What is Net Promoter Score?

Net Promoter Score is a customer loyalty metric built on one question: how likely is it that you would recommend this company, product, or service to a friend or colleague, answered on a 0-10 scale. Respondents are sorted into three bands. Promoters (9-10) are enthusiasts; passives (7-8) are satisfied but unmoved; detractors (0-6) range from lukewarm to actively hostile. The score is the percentage of promoters minus the percentage of detractors, giving a range from -100 to +100. Passives count in the denominator but not the arithmetic, a design choice with real consequences: converting a detractor to a passive raises the score exactly as much as converting a passive to a promoter.

Most implementations pair the score with an open follow-up ("what is the primary reason for your score?"), and practitioners will tell you the verbatims are where the actual value lives.

A Brief History

NPS was introduced by Fred Reichheld of Bain & Company in a 2003 Harvard Business Review article, "The One Number You Need to Grow", which argued that the recommend question outperformed longer satisfaction batteries at predicting growth, and that companies needed one simple loyalty number the whole organization could rally around. The idea landed at exactly the right moment: executives exhausted by forty-question satisfaction trackers embraced a single digit, and within a decade NPS was embedded in earnings calls, executive dashboards, and compensation plans across much of the corporate world. Academic researchers have spent the years since testing Reichheld's empirical claims, with decidedly mixed results; the metric's predictive superiority over alternatives has not held up well under independent scrutiny, even as its organizational usefulness as a shared, simple benchmark has kept it dominant.

Relationship vs. Transactional NPS

The question gets deployed in two distinct ways, and conflating them muddles everything. Relationship NPS surveys a sample of the customer base on a schedule (quarterly, say) and asks about the overall relationship; it is the trend-line, benchmark-friendly version. Transactional NPS fires after a specific event (a support ticket, a delivery, onboarding) and measures that touchpoint. Both are legitimate; a company that mixes their results into one number is averaging apples into oranges. Transactional deployments also carry the greater fatigue risk, because every team wants their touchpoint measured and the customer receives the sum of those wants.

Using NPS Well

1. Treat the verbatims as the product.
The score tells you that something moved; the open-text answers tell you what. Mine them systematically, with proper thematic analysis rather than a skim for quotable praise, and route themes to the teams that own them.

2. Mind who is answering.
NPS surveys are short, so response rates are decent, but the responding population still skews toward the engaged. Non-response bias can move an NPS more than any product change; watch response rates alongside the score, and be suspicious of improvements that coincide with declining participation.

3. Respect the noise floor.
With typical B2B sample sizes, an NPS bouncing between +31 and +38 quarter to quarter may be describing nothing but sampling variance. Report the sample size and margin of error next to the score, and resist narrating every wiggle. The concepts under statistical significance apply to NPS exactly as they do to any survey estimate.

4. Benchmark within your industry, or not at all.
Scores vary enormously by sector, country, and survey channel. A +30 can be elite in one category and mediocre in another, so cross-industry comparisons mostly generate noise and envy. Internal trend over time, measured consistently, is the more honest use of the number; treat it as one input to a wider benchmarking practice.

5. Never bonus the number without guarding the measurement.
The moment compensation attaches to NPS, the survey process attracts pressure: hand-picked send lists, coached responses ("anything less than a 9 means I failed you"), surveys timed after happy moments. Goodhart's law is undefeated. If NPS must feed incentives, separate the people measured by it from the people administering it.

The Benefits

NPS is short enough that people actually answer it, simple enough that an entire organization can understand it, and standardized enough to trend over time. It gives customer sentiment a seat in rooms where a forty-item satisfaction battery never gets read, and its follow-up question, taken seriously, is a steady pipeline of prioritized customer feedback. As a common language between product, support, sales, and the board, it has few rivals.

The Limitations

A single question measuring stated intention is a blunt instrument: it does not measure actual recommending behavior, it compresses eleven scale points into three coarse bands (a 0 and a 6 are the same "detractor"), and its famous claim to predict growth has substantial academic evidence against its superiority. The score moves with survey channel, timing, and culture as much as with the underlying sentiment. And it diagnoses nothing by itself: an NPS without verbatims and follow-up research is a mood ring with a decimal point. Treat it as a thermometer, useful for noticing a fever, useless for identifying the disease.

NPS is one instrument in the wider practice of customer experience measurement, which also draws on CSAT, effort scores, and the Voice of the Customer program that collects and acts on all of them.

The Short Version

NPS survives every wave of criticism because it does one organizational job superbly: giving everyone the same simple number to care about. Use it for that. Track it consistently, read the verbatims religiously, report the uncertainty honestly, keep incentives away from the measurement, and when the number moves, reach for real research (interviews, usability tests, targeted surveys) to find out why. The one number you need to grow was always a slight exaggeration; one number you can all agree to investigate is nearer the truth, and still worth having.