NPS vs CSAT explained

Why the distinction matters

In product analyst and PM interviews at companies like Stripe, Notion, and Airbnb, one of the most reliable filter questions is: "which customer satisfaction metric do you use, and why?" A junior says "NPS, it's the standard." A mid-level candidate explains that NPS captures long-term loyalty while CSAT captures satisfaction with a specific interaction — and that these are not interchangeable. A senior candidate walks through when each one is the wrong choice.

The same confusion shows up at work. Product teams track NPS as a quarterly health number. Support teams track CSAT after every closed ticket. Marketing teams build a blended dashboard. Then a release ships, CSAT drops 8 points, and the exec asks why NPS is still flat. The answer is structural: NPS is a lagging indicator of loyalty that takes a quarter or two to move, while CSAT is a leading indicator of the specific friction users just felt.

The short answer

• NPS (Net Promoter Score) — long-term loyalty, asked at the relationship level: "How likely are you to recommend us to a friend?"
• CSAT (Customer Satisfaction) — situational satisfaction, asked right after an interaction: "How satisfied were you with X?"

NPS belongs on the board deck. CSAT belongs in the support and onboarding dashboards.

Formulas and scales

NPS

The classic Bain & Company formulation. Scale is 0-10, respondents are bucketed:

• 9-10: Promoters
• 7-8: Passives
• 0-6: Detractors

NPS = % Promoters − % Detractors

The range is -100 to +100. Passives are deliberately excluded — the metric is asymmetric because real growth comes from active advocates, not lukewarm survivors.

CSAT

Scale is usually 1-5 (stars) or 1-7. The standard formulation counts the top two boxes:

CSAT = (Respondents giving top-2 box) / (Total respondents) × 100%

Some teams report the raw average instead. Both are valid as long as you stay consistent — switching mid-quarter breaks every trend chart you own.

A subtle point most blog posts skip: NPS is computed on the percentage difference between two buckets, so it's much more sensitive to small respondent counts than CSAT. With fewer than 200 responses, NPS swings wildly between weeks even when nothing changed.

Key differences

When to use NPS

NPS earns its place when you need one number that travels well across a board deck, an investor update, or a year-over-year comparison. It works best at the whole-relationship level — sent quarterly to the full active base, segmented by plan tier or cohort. It is also the metric of choice when you want a defensible external benchmark: Apple sits around 60, Tesla around 50, Netflix around 35, most B2B SaaS in the 20-40 band.

NPS is the right tool when:

The audience is executives or investors who want a single trend line. The cadence is quarterly or longer — long enough for behavior to actually shift. The decision is strategic: should we keep investing in this product surface? Are we losing ground to a competitor? Is the relationship deepening?

When to use CSAT

CSAT shines when the event is specific and recent — a ticket closes, an order ships, an onboarding step completes. You want a fast signal that this particular experience worked. CSAT is your default for:

The end of a support interaction (industry-standard touchpoint). The first 24-48 hours after onboarding completion. The checkout confirmation page for an e-commerce flow. After a major UX change is shipped to part of the base — pairs naturally with an A/B test. Whenever you need a leading indicator that surfaces in days, not quarters.

CSAT is also far easier to localize: "Were you satisfied with this delivery?" translates cleanly across markets, where "Would you recommend us to a friend?" has cultural baggage (in some markets, recommending services is socially uncommon, which depresses NPS artificially).

Using them together

The combination most healthy teams converge on:

1. CSAT after every key event — ticket close, onboarding, first purchase, post-release. Watch the weekly trend per segment.
2. NPS quarterly to the full active base — gated to users with at least 30 days of activity so first-day churn doesn't poison the result.
3. A correlation panel — when CSAT in a specific segment falls 3+ weeks in a row, expect NPS for that segment to dip the following quarter. If it doesn't, the CSAT drop was probably noise.

CES — the third metric

CES (Customer Effort Score) asks "How easy was it to resolve your issue?" on a 1-7 scale. It was popularized by the Harvard Business Review paper "Stop Trying to Delight Your Customers," which argued that reducing effort predicts retention better than driving delight in support contexts.

For a support org, the priority stack is usually CES first, CSAT second, NPS third. For a product org, the order flips: NPS at the top, CSAT in the middle, CES only where friction is the explicit hypothesis (checkout, onboarding, password reset).

When they diverge — a worked example

A consumer SaaS ships a redesign of the navigation. The redesign exposes a new AI feature on the home screen but accidentally buries the export button three clicks deep. Within two weeks:

• CSAT on the "export" event drops from 88% to 71% — users who tried to export had a bad time.
• NPS for the quarter is flat at +34 — the rest of the base hasn't even noticed the change, and the AI feature has earned a few new promoters that offset the detractors.

Three months later, the export-heavy power-user segment has churned to a competitor. NPS drops to +22, and the post-mortem traces the cause back to the original CSAT signal that nobody acted on. This is the textbook case for treating CSAT as a leading indicator and wiring it into release-monitoring dashboards, not just into the support team's queue.

Common pitfalls

The most common NPS mistake is comparing your number to a published industry benchmark without normalizing the survey channel. NPS collected via in-app prompt runs 10-20 points higher than NPS collected via email, because in-app captures users who are currently engaged. If your competitor publishes "NPS 55 via email" and you publish "NPS 62 via in-app," you have not won — you have measured a different thing. The fix is to lock one channel as your reporting channel and footnote it everywhere.

A related trap is survey fatigue inflation. When you send NPS to users who just received three CSAT prompts the same week, response rates collapse and the people who do respond are systematically skewed toward extremes — your power users and your active complainers. The fix is a survey governance policy: no more than one survey per user per 30 days, and never mix NPS and CSAT prompts in the same flow.

A third issue is treating CSAT as a vanity metric you can game. Teams under pressure start filtering the CSAT trigger to only-resolved tickets, only-satisfied-looking sessions, or only users who completed the happy path. The number climbs, leadership relaxes, and the actual problems compound underneath. The fix is locking the trigger logic in code, putting it under change control, and publishing the trigger definition next to the metric.

A fourth pitfall is the Apple Mail Privacy Protection problem. Since 2021, Apple Mail pre-fetches images in emails, which inflates open rates and changes the selection bias of email-triggered surveys. NPS sent via email now over-represents Apple users systematically. The fix is to either move to in-app for NPS, or to weight responses back to your true user distribution.

Finally, teams routinely read too much into a single quarter's NPS move. NPS calculated on fewer than 400 responses has a margin of error of roughly ±5 points, so a "drop from 42 to 38" might be statistical noise. The fix is publishing confidence intervals alongside the headline number — or better, only reporting NPS rolling-12-week so weekly noise washes out.

Benchmarks by industry

Benchmarks shift over time and by region. Use them as a sanity bracket, not a target.

If you want to drill metrics interview questions like this every day — including the SQL queries behind NPS and CSAT calculation — NAILDD is launching with 500+ problems for product analysts and PMs.

Related reading

• How to calculate CSAT in SQL
• How to calculate CES in SQL
• How to calculate CSAT by issue type in SQL
• Product metrics PM interview questions
• How to choose a North Star metric

FAQ

Can I use only NPS and skip CSAT?

You can, but you lose the ability to diagnose what went wrong. NPS will tell you the relationship is deteriorating; CSAT will tell you which interaction caused it. Most teams that try to run on NPS alone end up flying blind for 8-12 weeks between surveys and react too late. The cost of running CSAT alongside is low — one extra prompt at well-chosen touchpoints — and the diagnostic value is high.

Is NPS 50 a good score?

Above 30 is healthy. Above 50 is genuinely strong. Above 70 is Apple territory and almost certainly indicates either a niche enthusiast base or a survey selection effect. For a typical B2B SaaS, an NPS in the 25-45 band is normal; for an airline or telecom, +15 is excellent. Always compare against the right industry and the right survey channel before celebrating.

Should CSAT use a 1-5 or 1-7 scale?

Both are defensible. The 1-5 star scale is more intuitive for consumers and matches the App Store and Amazon mental model. The 1-7 scale gives more granularity and is preferred in academic and large-scale enterprise contexts because it produces more variance and is easier to analyze statistically. Pick one, stick with it for at least 12 months, and publish the scale in every report so readers can compare correctly.

How does NPS relate to revenue?

The empirical relationship in most consumer SaaS studies is that a 10-point NPS increase correlates with roughly 3-5% additional revenue growth over the following year, mediated through retention and word-of-mouth acquisition. The relationship is correlational, not causal — high-NPS companies also tend to have better products, better pricing, and better positioning, all of which independently drive revenue. Don't promise a CFO that fixing NPS by 10 points will mechanically add 5% to ARR.

What sample size do I need for NPS to be stable?

A practical minimum is 400 responses per reporting segment per quarter, which gives you roughly ±5 points of margin of error. For executive-level reporting on the whole base, aim for 1,000+ responses. Below 200 responses, NPS swings so much week to week that trend analysis is meaningless and you should report a rolling 12-week average instead of a point estimate.

Can NPS go negative, and what does that mean?

Yes — the theoretical floor is -100. A negative NPS means more detractors than promoters, which is genuinely bad and usually indicates either a serious product-market fit problem, a recent breaking change, or a survey targeting a churned-user list. The action is to segment immediately: is it the whole base, or one geography, or one plan tier? Negative NPS isolated to one segment is usually fixable; negative NPS across the board is a strategic emergency.