CSAT vs NPS vs CES: Which Metric Actually Predicts Churn

CSAT, NPS, and CES are three customer experience metrics that measure different things on different timescales. CSAT captures transactional satisfaction right after an interaction, NPS captures long-term brand loyalty, and CES captures how much work the customer had to do to get something resolved.

Confusing them is the most common CX program mistake. Each was designed to answer a specific question — using one to answer another guarantees you act on noise.

The standard way to split them is transactional vs relational. CSAT is transactional: it asks about one moment. NPS is relational: it asks about the whole relationship. CES sits in between — transactional in scope, like CSAT, but measuring effort rather than feeling, which is a different question about the same interaction.

Each metric uses a different formula, which is why their scores are not comparable. CSAT is the percentage of satisfied responses, NPS subtracts detractors from promoters on a −100 to +100 scale, and CES is the average ease rating (or the percentage who rated the interaction "easy").

Run the same 100 survey responses through all three and you get three numbers that can't be reconciled. Say 70 people score you 5/5, 20 score 4/5, and 10 score 3 or below. On a CSAT basis that's 90 satisfied of 100 — a 90% CSAT. Frame the identical sentiment as NPS (treating 9–10 as the 70 enthusiasts, 7–8 as the 20 lukewarm, 0–6 as the 10 unhappy) and you get 70% promoters minus 10% detractors — an NPS of +60. Same customers, "90" versus "60," neither convertible to the other. No formula links them, because one counts a share of a population and the other subtracts one share from another — which is the practical reason you report each separately and never average them into one "satisfaction" figure.

No. CSAT and NPS are not interchangeable and they often move in opposite directions. A customer can give you a 5/5 CSAT on a single resolved ticket and a 4/10 NPS the same week, because they are satisfied with that one interaction but frustrated by the broader product or pricing.

This is the most common source of "our metrics don't agree" confusion in support teams. CSAT is sampling one event. NPS is sampling a feeling about the whole company built up over months. They are measuring different layers of the customer relationship and should be expected to disagree.

Three operational rules follow:

• Never report them as a single combined "satisfaction" number. Average them and you lose both signals.
• Different teams own each metric. Support owns CSAT and CES. Marketing and product own NPS.
• Triangulate, don't convert. There is no arithmetic that derives NPS from CSAT or vice versa — the scales and the formulas disagree. Use them as separate lenses on the same customer base.

NPS is considered outdated because the peer-reviewed research never validated its central claim — that it predicts growth better than other satisfaction metrics — and because Bain's own consultants quietly walked it back in 2021 with a follow-up called "Net Promoter 3.0."

The original claim was made in Fred Reichheld's 2003 Harvard Business Review article "The One Number You Need to Grow", which presented NPS as the single best predictor of revenue growth across industries. The number caught on inside corporate boardrooms because it was easy to explain.

The academic rebuttal came four years later. Keiningham, Cooil, Andreassen, and Aksoy published "A Longitudinal Examination of Net Promoter and Firm Revenue Growth" in the Journal of Marketing (Vol. 71, No. 3, July 2007), testing the claim across 21 firms in 6 industries with multi-year data. Their finding: NPS performs no better than the American Customer Satisfaction Index at predicting firm revenue growth, and in several industries it performs worse.

Eighteen years after the original article, Reichheld himself wrote "Net Promoter 3.0" in HBR (November 2021) acknowledging that survey-score gaming and inconsistent measurement had undermined the metric, and proposing a replacement called the Earned Growth Rate — an accounting-based metric calculated from revenue retention and new customer revenue from referrals. When the author of a metric publishes a successor that abandons the survey entirely, that is the strongest available signal about the original.

None of this makes NPS useless. It is still a fine directional indicator of brand health when measured consistently and not tied to bonuses. But the claim that it predicts growth better than CSAT — the original reason it spread — does not hold up.

Customer Effort Score (CES) is the metric most closely tied to churn and repeat purchase, on the strength of research from the Corporate Executive Board (now part of Gartner) published as the 2010 HBR article "Stop Trying to Delight Your Customers" and expanded in the book The Effortless Experience (Dixon, Toman, DeLisi, 2013).

Dixon, Freeman and Toman's HBR article reports a study of more than 75,000 customers, and two findings from it reframed the field: 96% of customers who had a high-effort service interaction went on to describe themselves as disloyal, against 9% of those whose experience was low-effort, and reducing effort predicted repurchase intent better than exceeding expectations did. The body of that article sits behind HBR's paywall, so read those as figures HBR publishes rather than as anything we re-derived — the link is there so you can check them yourself.

The mechanism matters more than the numbers, and it is the part you can act on without taking anyone's study on faith. Customers rarely leave because one interaction was bad. They leave because friction accumulates. A customer who has to repeat their issue four times across three agents will go no matter how polite the final response was, and CES is pointed directly at that. CSAT often misses it, because the customer can rate the resolution 5/5 and have already decided to switch — CSAT samples one interaction, and the interaction is not what they are leaving over. That difference in what each question can see is the whole argument for CES, and it does not need a multiplier attached to it.

Gartner has published its own guidance on the metric, "How to Measure and Interpret Customer Effort Score (CES)" (Deborah Alvord, 18 November 2024), whose public abstract makes the narrower point that high-effort experiences are expensive to serve and work against loyalty. The full note is client-gated. If you see a precise industry-by-industry churn correlation attributed to it, that detail is not in anything Gartner has published openly.

The caveat: CES is a tactical metric. It tells you a specific interaction was hard, not why the customer might love or leave you over time. For long-term loyalty signal, you still need NPS or a similar relational measure. For predicting whether last month's ticket-takers will churn next quarter, CES wins.

There is no credible cross-industry table of "good" CSAT, NPS and CES scores, and any page that hands you one is guessing. The one benchmark with real measurement behind it is the American Customer Satisfaction Index, which put the US national score at 76.1 in Q2 2026, down from 76.7 in Q1 — and that is an index score, not a CSAT percentage, so it is not something your own 85% can be laid next to.

ACSI has run continuously since 1994 out of the University of Michigan's Ross School of Business, and its national score page is built on roughly 200,000 customer interviews a year covering more than 400 companies across 40-plus industries. The same page carries the per-industry breakdowns. Those bands, not a cross-industry average, are the thing to compare yourself against.

Direction is as informative as level. ACSI's Q2 2026 release is headlined "Economic Alarm Bells are Blasting": the national score has slid for several consecutive quarters, customer complaints are at record levels for the series, and outside of the COVID period this is the steepest decline of the century so far. For scale, the same index read 78.0 in Q1 2024. If your own scores have drifted down a point or two this year, part of that is the tide — and the way to tell which part is yours is to plot your line against your sector's, not against last year's version of yourself.

NPS benchmarks in particular travel badly, because the promoter-minus-detractor formula compresses results unevenly across industries. A number that reads as excellent in one sector is unremarkable in another, and there is no conversion between them. Cross-industry "average NPS" figures circulate widely and none of them come with a published method, so they are not worth the anxiety.

Three rules for reading benchmarks honestly:

1. Compare to your own industry's band, not a global average.
2. Track the trend line over four quarters. Direction beats absolute number.
3. Segment by customer cohort — new customers always score higher than year-three customers.

For internal targets, the only defensible method is to set them from your own baseline. Measure for a quarter, take the median, set the target one step above it, and re-set it every two quarters. The working thresholds we use are CSAT above 85% and CES above 5.5 on a 7-point agree scale, but those are operating conventions we chose, not research findings — and you should apply exactly that scepticism to any threshold you find presented as a benchmark, including ours.

The standard playbook is to run CSAT after every resolved ticket, CES after specific high-effort moments (self-service, account changes, returns), and NPS quarterly to a sampled cohort. Each metric gets its own owner, dashboard, and decision loop — never a single combined score.

A working setup for a mid-sized support team looks like this:

1. CSAT — post-resolution one-question survey, sent automatically when a ticket closes. Owned by the support manager. Reviewed weekly. Drives coaching for individual agents.
2. CES — sent after specific journeys: first onboarding, billing changes, returns, password resets. Owned by the head of CX or operations. Reviewed monthly. Drives process and product fixes.
3. NPS — sent quarterly to a stratified sample (new, expanding, at-risk segments). Owned by marketing or the customer success leader. Reviewed quarterly. Drives executive narrative and renewal forecasts.

The hardest part is not the math, it's discipline. Three failure patterns to avoid:

• Survey fatigue. Every survey you send spends goodwill you cannot get back, and the customers who stop answering first are the moderately-satisfied majority — which biases whatever is left toward the angry and the delighted. Cap total surveys at two per customer per quarter across all metrics, counting every team's sends together, not each team's separately.
• Score-gaming. Tying agent bonuses to CSAT or NPS produces score inflation, not behavior change. Reichheld's "Net Promoter 3.0" names this as a reason NPS lost credibility in many programs.
• Acting on outliers. A single 0/10 NPS from an angry promoter-turned-detractor isn't a signal. A two-quarter slide of 4 points across the same segment is.

If you have fewer than 15 agents, run CSAT only. Don't add NPS or CES until you have a sustained 80%+ CSAT response rate and a clear pattern of where you're losing scores. One metric tracked well beats three dashboards nobody reads.

The Effortless Experience research is compelling, but operationalizing CES correctly requires segmenting by interaction type — a level of survey design discipline that small teams rarely have bandwidth for. NPS is even worse for small teams: with a sample of a few hundred customers, the math is volatile enough that a single bad week can swing your score 10 points and tell you nothing.

The practical sequence for a 3–15-agent team:

1. Month 1–3: Add a one-question post-resolution CSAT survey to every channel. Aim for a 25%+ response rate before doing anything else.
2. Month 4–6: Read every comment. Tag the top 3 friction patterns. Fix the worst one.
3. Month 7–9: Re-measure. If CSAT holds above 85% with sustained response rates, add CES on your highest-friction journey (usually billing or onboarding).
4. Month 10+: Only then consider NPS, and only quarterly.

Converge ships agent-triggered CSAT surveys built into the unified inbox — a 5-star post-resolution prompt sent as a platform message — for $49/month flat rate covering up to 15 agents. The point is not the tool; it's that you should pick the smallest possible measurement footprint and grow it deliberately.

The four mistakes that kill CX programs are: tying agent compensation to scores, surveying too often, comparing across industries, and acting on individual responses instead of trends.

Each one is documented in the literature and visible in any audit of a struggling CX program:

• Compensation tied to scores. Customers get pressured into 5/5 ratings. CSAT inflates, real signal disappears. Reichheld's "Net Promoter 3.0" points at score-gaming as a reason NPS programs lose credibility, and the same effect turns up wherever a score is attached to money.
• Surveying too often. The same customer hit with a CSAT after every ticket and an NPS every month stops responding entirely. Response rates collapse, then leaders make decisions on a self-selecting fragment of angry or hyper-engaged users.
• Cross-industry comparisons. The same NPS can be a strong result in a sector customers resent and a weak one in a sector they choose enthusiastically. Without a sector-relative reference point, the number is decorative.
• Reacting to individual responses. A 3-month rolling trend on a segmented cohort is signal. One angry email is not. Both deserve responses, but only one drives roadmap decisions.

A fifth and increasingly common mistake: assuming AI summaries of open-ended survey responses are sufficient. They aren't. AI clustering is useful for triage, but the actual product and process decisions still require a human reading 50–100 verbatim comments per quarter.