survey and should not be converted into CSAT. Customer effort score (CES) can help explain why satisfaction changes when process friction is the likely driver.

Choose the measure based on the decision. Do not add all three to every interaction simply because they are available.

Measuring CSAT across human and AI service

Use the same satisfaction standard for comparable human- and AI-handled journeys, but do not assume the underlying case mix is comparable.

If an AI agent handles routine order questions while people handle escalated disputes, a raw CSAT comparison will favor the easier population. Compare matched journeys, disclose eligibility and transfer rules, and show the share of demand represented by each group.

Also separate satisfaction from resolution. A customer may appreciate a courteous response even if the task remains incomplete. Track CSAT beside verified resolution, repeat contact, escalation quality, and time to outcome.

How Sierra customers use CSAT

CSAT is one of the main ways Sierra customers judge whether an AI agent is doing its job. A common requirement is that the agent meets the same bar as the human team.

• Guild: Guild's member support specialists averaged 4.8 out of 5 CSAT, and its COO required any AI agent to match that. Sierra's agent matched the 4.8 score on member chat interactions.
• Tubi: Three months after launch, CSAT was up seven percentage points and containment reached 80%.
• Capital on Tap: The AI agent consistently outperforms both human-only and human-escalated CSAT scores.

These three results use different methods. Guild reports an average rating, Tubi reports a change in percentage points, and Capital on Tap compares CSAT across groups of conversations. Check how a CSAT number was calculated before you compare it to another one.

Ghostwriter, Sierra's agent for building agents, can report how CSAT is trending and suggest changes to the agent to improve it.

How to improve CSAT

Start with the reasons customers give for low ratings, then work on the causes that affect the most customers.

• Resolve the issue on first contact. Unresolved issues, repeat contacts, and transfers commonly pull ratings down.
• Cut wait time. Long holds and slow replies lower satisfaction before the conversation starts.
• Read the comments by theme. Group open-text reasons into themes and fix the most frequent ones first.
• Prioritize the weakest journeys. Focus on journeys with low scores and high volume.
• Follow up with dissatisfied customers. A quick follow-up can recover the relationship and reveal the root cause.
• Let AI agents handle routine requests. An AI agent can resolve common requests such as order status or account updates at any hour, so customers don't wait and human agents have more time for complex cases. Hold the AI agent to the same CSAT standard as your human team.

Document how your CSAT is defined

Before publishing the metric, record these fields in a shared metric dictionary:

• Survey question and scale
• Definition of a satisfied response
• Numerator and denominator
• Eligibility and exclusion rules
• Survey trigger and contact limits
• Channels, journeys, and populations covered
• Response count and response rate
• Data source and metric owner
• Reporting period and minimum sample policy
• Date of any method change

Writing these down makes the score auditable. It prevents a survey redesign, channel shift, or denominator change from being misread as a customer-experience improvement.

CSAT is simple enough to calculate in a spreadsheet and important enough to shape executive decisions. Reporting the method with the number is what makes it trustworthy.