"NPS or CSAT — which should we use?" — anyone who has tried to measure customer satisfaction has hit this question.
Let us get to a conclusion early. "Which one to use" is, in fact, not a very important question. For most companies, the thing that actually matters is not the metric choice but two other things: (1) whether your response volume can produce a meaningful score at all, and (2) whether you have a process to actually read the open text attached to that score. This article lays out the differences between the two metrics — and then gets into that field reality.
The basics
NPS (Net Promoter Score)
"How likely are you to recommend us to a friend or colleague?" on an 11-point scale (0–10). Computed from promoters minus detractors. Measures overall customer loyalty.
CSAT (Customer Satisfaction Score)
"How satisfied were you with this experience?" on a 5- or 7-point scale. Measures satisfaction with a specific experience.
How they differ
| Dimension | NPS | CSAT |
|---|---|---|
| Measures | Loyalty / recommendation | Satisfaction with experience |
| Format | 11-point + open text | 5/7-point + open text |
| Timing | Whole relationship / on a cadence | Right after the experience |
| Year-over-year comparison | Easy | Easy (when same touchpoint) |
| Correlation with business outcomes | Strong (revenue growth etc.) | Weaker (operational quality) |
| Response volume needed | High (see below) | Works at lower volume |
NPS — strengths and limits
Strengths
- One question = low respondent burden
- Single number = common language for the exec team
- Industry benchmarks are widely available
Limits
- The reason behind the score is invisible from the number alone
- Cultural bias is large (Japanese-language NPS skews low globally)
- Insensitive to short-term campaign impact
- Unstable, high-error at low response volumes
CSAT — strengths and limits
Strengths
- The impact of specific touchpoints and initiatives is visible
- Combined with open text, improvement points are obvious
- Decomposable across departments (support, sales, product)
- A trend is readable even from a few dozen responses
Limits
- Susceptible to central bias (clustering on 3)
- Precision drops if not asked immediately after the experience
- Too low-altitude to use as a single executive metric
Three structural reasons NPS struggles at small and mid-size companies
Here is the heart of it. NPS is a global-standard metric, but when a small or mid-size company adopts it as-is, it very often ends up as a number that leaves everyone asking "so... is that good?" Three reasons.
Reason 1: Cultural bias breaks benchmark comparison
Japanese-language respondents have a strong psychological resistance to "giving a 10 out of 10," so NPS comes out structurally low. At identical real satisfaction, your score looks worse next to a Western company's published NPS. NPS's single biggest selling point — "you can compare against the industry average" — is only half usable in that context. If you compare, compare against your own prior-year period, not foreign benchmarks.
Reason 2: At low volume, error makes the score swing wildly
NPS spans a wide −100 to +100 range. With only a few dozen responses, a handful of promoters or detractors swapping moves the score by 10+ points. Report "NPS hit +8 this quarter" and you cannot tell whether that change is the result of an initiative or just noise. For a business with a few dozen customer touchpoints a month, NPS simply lacks the precision to tell "moved" from "didn't move."
Reason 3: Make it a KPI and the field starts manufacturing the number
Set NPS as a sales or CS KPI and Goodhart's Law kicks in (when a measure becomes a target, it stops being a good measure). Concretely: surveys go only to customers who seem happy, distribution to customers likely to score low gets "forgotten," and reps verbally ask "a 10 would really help us" at survey time. The number rises — and real customer satisfaction has not improved by one millimeter.
So choose by scale — the decision axis
Rather than "NPS or CSAT," the realistic move is to decide by your response volume.
| Customer touchpoints per month | Recommended metric mix |
|---|---|
| Up to 100 (small) | CSAT-centered. 5-point + reason right after the touchpoint. Don't force NPS in |
| 100–1,000 (mid) | Run CSAT as the spine; add NPS annually/semi-annually as a supplementary "fixed-point reading" |
| Over 1,000 (large) | NPS as an exec KPI, CSAT to decompose by touchpoint. The intended combination works |
NPS comes into its own at large businesses where responses arrive steadily. A small business that jumps on NPS gets worn down chasing noise. When in doubt, start with CSAT.
Combining them (for mid-to-large scale)
If your volume is sufficient, combining NPS and CSAT is the intended approach.
Pattern 1: NPS (semi-annual) + CSAT (per touchpoint)
Send NPS to all customers semi-annually to track year-over-year, while running CSAT at each touchpoint (support, post-purchase). When NPS drops, find the low-CSAT touchpoint and fix it — the default combination.
Pattern 2: NPS (post-onboarding) + CSAT (post-feature-use)
NPS right after onboarding to capture first impressions; CSAT per feature to drive UX improvement. Standard in SaaS and mobile apps.
The score is a tracking device — the insight lives in the open text
This is the part this article most wants to land.
NPS and CSAT alike contain not one improvement hint in the number itself. All a score tells you is the fact "better or worse than last time." Why it moved, and what to fix, all live inside the attached open text (the "why did you give that score" of Q2).
So before agonizing over metric choice, what you should actually do is:
- Whichever metric you use, always ask for the "reason" as open text
- Have a process to read, classify, and convert that open text into action
A program that chases the score and leaves the open text untouched is a doctor staring at a thermometer reading and never examining the symptoms. The metric is just an alert that "something happened somewhere." The diagnosis always happens in the comment field.
Example question set (combined)
[NPS]
Q1. How likely are you to recommend us to a friend or colleague?
(0 to 10)
Q2. Why did you give that score? (open text)
[CSAT — about your most recent support case]
Q3. How satisfied were you with the recent support case?
(5-point)
Q4. Anything we should change? (open text, optional)
Four questions, both loyalty and operational quality captured. Treat the Q2 and Q4 open text as the main body.
CES is also worth knowing
The third major metric is CES (Customer Effort Score) — "how much effort did it take you to accomplish your goal?" — strongly correlated with churn. Combining all three gives a more dimensional view of the customer experience (see CES complete guide).
Summary
- Before "NPS or CSAT," ask whether your response volume can produce a meaningful score
- Small businesses: start with CSAT. Add NPS once volume arrives steadily
- Used as-is, NPS tends to leave companies with a limbo number — cultural bias, error, and KPI-gaming
- For any metric, the improvement material lives in the open text, not the number
Repoan ships both NPS and SaaS CSAT templates and lets you deploy either in one click. The AI report feature aggregates and classifies open text in one pass — so an "open text over score" operating model runs at a realistic level of effort (see AI response analysis).
Related
- Employee version: eNPS — measuring and improving employee NPS
- Including CES: The complete guide to CES
- Reading what is behind the number: Survey error and bias types