CSAT vs NPS vs CES: Which Customer Experience Metric Should You Use?

Bottom line: CSAT, NPS, and CES measure three different things, so the right choice depends on the question you are asking. Use CSAT (Customer Satisfaction) to measure how happy a customer was with a specific interaction; NPS (Net Promoter Score) to gauge long-term loyalty and benchmark against competitors; and CES (Customer Effort Score) to find friction and predict retention after a task or support interaction. None is sufficient alone — the strongest programs run all three and, critically, follow every score with the open-ended "why."

What each metric is

CSAT asks customers to rate satisfaction with a specific experience — a purchase, a support ticket, an onboarding step — usually on a 1–5 scale. You report the percentage of respondents who chose the top one or two boxes (for example, 4–5). It is the most intuitive and flexible metric, ideal for evaluating a single touchpoint.

NPS, introduced by Fred Reichheld in the 2003 Harvard Business Review article "The One Number You Need to Grow," asks how likely a customer is to recommend you on a 0–10 scale. Respondents are grouped into Promoters (9–10), Passives (7–8), and Detractors (0–6); NPS = %Promoters − %Detractors, producing a score from −100 to +100. It measures the durable, relationship-level sentiment that correlates with growth.

CES, popularized by the 2010 HBR article "Stop Trying to Delight Your Customers," asks how much effort a customer had to expend, typically as agreement with a statement like "The company made it easy to handle my issue." The underlying finding: reducing customer effort is a stronger driver of loyalty than exceeding expectations.

When to use each

Use CSAT when you want to evaluate a specific, recent interaction — a support agent, an onboarding flow, a feature, or a purchase. It is the right tool for "did this particular thing land well?"

Use NPS when you want a north-star measure of overall customer health, want to benchmark against competitors and industry norms, or want to identify your most loyal promoters for referrals and case studies.

Use CES when you want to find and remove friction — in support, checkout, or self-service — and predict whether customers will stick around. High effort is one of the strongest predictors of churn.

Benchmarks (and why they are a trap)

• CSAT: 75% or higher is generally considered strong and 85%+ excellent, though norms vary widely by industry.
• NPS: Scores are best read relative to your industry; a "good" absolute NPS in software looks very different from one in insurance, which is exactly why NPS shines as a benchmarking tool.
• CES: There is no universal cross-industry benchmark because platforms use different scales. Gartner has suggested that on a percentage-based version, scores below 70% flag room for improvement and above 90% signal a strong position.

Treat all benchmarks cautiously: scale design, question wording, and timing change scores more than real performance does. Your own trend over time is more trustworthy than any external number.

The shared blind spot: the score tells you what, not why

Every one of these metrics produces a number — and a number alone cannot tell you what to fix. An NPS of 32 does not say whether detractors are angry about pricing, reliability, or support. A CSAT of 60% does not say which part of onboarding frustrated people. The insight lives in the open-ended follow-up, and that is exactly where traditional survey programs collapse: the verbatim comment box gets a one-line answer (or nothing), and analyzing thousands of free-text responses by hand is so slow that most teams never do it.

How Koji turns scores into the reasons behind them

Koji is an AI-native research platform that keeps the quantitative metric and adds the conversation that explains it — at scale, automatically.

• Run the metric as a structured question. Koji's six structured question types — open_ended, scale, single_choice, multiple_choice, ranking, and yes_no — let you field a clean NPS (0–10 scale), CSAT (1–5 scale), or CES (agree–disagree scale) and capture the exact, comparable numbers you would expect. See the structured questions guide.
• Auto-probe every score. Instead of a dead comment box, Koji's AI interviewer asks an intelligent follow-up tuned to the answer: a Detractor is gently asked what went wrong; a Promoter is asked what they would tell a friend. Every score arrives with its reason attached.
• Thematic analysis at scale. Koji automatically clusters thousands of those reasons into themes and quantifies them — "41% of Detractors cited slow support" — so you can act on the drivers, not just watch the number move.
• Real-time reporting means you see the score and the story the same day, in voice or text, without a researcher manually reading transcripts.

While legacy survey tools like SurveyMonkey or Delighted give you a metric and a pile of unread comments, an AI-native platform like Koji gives you the metric, the verbatim reasons, and the ranked themes behind every point of movement — and teams using AI-assisted research report dramatically faster time-to-insight. You do not need a dedicated research team to run it.

How to combine all three

The metrics are complementary, not competing:

• NPS as the relationship-level north star, surveyed periodically.
• CSAT at key transactional touchpoints (post-purchase, post-onboarding, post-support).
• CES specifically where friction kills retention (support resolution, checkout, self-service).

Map each to a moment in the customer journey, always pair it with an AI-moderated "why," and you get a complete picture: how customers feel overall, how each interaction lands, and where effort is driving them away.

Related Resources

• Structured Questions Guide — field NPS, CSAT, and CES with the right scale types
• NPS Survey Guide — design and benchmark Net Promoter Score
• CSAT Survey Guide — measure satisfaction at every touchpoint
• Customer Effort Score Guide — reduce friction and predict churn
• NPS Follow-up Interviews — turn scores into reasons
• Voice of Customer Research Program — build a continuous listening system