NPS vs CSAT vs CES: choose the question your CX team needs answered

A customer can like the person who helped them and still find the process exhausting. They can complete a task easily and still decide that the product is poor value. One customer-experience score cannot explain all three situations.

Choose the measure around the decision. Net Promoter Score (NPS) captures stated willingness to recommend. Customer Satisfaction (CSAT) asks how satisfied someone is with a specified experience. Customer Effort Score (CES) asks about the difficulty or ease of getting something done. Their value comes from the question, timing and customer group behind the number.

Before adding another dashboard tile, decide what your team would change if the result were poor. A relationship problem, an unsatisfactory resolution and a difficult task call for different evidence and different owners.

Three questions about the same business

Consider Northline, a fictional broadband provider trying to improve customers' first month. Its team wants to know whether customers would recommend the service, whether support interactions leave them satisfied and where router activation requires too much work. The counts below illustrate scoring rather than report Northline customer research.

Those questions belong at different moments. A recommendation question immediately after a friendly support call may capture gratitude for the agent. An activation-effort question months later relies on distant recall. A satisfaction question that says only “How was your experience?” leaves the customer to choose which experience to rate.

Northline needs to define the object of each measure before comparing its results.

NPS: would customers recommend the service?

The standard NPS question asks how likely someone is to recommend the company, product or service to a friend or colleague, using a 0–10 scale. Scores of 9–10 are promoters, 7–8 are passives and 0–6 are detractors. NPS is the percentage of promoters minus the percentage of detractors, as set out in Bain's measurement guidance.

Suppose Northline receives 200 valid ratings: 90 promoters, 70 passives and 40 detractors.

NPS = (90 ÷ 200 × 100) − (40 ÷ 200 × 100) = +25.

Passives remain in the denominator. Removing them would inflate the result to about +38.5. NPS is a score ranging from −100 to +100; report this example as +25 NPS, rather than “25% satisfaction.”

Use a defined relationship-survey population, with enough experience of the service to answer the question. Keep the timing and eligibility stable. For Northline, a sample of established customers answers a different question from people who have just activated a router or contacted support.

Ask what drove the rating and inspect the reasons. Price, connection reliability and complaint handling could all affect recommendation. The score can identify a relationship concern, but it does not identify the cause or establish that someone will actually renew, refer a friend or cancel. Compare it with subsequent behavior where appropriate.

CSAT: were customers satisfied with this experience?

For a support interaction, Northline could ask: “How satisfied were you with the help you received on this request?” Specify whether the intended object is the agent's help, the resolution or the whole contact. They can receive different answers.

A common CSAT format uses a 1–5 scale from very dissatisfied to very satisfied and reports the percentage selecting 4 or 5. Qualtrics' CSAT guidance describes this top-two-response calculation and distinguishes satisfaction with a specific experience from the wider relationship.

If 80 of 100 valid respondents select 4 or 5:

CSAT = 80 ÷ 100 × 100 = 80%.

The denominator is all valid answers to that satisfaction question, including neutral and dissatisfied ratings. It is not just customers who selected a positive or negative answer. A mean rating, such as 4.1 out of 5, is a different summary and should carry that label.

For a discrete contact, ask while the interaction is recent. Check separately whether the underlying issue was resolved. A customer might appreciate a helpful agent while waiting for an engineer to restore the connection. Northline would make the wrong operational decision if it treated satisfaction with the conversation as proof that the broadband problem had ended.

Also examine who receives the survey. A trigger that includes only neatly resolved cases can miss customers whose requests remain open or who abandoned the process. Report the eligible contacts and responses so that “80% satisfied” remains a statement about the respondents who answered.

CES: how easy was the task to complete?

CES focuses on a particular interaction or task. Qualtrics' effort-score guidance describes its use after requests, purchases, returns and issue resolution, while the experience is fresh.

Northline could ask after an activation attempt: “How easy or difficult was it to activate your router?” For this program, use 1 = very difficult and 5 = very easy, then report the mean of valid answers. If 80 customers provide ratings that sum to 312:

Mean activation ease = 312 ÷ 80 = 3.9 out of 5. Higher means easier.

Keep the full distribution alongside the mean. Two groups can have the same average even when one contains many very difficult experiences and the other is mostly neutral.

CES is used with different question formats, scale directions and summaries. A question about how much effort someone expended can run in the opposite direction from one asking whether the company made things easy. Some programs report a favorable-response percentage instead of a mean. Record the exact item, anchors and calculation; do not turn 3.9 out of 5 into “78% of customers found activation easy.”

Include failed attempts in the research design. If the survey appears only after successful activation, the customers with the most serious friction may never see it. Compare the survey with activation completion, repeat attempts, time spent and subsequent support contacts.

Make the denominators visible before interpreting the gap

Northline's three example results have three different bases: 200 recommendation ratings, 100 support-satisfaction ratings and 80 activation-ease ratings. They may concern different people, episodes and periods. There is no meaningful arithmetic in averaging +25, 80% and 3.9 into a single “CX score.”

Keep nonresponses, “not applicable” and incomplete answers separate. They are not zero ratings. A respondent who skips the NPS item but answers CSAT contributes to the CSAT base only. Report responses relative to eligible invitations, and check repeated contacts or multiple responses from the same customer before making customer-level claims.

Suppose 200 of 800 invited customers answer the relationship survey. The response rate is 25%, while NPS remains +25. The matching numbers mean different things. Neither tells you how the 600 nonrespondents would have rated the service.

A trend also needs a stable comparison. If the next support sample contains more simple password questions and fewer outages, higher CSAT may reflect the case mix. Compare relevant issue types, channels and customer stages before attributing the change to a new support policy. Small score movements also need appropriate uncertainty analysis before a team treats them as a meaningful improvement.

Use the score to choose an investigation

For Northline, difficult activation with otherwise satisfactory support points toward setup instructions, equipment, the activation flow or handoffs. The team should examine failed attempts and customer accounts before choosing a fix. If activation is easy but recommendation remains weak, it needs to investigate the wider service proposition, including reliability, price and expectations.

Read explanations beside the numbers. A score tells you that a problem may exist; a useful customer account helps locate the mechanism. Our consumer-insights guide explains how to move from an account to a testable interpretation and a concrete change.

Sapien can help explore proposed experience changes with a data-grounded simulated audience, compare reactions and develop hypotheses about friction or expectations. Those modeled reactions should remain separate from scores collected after actual customer experiences. A synthetic respondent's recommendation rating is not Northline's customer NPS, and a simulated easy journey is not evidence that its routers activate successfully.

If the immediate decision is how to reduce activation friction, start with a specific effort question and the corresponding behavior. Use CSAT to understand the defined support experience and NPS to track the wider recommendation relationship. Add each measure when someone owns a decision it can inform, then return to customers and observed outcomes to determine whether the change worked.