What Is eNPS? The Formula, the Bands, and the Part About Small Teams

Short answer: eNPS, or employee Net Promoter Score, is one question scored on a 0 to 10 scale. People who answer 9 or 10 are promoters, 7 or 8 are passives, and 0 through 6 are detractors. The score is the percentage of promoters minus the percentage of detractors, which puts it somewhere between -100 and +100.

That is the whole metric, and every explainer says so. Here is the part they leave out: an eNPS compresses a team into three integers, and three integers plus a follow-up question is often enough to work out who the detractor was. That is not a flaw in how you run it. It is a property of the arithmetic, and it gets worse as the group gets smaller.

The question

The standard wording is a recommendation question:

How likely are you to recommend this organization as a place to work?

Answered on a 0 to 10 scale, usually with "Not at all" at one end and "Extremely likely" at the other. Many programs add one open follow-up asking what most influenced the score.

The recommendation framing is the point. You are not asking whether someone is satisfied, which people answer politely. You are asking whether they would put their own reputation behind the place to somebody they know, which they answer more carefully.

The formula

Take everyone who answered. Sort them into three bands:

Band	Scores	What it means
Promoters	9 to 10	Would actively recommend
Passives	7 to 8	Content, not enthusiastic
Detractors	0 to 6	Would not recommend, and may say so

Then:

eNPS = (% promoters) - (% detractors)

Passives count toward the denominator and toward nothing else. They dilute the score without moving it in either direction, which is deliberate: the metric is about advocacy, and a passive is not advocating.

A worked example

Fifty people answer. Twenty score 9 or 10, twenty score 7 or 8, ten score 6 or below.

• Promoters: 20 of 50, or 40%
• Detractors: 10 of 50, or 20%
• eNPS = 40 - 20 = +20

Note what the twenty passives did. They made up 40% of your respondents and contributed nothing to the number. If ten of them slipped to an 8 next quarter, your score would not move at all. If ten of them slipped to a 6, it would fall by twenty points. The metric is insensitive right up until it is very sensitive, which is one reason a single quarter's movement is weak evidence of anything.

eNPS versus NPS

Same arithmetic, different population and different stakes.

NPS asks customers whether they would recommend a product. eNPS asks employees whether they would recommend their employer. The mechanics carry over cleanly, but two things do not:

Your sample is small and fixed. A consumer brand surveys thousands of customers drawn from a much larger pool. You survey the forty people who work here, and they are the same forty every quarter. Sampling noise that averages out across thousands does not average out across forty.

Your respondents are not anonymous by default. A customer filling in an NPS survey has nothing at stake. An employee scoring their employer a 2 is documenting dissatisfaction inside a system their employer administers. Whether that is safe depends entirely on what the survey tool can see, which is the subject of the rest of this article.

What counts as a good eNPS

Honestly: nobody can tell you, and you should be suspicious of anyone who does with confidence.

The commonly repeated bands (above zero is fine, above 20 is good, above 50 is excellent) come from consumer NPS convention rather than from employee research. Published eNPS benchmarks vary enormously by industry, country, company size, and by how the vendor collecting them defines its sample. Two organizations reporting +30 may have measured populations with nothing in common.

What the number is genuinely good for is its own trend, measured the same way each time. Your eNPS this quarter against your eNPS last quarter, with the same question, the same scale and a comparable response rate, tells you something real. Your eNPS against an industry average from a vendor's marketing page tells you very little.

This is also why the scale must not drift. An eNPS asked on a 0 to 10 scale and an eNPS asked on a 1 to 5 scale are not the same measurement, and the bands do not survive the translation: there is no meaningful "9 or 10" on a five point scale. If a tool lets you edit the range of a question you labelled eNPS, your trend line can break without anything appearing to go wrong.

The part about small teams

Here is the problem nobody puts in the glossary entry.

Suppose you run an eNPS on a team of eight and publish the result to that team. The score comes back -25. That is enough to reconstruct the bands: with eight respondents, each person is 12.5 points, so -25 means the promoter count minus the detractor count is -2. Add a "what most influenced your score" follow-up, and one distinctive answer, and the pool of people who could have written it is very small.

Now add the thing that actually happens in practice: you run it again next quarter. The score moves from -25 to 0. Somebody changed their mind, and if one person left the team in between, the arithmetic will tell you a great deal about what they thought.

None of this requires anybody to behave badly. It requires only a manager with a calculator and a plausible reason to care. The three-integer summary that makes eNPS easy to put on a slide is the same property that makes it identifying at small n.

What to do about it

Set a reporting threshold and enforce it in the tool, not in the process. A rule that results are only shown above a minimum number of respondents is worth nothing if it depends on somebody remembering. It should be the software refusing to render the panel.

Be careful with cohort filters. The threshold on the whole poll does not protect a slice of it. Filtering a fifty person result down to "Engineering, joined this year" can produce a group of three, and the eNPS for that group of three is a very small number of people's opinions with a label attached.

Do not pair an eNPS with demographic questions unless you have thought it through. Department plus tenure plus location is frequently enough to identify one person on its own, before the score is even considered.

Say what you are doing. If the results are visible to managers, say so. If they are not anonymous, do not call them anonymous. People calibrate their answers to what they believe about who is reading, and a survey that overclaims its own privacy collects politer, less useful data than one that is honest about being confidential. We wrote about the difference in confidential versus anonymous and about how to tell if an employee survey is anonymous.

How often to run it

Quarterly is the usual answer and it is a reasonable one. The metric is a trend instrument, so it needs enough readings to have a trend and enough space between them for anything to have changed.

Monthly tends to produce noise plus survey fatigue. Annually produces a number that arrives too late to act on and gets attributed to whatever happened in the preceding fortnight. If you want a faster signal, a short recurring pulse is a better instrument than a more frequent eNPS. We cover that in anonymous pulse surveys versus annual engagement surveys.

How InviziPoll handles it

Disclosure: I am the founder of InviziPoll, so weigh this section accordingly. The reason it is here is that the design decisions are the article's argument made concrete.

eNPS is its own question type, and the range cannot be edited. A question created as an NPS is fixed at 0 to 10. There is no minimum or maximum to author, because the promoter and detractor bands are only meaningful on that scale, and a range that can be edited is a range that eventually is. The results view, the wave-over-wave trend and the export all read the same bands.

The score is computed from the responses, never estimated. Promoters, passives and detractors are counted, and the score is reported only when the range is exactly 0 to 10.

Importing a 0 to 10 rating column does not silently make it an eNPS. Bringing a survey in from another tool leaves a 0 to 10 rating as an ordinary scale question. Guessing would change what you are measuring without telling you, and if the guess were wrong you would find out from a trend line that had quietly re-based itself.

There is a minimum respondent count before a result is shown, and it applies to cohort slices as well as to the whole poll. The relevant detail is that it is enforced where the results are assembled rather than left to a setting somebody can turn off in a hurry.

Responses are encrypted so that we cannot read them. That is the part of the design that the rest depends on: a threshold enforced by a vendor who can read the underlying answers is a policy, and a threshold enforced over data the vendor cannot decrypt is an architecture. The difference matters exactly when somebody is motivated to look. See how zero-knowledge surveys work.

If you want to run one, there is a prebuilt eNPS pulse template: the recommendation question plus one follow-up asking what most influenced the score, which takes about a minute to answer. The recurring pulse series documentation covers running it on a schedule so the readings stay comparable, and cohort thresholds covers what happens when you slice the results.

The short version

eNPS is one question, three bands and one subtraction. It is a useful trend instrument for your own organization and a nearly useless comparison against anybody else's. Keep the scale fixed, keep the readings comparable, and remember that the compactness that makes the number easy to report is the same thing that makes it identifying when the group is small.