# NPS Survey Methodology: How the Score Is Built, How Precise It Is and Where It Falls Short

NPS survey methodology: Bain's question and scoring, the margin of error, whether a change between waves is real, dashboards, and the method's critics.

Published 2026-07-03 · Updated 2026-10-12
Canonical: https://hercules.works/blogs/nps-survey-methodology

Net Promoter Score asks "How likely are you to recommend us to a friend or colleague?" on a 0 to 10 scale. Scores of 9 or 10 are promoters and 0 to 6 are detractors, and the score is the promoters' share less the detractors' share. With 400 answers and a variance near 0.65, the score carries a margin of about ±8 points, so small moves between waves are usually noise. Researchers have also questioned whether NPS predicts growth better than other loyalty measures. Hercules Works lists NPS surveys among its customer satisfaction methods.

## One Question, Three Groups and a Number That Moves More Than You Think

NPS is easy to ask and easy to compute, which is why it ends up on so many management dashboards. The harder parts are knowing how much of a quarter-on-quarter change is real, what to show next to the number, and what the score cannot tell you.

This guide covers the definition from its source, a two-wave worked example with intervals, a sample-size table, what an NPS dashboard should carry, and the published criticism. If you only need the basic margin-of-error calculation, our [Qualtrics competitors guide](/blogs/qualtrics-survey-competitors-in-india/) works one through for a single wave.

## The Definition, From Bain

Fred Reichheld introduced the measure in "The One Number You Need to Grow" in *Harvard Business Review*, December 2003 ([HBR](https://hbr.org/2003/12/the-one-number-you-need-to-grow)). Bain & Company publishes the method on its Net Promoter System site ([Bain](https://www.netpromotersystem.com/about/measuring-your-net-promoter-score/)):

| Element | Bain's wording |
|---|---|
| Question | "How likely are you to recommend us to a friend or colleague?" |
| Scale | "a zero-to-ten scale" |
| Promoters | 9 to 10 |
| Passives | 7 or 8 |
| Detractors | 0 to 6 |
| Score | "the percentage of customers who are promoters (those who scored 9 or 10) minus the percentage who are detractors (those who scored 0 to 6)" |

Bain says it uses 0 to 10 partly so people do not pick "one" thinking it is the top of a one-to-ten scale. Keep the 11 points, the end labels and the cut-offs identical across waves, languages and tools. If one wave labels 0 as "Not at all likely" and another as "Never", the two scores are not comparable.

A score can run from −100 (everyone a detractor) to +100 (everyone a promoter). Passives count in the base but in neither group.

## How Precise Is One NPS?

Brendan Rocks, in "Interval Estimation for the 'Net Promoter Score'" (*The American Statistician*, 2016), gives the variance of NPS as the promoter share plus the detractor share minus NPS squared, with NPS on a −1 to 1 scale ([Rocks, arXiv preprint](https://arxiv.org/abs/1601.07235)). The 95% margin is 1.96 × √(variance ÷ n). Because promoters and detractors pull in opposite directions, NPS is noisier than a single percentage from the same sample.

Assuming a variance of 0.65, the value in the worked example below, margins look like this:

| Answers | 95% margin on one score | 95% margin on a change between two equal waves |
|---|---|---|
| 100 | ±15.8 points | ±22.4 points |
| 200 | ±11.2 | ±15.8 |
| 400 | ±7.9 | ±11.2 |
| 800 | ±5.6 | ±7.9 |
| 1,600 | ±4.0 | ±5.6 |
| 2,000 | ±3.5 | ±5.0 |

Read the last column before you set a target. To call a 5-point rise significant at 95% confidence, each wave needs about 2,000 answers, and that is before allowing for statistical power. Rocks also tested interval methods on a large database of companies' scores and concluded that the simple (Wald) interval performs poorly and "should be avoided", recommending an adjusted version: add 3/4 to the promoter and detractor counts and 3/2 to passives before computing it. At a few hundred answers the two give almost the same result; for small cuts the adjustment matters more.

## Worked Example: Did NPS Really Rise This Quarter?

The answer counts here are hypothetical. A brand surveys 400 category users each quarter.

| | Wave 1 | Wave 2 |
|---|---|---|
| Promoters (9 to 10) | 152 (38%) | 168 (42%) |
| Passives (7 to 8) | 140 (35%) | 136 (34%) |
| Detractors (0 to 6) | 108 (27%) | 96 (24%) |
| NPS | +11 | +18 |
| Variance | 0.38 + 0.27 − 0.11² = 0.638 | 0.42 + 0.24 − 0.18² = 0.628 |
| 95% margin | ±7.8 | ±7.8 |
| Adjusted interval (Rocks) | +3.1 to +18.7 | +10.1 to +25.6 |

The change is +7 points. Its standard error is √(0.638 ÷ 400 + 0.628 ÷ 400) = 0.056, so the 95% margin on the change is ±11.0 points. A 7-point rise could be a real gain or sampling noise; this design cannot tell.

Now cut wave 1 to the 40 respondents in one city: 15 promoters, 13 passives, 12 detractors. NPS is +8, and the margin is about ±25 points. That interval runs from about −18 to +33. A dashboard that ranks cities on bases like this is ranking noise.

What to do instead: pool two waves for city cuts, set a minimum base below which a cut is not shown, and report the promoter and detractor shares separately so readers can see which one moved.

## What an NPS Dashboard Should Show

Many teams search for an NPS dashboard or analytics tool when the real need is a short list of things the dashboard must not hide:

| Element | Why |
|---|---|
| Score with its interval | Without it, every movement looks meaningful |
| Base size for every cut | Readers can judge small segments themselves |
| Promoter %, passive % and detractor % | The same NPS can come from very different mixes |
| Trend with intervals on each point | Shows whether a change exceeds noise |
| The verbatim "why" answers, coded into themes | The score says how many; the open answer says why |
| Fieldwork dates and question wording | Changes in method break the trend line |

The third row matters more than it looks. Rocks points out that a single NPS "can come from many (potentially rather different) distributions". An NPS of 0 can mean everyone is a passive, or half promoters and half detractors. Those two businesses need opposite actions.

## What Critics Have Found

Bain's page says the behaviour behind the scores is "linked to economic value", and the 2003 article that launched it was called "The One Number You Need to Grow". Keiningham, Cooil, Andreassen and Aksoy describe Reichheld as advocating NPS as "the single most reliable indicator of firm growth", and tested that. They used longitudinal data from 21 firms and more than 15,500 interviews from the Norwegian Customer Satisfaction Barometer to replicate the Net Promoter analyses. In their words, "the research fails to replicate his assertions regarding the 'clear superiority' of Net Promoter compared with other measures in those industries" (*Journal of Marketing*, vol. 71, no. 3, 2007; [abstract](https://research.hub.ku.edu.tr/entities/publication/6c8605c0-154e-4020-bcd9-2e7e3e1f1550/full)).

Two more limits follow from the arithmetic, not from any study:

- **Information is thrown away.** A move from 0 to 6 does not change the score, and neither does a move from 7 to 8. Keep the full 0 to 10 distribution in your data even if you report NPS.
- **It does not say why.** Always ask a follow-up open question, and read the answers yourself before relying on theme counts.

We have not found a public NPS benchmark for Indian categories that states its question wording, sample and fieldwork method. Without those, a "good NPS in India" figure cannot be compared with yours. Compare against your own earlier waves, or against competitors measured in the same survey.

## Your Customers or the Category: Two Different Studies

An NPS sent to your own customer list after a purchase or a support call measures their experience with you, and lets you follow up with the person who gave a 2. That needs your customer data and a consent basis for contacting them.

An NPS from a consumer panel measures how category users rate you and your competitors, asked the same way at the same time. It suits competitive tracking. It cannot close the loop with individuals, because panel respondents are not your customers of record and their identities are not shared.

## Running This Study on Hercules

Our [customer satisfaction page](/use-cases/customer-satisfaction) lists Net Promoter Score surveys, with an NPS score and promoter/detractor breakdown among the deliverables. The study that fits our panel is the second kind above: a competitive NPS among category users. Clients receive individual answers without identity data ([panel methodology](/panel-methodology)), so you can rebuild the score and intervals yourself, but you cannot call a detractor back.

The worked example at 400 answers a quarter is a 400-response Pro bundle each wave: ₹48,000 per wave, ₹1,92,000 for four quarters. If the 2,000-a-wave arithmetic above fits your decision better, price that before you set a target. The Free plan includes 30 responses, enough to see the 0 to 10 scale and its end labels on a phone in each language.

## Frequently asked questions

### How is NPS calculated?

Ask how likely the person is to recommend you on a 0 to 10 scale. Take the percentage who answer 9 or 10 and subtract the percentage who answer 0 to 6. The result runs from −100 to +100.

### What is the margin of error on an NPS?

1.96 times the square root of (promoter share + detractor share − NPS²) divided by the number of answers, with NPS on a −1 to 1 scale. At 400 answers and a variance near 0.65 it is around ±8 points.

### How many responses do I need to track NPS?

It depends on the smallest change you need to detect. With equal waves, spotting a 5-point change at 95% confidence takes about 2,000 answers per wave; 400 per wave only detects changes of about 11 points or more.

### Is NPS a reliable predictor of growth?

Bain says the behaviours behind the scores are linked to economic value, and Reichheld's 2003 HBR article presented NPS as the number you need to grow. A 2007 Journal of Marketing study using Norwegian data could not replicate the claim that NPS is clearly superior to other loyalty measures in the industries Reichheld cited.

### What is a good NPS score in India?

There is no published, method-consistent benchmark we can point to. Compare with your own previous waves or with competitors measured in the same survey.

### Should my NPS dashboard show only the score?

No. Show the interval, the base size, the promoter and detractor shares, and coded reasons from the open question. The score alone hides most of what changed.
