NPS vs CSAT vs CES: Which Metric to Use When

A telecom operator can post an NPS of +42 in the same quarter its app store fills with three-star reviews about five-minute hold times nobody fixed. Both numbers are real. Neither is wrong. They are simply not answering the same question — and most organisations never stop to ask which question they actually need answered before they pick up a survey tool.

The debate over NPS versus CSAT versus CES usually gets framed as a contest: which one is the "true" measure of customer experience. That framing is the problem. Each metric captures a different psychological event, at a different point in the relationship, built to inform a different decision. The question worth asking isn't which score is best. It's which score tells you what you need to know, right now, about this touchpoint.

Short answer: use CSAT to gauge satisfaction with a specific, bounded interaction; use CES to measure how much effort a customer had to expend to get something done; use NPS to track overall relationship loyalty and willingness to recommend. Match the metric to the decision it needs to inform — not to a company-wide mandate for one number.

What do NPS, CSAT, and CES actually measure?

Each metric is a different lens on the same customer, not three attempts at the same measurement.

• Net Promoter Score (NPS) asks a single question — "how likely are you to recommend us to a friend or colleague?" on a 0–10 scale — and subtracts the percentage of detractors (0–6) from promoters (9–10). It was introduced by Fred Reichheld and popularised in "The One Number You Need to Grow" (Harvard Business Review, December 2003). It measures the relationship as a whole, coloured by everything the customer has experienced with you recently.
• Customer Satisfaction Score (CSAT) asks how satisfied a customer was with a specific transaction, typically on a 1–5 scale. Its methodology draws on decades of academic satisfaction research, including the American Customer Satisfaction Index, run by the University of Michigan since 1994. CSAT is episodic: it measures a moment, not a history.
• Customer Effort Score (CES) asks how easy it was to get something done — resolve an issue, complete a purchase, find an answer. It emerged from research by the Corporate Executive Board and was published by Matthew Dixon, Karen Freeman and Nicholas Toman in "Stop Trying to Delight Your Customers" (Harvard Business Review, July–August 2010). It measures friction, not feeling.

NPS tells you whether someone loves you enough to say so in public. CES tells you whether you made them work too hard to find out.

Why does NPS get chosen by default — and why is that often the wrong call?

NPS became the board-level metric because it is simple, benchmarkable across industries, and easy to report in a single line to a chief executive. None of that makes it the right instrument for diagnosing a specific touchpoint.

NPS measures the relationship, which means it is a lagging indicator. Asking "would you recommend us?" immediately after a password reset conflates two entirely different judgments: how the customer feels about this one interaction, and how they feel about the entire brand relationship — pricing, past experiences, competitor comparisons, even mood that day. A perfectly executed support call can still produce a low NPS response from a customer who is angry about last month's billing error. The metric isn't lying. It's just answering a broader question than the one you asked.

This is why NPS works well as a governance-level health check and poorly as a diagnostic tool. It tells the board whether the relationship is trending up or down. It tells the operations team almost nothing about which of last Tuesday's forty touchpoints caused the dip.

When should CSAT be the tool of choice?

CSAT earns its place immediately after a bounded transaction — a support call, a delivery, an onboarding step, a single page in a digital journey. It answers "did this specific interaction meet expectations?" and nothing broader.

That narrowness is the value. If a bank wants to know whether its new mortgage-application flow is working, NPS asked six months later is contaminated by everything that happened since. CSAT asked the moment the application is submitted isolates the variable you actually changed. It is the right instrument when you need to diagnose a specific process, not audit the whole relationship — which is exactly why it pairs so well with structured journey mapping that identifies where each transaction begins and ends.

The limitation is equally specific: CSAT is a satisfaction judgment, and satisfaction is a weaker predictor of future behaviour than effort. A customer can report being "satisfied" with a call that took forty minutes to resolve a five-minute problem — and still quietly start shopping for a competitor.

When does Customer Effort Score outperform both?

CES tends to be the stronger predictor of repurchase and loyalty behaviour specifically in service and support contexts, because friction — not the absence of delight — is what drives customers away. In their HBR research, Dixon, Freeman and Toman found that service interactions were far more likely to erode loyalty through effort and friction than to build it through above-and-beyond gestures; delighting customers moved the needle far less than simply removing obstacles from their path.

There is a behavioural mechanism behind this, and it is worth naming: loss aversion. Kahneman and Tversky's foundational work on prospect theory established that losses are felt roughly twice as intensely as equivalent gains. Effort is a small, immediate loss — time, cognitive strain, a second phone call. Delight is a distant, uncertain gain. Loss aversion means the small loss wins the fight for a customer's attention every time, which is precisely why an effortless password reset does more for retention than a surprise upgrade ever will. This is the same logic explored in our piece on loss aversion in pricing — the asymmetry between losses and gains shapes far more of customer behaviour than most CX programmes account for.

Deploy CES immediately after any high-friction task: a claim, a return, a self-service flow, a complaint resolution. It is the metric that catches the thing NPS and CSAT both miss — not whether the customer was happy, but whether you made them work for it.

How does the peak-end rule explain why these three metrics disagree with each other?

They disagree because they are sampling different moments of memory, not because one of them is faulty. Daniel Kahneman, Barbara Fredrickson, Charles Schreiber and Donald Redelmeier demonstrated in their 1993 study, "When More Pain Is Preferred to Less: Adding a Better End" (published in Psychological Science), that people judge an experience not by its average but by its peak intensity and, disproportionately, by how it ends. This is the peak-end rule.

Apply that to your survey programme and the disagreement between metrics stops looking like noise. NPS asked at a relationship checkpoint reflects whichever moment was most emotionally intense recently — often the most recent one, good or bad. A customer whose last twenty interactions were flawless but whose most recent renewal call was a nightmare will report NPS coloured heavily by that single ending, even while their CSAT scores across the previous year sit comfortably high.

A relationship metric asked at the wrong moment doesn't measure the relationship — it measures whichever memory happens to be loudest that week.

This is also why a company can hold a respectable NPS while individual CES scores in its contact centre are quietly deteriorating: the peak-end effect means one recovered escalation, handled with real care, can outweigh several routine but slightly effortful interactions in the customer's overall verdict — right up until the pattern of friction becomes the peak memory itself.

How do you decide which metric to deploy at each touchpoint?

Rather than choosing one metric for the whole organisation, map the decision each touchpoint requires and instrument accordingly. In practice, that means working through the journey deliberately:

1. Map the journey and separate bounded transactions from relationship-level moments. A password reset is a transaction; a contract renewal is a relationship moment. Structured journey mapping makes this distinction visible rather than assumed.
2. Attach CES to every high-friction task step — self-service portals, claims, returns, complaint handling, anything where the customer is trying to get something done under time pressure.
3. Attach CSAT to discrete service interactions with a clear beginning and end — a single support call, a delivery, an onboarding step — where you need to isolate one variable.
4. Reserve NPS for relationship checkpoints — post-renewal, quarterly for subscription businesses, annually for lower-frequency relationships like banking or insurance — where you are deliberately asking about the whole picture.
5. Weight your instrumentation by moments of truth, not evenly across every touchpoint. Surveying every click produces fatigue and noise; surveying the handful of moments that actually shape perception produces signal.
6. Feed all three into one closed-loop system, not three disconnected dashboards owned by three different teams. A unified feedback management approach is what turns three separate scores into one coherent picture of where the experience is breaking.

Done this way, the three metrics stop competing and start triangulating: NPS shows you the trend line, CSAT shows you which chapter caused the dip, and CES tells you whether the fix is about warmth or about removing obstacles.

Why does closing the loop matter more than the metric you chose?

Because none of the three predicts anything on its own. A score is a diagnosis, not a treatment. Organisations that survey rigorously but never act on the results are running measurement theatre — and customers notice when they are asked the same question repeatedly with no evidence anything changed.

Closing the loop means every detractor, every low-effort flag, and every poor CSAT response triggers a defined response within a set service level, not an eventual review in next quarter's steering committee. That discipline is what separates a Voice of Customer programme from a Voice of Customer archive. Our earlier piece on moving from listening to action with VoC data covers the mechanics; the principle here is simpler — a metric only earns its keep once it changes a decision.

This is also where governance discipline matters more than survey design. A Voice of Customer strategy that defines who owns each metric, what threshold triggers escalation, and how fast a response is expected will outperform a technically superior survey instrument with no accountability behind it.

Is there a single "best" CX metric?

No — and the search for one is the real failure mode, more damaging than picking the "wrong" metric for any given touchpoint. Reducing an entire customer relationship to one number flattens exactly the diagnostic power that having three distinct instruments provides. NPS, CSAT, and CES are not rival contenders for a single throne. They are three different questions, asked at three different moments, in service of three different decisions.

Organisations that mature past the single-score obsession tend to ask a more useful question: where in our journey are we even instrumented to notice a problem, and where are we flying blind between surveys? That question is worth answering honestly, ideally with a structured lens like a CX maturity assessment, before adding a fourth metric to the pile.

The number that matters is the one that moves

Pick the metric for what it will make someone do differently on Monday morning, not for how cleanly it fits on a slide. NPS that never triggers a root-cause review is decoration. CSAT that nobody reads after a bad week is data entry. CES that flags friction nobody has the mandate to fix is just a more precise way of watching customers leave. The metric earns its place in your programme only once it changes a decision — everything before that is measurement for its own sake.