Customer satisfaction metrics help you understand how customers judge an experience, where they encounter friction, and which problems deserve attention. A useful report combines direct feedback with service and business outcomes, while keeping those different kinds of evidence separate.
This guide covers ten measures, with formulas, illustrative calculations, limitations, and a practical survey plan. CSAT measures reported satisfaction, CES measures perceived effort, and NPS measures willingness to recommend. Churn, lifetime value, and resolution metrics add context; they do not directly tell you whether a customer is happy.
Choose a metric around the question you need to answer
| Measure | Question it helps answer | Main limitation |
|---|---|---|
| CSAT | How satisfied were respondents with this experience? | Respondents may differ from everyone invited. |
| CES | How easy or difficult was a particular task? | Question wording changes the score’s direction. |
| NPS | How likely are respondents to recommend us? | Intention is different from an actual referral. |
| Customer churn | How many customers in a defined cohort left? | The rate does not establish why they left. |
| Customer lifetime value | What future value does a model estimate? | Results depend on assumptions and the value definition. |
| First contact resolution | Was the issue resolved at its first contact? | A ticket status alone may not prove resolution. |
| Customer health score | Which relationships need a closer look? | A custom index is not automatically a validated prediction. |
| Time to resolution | How long did the issue take to resolve? | A completed-ticket view omits unresolved work. |
| Customer satisfaction index | How do defined satisfaction dimensions combine? | Different indexes are not interchangeable. |
| Social sentiment | What themes and opinions appear in collected mentions? | Public mentions do not represent the whole customer base. |
You do not need all ten on a daily dashboard. Start with a specific experience, one direct feedback measure, and an operational measure that can help explain the result. Keep broader relationship and financial indicators on a separate reporting cadence.
1. Customer Satisfaction Score: measure a specific experience
Ask customers to rate their satisfaction with a named experience, such as a support interaction or delivery. Avoid combining unrelated experiences in one question. On a five-point scale from very dissatisfied to very satisfied, a common percentage method counts ratings of four and five as satisfied.
CSAT = satisfied valid responses ÷ all valid responses to that question × 100. Qualtrics documents this top-two-response calculation. A different scale or threshold needs its own documented definition.
Illustrative example: 84 of 100 valid responses rate the interaction four or five. CSAT is 84%. If 400 eligible interactions received a survey invitation, the response rate is 100 ÷ 400 × 100 = 25%. The 84% describes respondents, not all 400 interactions.
Report the invitation count, valid response count, rating distribution, and comments alongside CSAT. Decide how repeated submissions are deduplicated. Separate product dissatisfaction from the quality of the support experience before using a result for agent coaching.
2. Customer Effort Score: identify friction in a task
CES concerns the effort involved in completing a task. Qualtrics describes CES as a single-item measure that can apply to resolving an issue, fulfilling a request, or making a purchase. Use a question that names the task and an optional follow-up about the obstacle.
Define the scale before calculating anything. For an agreement statement about the company making a task easy, a scale from one, strongly disagree, to seven, strongly agree, makes a higher score better. For an effort question with one meaning very little effort and seven meaning very high effort, a lower score is better. Never mix these directions in one trend.
An example reporting method: mean ease score = sum of valid ease ratings ÷ valid response count. Ratings of 6, 5, 7, 4, and 3 give 25 ÷ 5 = 5.0 out of seven. This is an illustrative averaging method, not a universal CES formula. Publish the anchors, distribution, and sample size rather than labeling the number simply “CES 5.”
Inspect whether customers repeated information, changed channels, or struggled with permissions. The score helps locate friction; it does not prove a particular customer will renew.
3. Net Promoter Score: track recommendation intent
NPS asks about likelihood to recommend on a zero-to-ten scale. Bain’s scoring definition groups nine and ten as promoters, seven and eight as passives, and zero through six as detractors.
NPS = percentage of promoters − percentage of detractors. Both percentages use all valid responses, including passives, as their denominator. The resulting score ranges from −100 to +100.
Illustrative example: among 100 responses, 55 are promoters, 25 passives, and 20 detractors. NPS is 55 − 20 = 35. It is a score of 35, not 35% customer satisfaction.
Add an open question about the reason for the rating. Keep a relationship survey separate from one triggered by a particular support interaction. An overall recommendation score can reflect price, product reliability, and brand expectations as well as service. For a broader behavioral view, see how to measure customer loyalty.
4. Customer churn: observe departures without assuming the cause
For a starting-cohort report, define a set of active customers at the start of the month and the event that counts as leaving.
Starting-cohort churn = customers from that starting cohort who churn during the month ÷ customers in the starting cohort × 100. Six departures from a starting cohort of 200 give 3%. New customers acquired during that month do not enter this particular denominator or numerator.
Specify whether you count an account, user, or subscription, and whether churn happens at a cancellation request or the end of paid service. Report reactivation separately. Billing problems, organizational changes, and product fit can all matter, so connect the number with cancellation evidence rather than labeling every departure dissatisfaction. See how to reduce customer churn for the diagnostic workflow.
Provider definitions can differ: Stripe’s analytics documentation includes new subscribers in its rolling 30-day churn denominator. Reconcile definitions before comparing reports.
5. Customer lifetime value: keep the estimate separate from satisfaction
CLV is a model of customer value across a relationship. State whether your model estimates revenue, gross profit, or discounted contribution. Historical revenue already received is different from a forecast of future revenue.
A simplified subscription revenue estimate is monthly average revenue per subscriber ÷ monthly subscriber churn rate, using the rate as a decimal. Stripe documents this ARPU/churn approach as an estimate. With $100 monthly ARPU and 5% monthly churn, the illustrative estimate is $100 ÷ 0.05 = $2,000.
This shortcut assumes sufficiently stable revenue and churn. It excludes acquisition costs, service costs, margin, and discounting, so $2,000 is not estimated profit. Zero observed churn over a short period does not establish infinite lifetime value. Use the model to frame investment questions; it cannot show that a service change caused a financial improvement.
6. First contact resolution: check the customer outcome
FCR = eligible issues resolved during their first contact, with the defined validation window complete, ÷ all eligible issues in that mature first-contact cohort × 100. Keep unresolved issues in the denominator. Publish exclusions and cases that lack enough evidence to classify.
Define the first contact for each channel, the same-issue matching rule, and how long customers have to test the solution. A later contact about the same unresolved issue should not become a fresh success because it has a new ticket ID. An unrelated question or courtesy reply needs a different classification.
Zendesk’s one-touch metric has reply-count rules that can include solved tickets with no replies. It therefore needs reconciliation with your outcome-based FCR definition. The separate first contact resolution guide covers measurement and improvement in more detail.
7. Customer health score: document what a composite means
A health score combines selected indicators, such as achievement of a customer goal, stakeholder feedback, product use, and unresolved blockers. Normalize components to the same scale, disclose weights, and define missing-data handling.
Illustrative custom index: with each component on a zero-to-100 scale, 50% goal progress, 30% stakeholder feedback, and 20% blocker status produce 0.5 × 80 + 0.3 × 60 + 0.2 × 100 = 78. The weights are invented for this example and need validation for actual use.
A score of 78 is an index, not a 78% renewal probability. Missing feedback should remain visible rather than silently becoming a neutral rating. Low login frequency may be normal for an automated workflow; frequent visits may reflect repeated errors. Use the score to prioritize investigation.
8. Time to resolution: show both completed and unfinished work
Define the starting event and final resolution event, then calculate each issue’s duration. Mean resolution time = sum of included issue durations ÷ included resolved-issue count. A median and a clearly specified upper percentile can reveal parts of the distribution that the mean hides.
Label elapsed time versus business time, time zone, holidays, pause rules, and reopening behavior. SLA clock time may exclude periods that the customer still experiences as waiting. Keep those clocks separate.
A resolution report for issues completed this month includes some older issues and omits those still open. Pair it with the number and age of unresolved issues. Break out incident severity or issue type before investigating whether a slower result corresponds with lower satisfaction. The response-time guide covers the earlier wait for a useful first reply.
9. Customer Satisfaction Index: distinguish a local score from ACSI
A business can build a custom satisfaction index around defined dimensions, such as reliability, support, and value. Publish its questions, scales, normalization, weights, and inclusion rules. Display component scores alongside the total so one improving dimension cannot hide another deteriorating one.
The American Customer Satisfaction Index methodology uses three satisfaction questions with weights estimated through proprietary technology. A homemade average of CSAT, NPS, and CES is not an ACSI score and should not be compared with its published scores as though the methods match.
Changing questions or weights changes the measurement. Version the index and explain any break in the trend.
10. Social media sentiment: monitor themes with coverage limits
Collect mentions using documented channels, search terms, language coverage, and dates. Deduplicate repeated posts, distinguish customers from other commenters where evidence permits, and preserve context for review.
An example local measure is negative classified mentions ÷ all included classified mentions × 100. Disclose whether neutral and mixed mentions enter the denominator. Amazon Comprehend, for example, distinguishes positive, negative, neutral, and mixed sentiment; a three-label system will produce a different report.
Review a sample of automated labels, including sarcasm and posts about multiple topics. A spike around one incident can identify a problem worth investigating, but public sentiment is not the percentage of your customers who are satisfied. An absence of complaints is not proof of satisfaction.
Build a survey and reporting plan that supports action
- Choose the experience and population. Name the journey, eligible respondents, invitation trigger, exclusions, and contact frequency.
- Keep the instrument consistent. Use clear wording and scale anchors. Offer an optional reason question, and avoid leading customers toward a positive answer.
- Allow time to judge the outcome. A conversation rating can follow an interaction; feedback about whether a fix worked may need time for the customer to try it.
- Publish coverage. Show delivered invitations, valid responses, missing answers, and counts by channel or customer group. Survey response rate and CSAT use different denominators.
- Investigate before assigning cause. Read comments and inspect relevant cases. Account for changing issue mix, survey timing, and respondent composition.
- Assign an improvement and follow-up. Record the problem, owner, proposed change, review date, and quality safeguard. Contact customers who request follow-up and report what was learned.
Copyable reporting record: experience; metric and formula version; population and dates; scale and direction; invitation and response counts; result and distribution; operational context; recurring feedback theme; owner; next action; follow-up evidence.
For external comparisons, check that question wording, sampling, scales, and reporting windows align. The customer service benchmarking guide explains how to select a relevant reference. A credible satisfaction report gives the team a problem to investigate and an action to evaluate, alongside the limits of the evidence.
