NPS vs CSAT vs CES: Which Metric Actually Predicts Customer Retention?

In 2003, Fred Reichheld published a Harvard Business Review article called “The One Number You Need to Grow.” It introduced Net Promoter Score — a single question (“How likely are you to recommend us?”) producing a single number (-100 to +100) that Reichheld argued correlated with revenue growth better than any other customer metric. Within a decade, NPS had become the default customer satisfaction metric at most Fortune 500 companies. Bain & Company, which Reichheld co-founded, built a consulting practice around it.

There’s one problem: after 20+ years of accumulated data, the research on whether NPS actually predicts revenue growth is… mixed at best. A 2019 paper in the Journal of Marketing (Keiningham et al.) found that Customer Effort Score is a stronger predictor of repurchase intent than NPS in most B2C transactional contexts. A separate Bain & Company analysis found that NPS outperforms CSAT in predicting long-term growth across B2B accounts. A 2022 MIT Sloan Management Review analysis found that the predictive validity of NPS varies dramatically by industry, with some sectors showing near-zero correlation between NPS and growth.

So who’s right? The answer, which is simultaneously more and less satisfying than you might hope: all three metrics work — for different things, at different points in the customer journey. This article gives you the unvarnished comparison, the research behind each metric, and the practical framework for using them together.

Quick Answer: NPS (Net Promoter Score) measures loyalty and long-term growth potential. CSAT (Customer Satisfaction Score) measures satisfaction with a specific interaction. CES (Customer Effort Score) measures how easy it was to accomplish something. For predicting customer retention: CES wins in B2C transactional contexts. NPS wins for overall relationship health and B2B accounts. CSAT wins for measuring individual support tickets or feature releases. Most teams should use all three at different touchpoints — and add video feedback to understand the why behind all three numbers.

NPS Explained (And Its Hidden Weaknesses)

NPS is calculated by asking one question: “On a scale of 0–10, how likely are you to recommend [company] to a friend or colleague?” Respondents scoring 9–10 are Promoters; 7–8 are Passives; 0–6 are Detractors. The formula: percentage of Promoters minus percentage of Detractors. Score range: -100 (everyone is a Detractor) to +100 (everyone is a Promoter).

Good NPS varies significantly by industry. According to Bain & Company’s 2026 industry benchmarks: software/SaaS averages +41, e-commerce averages +45, financial services averages +34, healthcare averages +27, telecommunications averages +18. Comparing your NPS to an industry average without knowing the methodology — sample size, survey timing, response rate — is methodologically questionable. But it’s the comparison everyone makes.

NPS is a relationship metric, not a transactional one. It measures how a customer feels about you overall, not how they felt about the interaction they just had. That’s its strength for B2B: a customer who’s been with you for three years and has opinions about your product, support, pricing, and roadmap — their NPS reflects the full relationship, not a recent moment.

The weaknesses are real, and the industry doesn’t discuss them enough. First: NPS is a lagging indicator. By the time NPS drops, the churn has often already started. The customer who’s going to cancel next month has already formed a 6/10 opinion; the NPS survey just captures it after the decision is largely made. Second: the follow-up question — “What’s the main reason for your score?” — is where the real insight lives, and most companies either don’t ask it or don’t analyze the text responses at scale. A 10-point NPS drop is alarming. “What’s the main reason?” responses that cluster around “onboarding was confusing” is actionable.

Third, and rarely discussed: NPS response bias is significant. Happy customers respond to surveys. Frustrated customers churn silently. The population of customers who answer your NPS survey overrepresents engaged customers, which means your NPS score is almost certainly higher than the true opinion of your full customer base. Bain has acknowledged this in their own methodology documentation.

CSAT Explained (And When It Misleads You)

CSAT is simpler: after a specific interaction (support ticket resolved, onboarding call completed, feature released), ask “How satisfied were you with [this interaction]?” on a 1–5 scale. Calculate: (number of satisfied responses, i.e., 4s and 5s) / (total responses) × 100. A CSAT of 85% means 85% of respondents rated the interaction a 4 or 5.

CSAT’s strength is its specificity. It measures a discrete moment, not a diffuse relationship. If your CSAT on support tickets drops from 88% to 72% after you changed your response time SLA, you know exactly what to fix. NPS wouldn’t tell you that — or it would tell you three months later, after the drop had already compounded.

The weakness: response bias is severe. Happy customers respond to CSAT surveys. Customers who had a genuinely bad experience — the ones who most need to be heard — often churn without responding. The SHRM research on candidate experience applies equally to customer experience: the most negative segment is systematically underrepresented in survey populations. Your 85% CSAT might reflect the 30% of customers who bothered to respond, skewing toward those who were satisfied enough to engage.

CSAT also tells you almost nothing about loyalty. A customer can rate a support ticket 5/5 and still churn next month because your product doesn’t solve their core problem. CSAT is the right tool for measuring execution quality on specific interactions. It is not the right tool for predicting whether someone will renew.

CES Explained (The Underrated Metric)

Customer Effort Score was introduced in a 2010 Harvard Business Review article by Dixon, Freeman, and Toman — “Stop Trying to Delight Your Customers” — and the central finding was counterintuitive: reducing customer effort is a stronger predictor of loyalty than delighting customers. Exceeding expectations marginally improves loyalty. Making something hard to do destroys it.

The CES question: “How easy was it to [accomplish what you were trying to do]?” on a 1–7 scale (1 = very difficult, 7 = very easy). Sometimes stated as: “The company made it easy for me to handle my issue” (agree/disagree). The calculation is simpler than NPS: average the scores, or calculate the percentage who rated 5–7.

The HBR research found that 96% of customers who had a high-effort experience became more disloyal, compared to only 9% who had a low-effort experience. That asymmetry — effort destroys loyalty far more reliably than delight creates it — is the core insight. Customers don’t need to love you. They need to not find you frustrating.

CES wins for: SaaS onboarding (how easy was it to get started?), support resolution (how easy was it to get your issue resolved?), and checkout flows (how easy was it to complete your purchase?). These are high-effort moments — the ones where friction most directly predicts churn. A customer who found onboarding difficult is 40% more likely to churn within 90 days than one who found it easy, regardless of how much they like your product’s features.

Head-to-Head: NPS vs CSAT vs CES

Dimension	NPS	CSAT	CES
What it measures	Loyalty and advocacy potential	Satisfaction with a specific interaction	Ease of accomplishing a task
Question asked	“How likely to recommend?” (0–10)	“How satisfied were you?” (1–5)	“How easy was it?” (1–7)
When to use	Quarterly relationship pulse	Post-interaction (support, onboarding)	Post-task (checkout, setup, resolution)
Predicts churn	Moderate — lagging indicator	Weak — interaction-specific	Strong — best predictor of churn in B2C
Predicts growth	Strong in B2B, mixed in B2C	Weak	Moderate
Industry benchmarks available	Yes (Bain, Satmetrix)	Yes (varies by industry)	Limited
Best for B2B	Yes	Interaction-level only	Onboarding and support
Best for B2C	Moderate	Post-purchase, post-support	Yes — strongest predictor
Response bias risk	High	Very high	Moderate
Primary weakness	Lagging; doesn’t explain why	Doesn’t predict loyalty	Limited benchmark data; narrow scope

The Problem All Three Share (And Why Video Changes It)

NPS, CSAT, and CES are all numeric. They tell you what happened — a score went up or down. They almost never tell you why. A customer gives you 6/10 NPS. Is that because the onboarding was confusing? The pricing increased? A competitor offered them something better? The product is missing a feature they depend on? The support team was slow? The score contains none of that information.

The follow-up open-text question is the standard solution — “What’s the main reason for your score?” — and it’s better than nothing. Text responses give you some signal. But analyzing hundreds of open-text responses at scale requires either expensive human review or NLP tooling that produces broad categories without nuance. “Product issues” is a category. What specific product issue, from whose perspective, in what context, is the insight.

Video feedback does what text can’t. A 90-second recorded response to “What’s one thing we could do better?” contains: the specific issue, the emotional weight attached to it, the context in which it arose, the customer’s language for describing it (which is different from your internal product language), and — often — an implicit solution. Watch 20 of those videos in a week and patterns emerge that 200 text responses would bury in aggregate.

RecRam’s video forms make collection frictionless — a link sent via email or embedded on any page, no account required from the respondent, recorded on any device. RecRam’s AI analysis transcribes every response, clusters by theme, and surfaces sentiment patterns across the full response set. The result: you get the quantitative signal from your NPS, CSAT, and CES scores, and the qualitative texture from video that explains what the numbers mean and what to do about them. That combination — score plus story — is what customer insight actually requires.

How to Use All Three Together (The Customer Feedback Stack)

The teams that get the most from customer feedback metrics don’t choose between NPS, CSAT, and CES. They deploy each at the moments where it’s strongest.

Relationship NPS: Quarterly, to all accounts. This is your pulse on overall relationship health. Send it 45–60 days after onboarding (to capture the post-honeymoon reality), then quarterly. For B2B teams managing accounts, NPS by account tier — enterprise, mid-market, SMB — tells a more useful story than aggregate NPS. An enterprise account with NPS of 7 is a retention risk worth calling. The same score from 50 SMB accounts might reflect a pricing perception issue that marketing can address.

Transactional CSAT: Post-support ticket, post-onboarding, post-training session. Anything with a discrete beginning and end is a CSAT moment. Automate the trigger — ticket closed, onboarding call completed, implementation signed off — and the survey sends itself. Use CSAT as a leading indicator for support quality and implementation effectiveness, not as a proxy for overall customer health.

CES: Post-product interaction, post-checkout, post-setup. The moments of highest friction in your product experience. New user completes their first workflow — CES. Customer completes a complex configuration — CES. First purchase completed — CES. These are the exact moments where effort converts into loyalty or churn. You need the signal while you can still act on it.

Video feedback: At-risk accounts, post-churn interviews, new feature beta users, NPS detractors. Video feedback is highest value when the stakes are highest. An NPS detractor who records a 90-second video explaining their score is giving you a roadmap to win them back — or at minimum, to prevent the next customer from having the same experience. A churned customer who records a post-exit video is giving you information that no retention survey could capture, because they no longer have any reason to soften their answer. RecRam’s product solution is built for exactly this workflow — connecting quantitative survey signals to qualitative video responses at the moments that matter most.

Frequently Asked Questions

What is a good NPS score?

It depends on industry. According to Bain & Company’s 2026 benchmarks: above +50 is excellent in most sectors, +20 to +50 is good, 0 to +20 is acceptable, and negative scores indicate significant loyalty risk. SaaS companies typically benchmark between +35 and +55. Telecom and cable companies often have negative NPS — which reflects industry dynamics more than individual company performance. Always compare within industry, not to an absolute standard.

Should I use NPS or CSAT for customer success teams?

Both, for different purposes. CSAT is the right operational metric for CS teams — it measures the quality of individual interactions (onboarding calls, QBRs, support escalations) and gives CSMs immediate feedback on their execution. NPS is the right strategic metric — it reflects the overall account relationship and predicts renewal risk at an account level. Many high-performing CS organizations track CSAT weekly, NPS quarterly, and use the gap between a high CSAT and a declining NPS to identify accounts where interactions are going well but the underlying product-market fit is eroding.

What response rate should I expect from NPS surveys?

Email NPS surveys to existing customers typically generate 15–35% response rates, depending on relationship strength, timing, and subject line. In-app NPS surveys (triggered while the customer is using the product) generate 30–60%. SMS NPS surveys can reach 45–70% in consumer contexts. Response rates below 10% should trigger a methodology review — low response rates produce unreliable scores and make it impossible to segment meaningfully by account type, product line, or cohort.

How often should I run NPS surveys?

Quarterly relationship NPS for existing customers is the standard. More frequent than quarterly risks survey fatigue; less frequent than quarterly means you’re missing early churn signals. For transactional NPS (sent after a specific event), timing is event-triggered rather than calendar-based. One caution: don’t survey the same customer with relationship NPS, transactional CSAT, and CES within the same 30-day window — survey fatigue is real and reduces response rates across all your metrics.

Can AI replace customer surveys entirely?

Not yet — and probably not for the near future. AI can analyze existing feedback at scale (support transcripts, review text, video responses), but it can’t generate the structured, comparable signal that comes from a consistent survey instrument applied at consistent moments. The most effective approach is AI-assisted analysis of qualitative feedback combined with structured quantitative surveys — not a replacement of one by the other. The combination of a 15-point NPS drop (what) and AI-clustered video feedback explaining the pricing perception shift (why) is more actionable than either source alone.