Employee Net Promoter Score (eNPS)

What does the eNPS question actually ask?

It asks about willingness to recommend the organization as a place to work, which is narrower than it is usually taken to mean. Recommending an employer is a social act performed in front of somebody else, and it carries reasons that have little to do with the respondent's own experience: whether they think the person asking would suit the place, whether the name is useful on a resume, whether a referral payment is involved, whether they simply dislike making recommendations at all. Someone can be well treated, productive and planning to stay for years and still recommend the place to nobody, and someone can recommend it warmly while already interviewing elsewhere. The index measures what it says on its face. Every additional meaning attached to it is inference.

What does the eNPS arithmetic discard?

The distribution. Responses are sorted into promoters, passives and detractors, and the reported figure is a subtraction between two of those groups, so the shape of the population disappears into a single difference. Two workforces with entirely different profiles can therefore return the identical number: one where opinion is mild and clustered in the middle, another split hard between people who would recommend the place without hesitation and people who would warn others away. Those are not the same organization and they do not call for the same response, but the index cannot tell them apart. This is a structural property of the calculation rather than a flaw in how anyone runs it, which is why the underlying counts should always travel with the figure.

Why does eNPS swing so hard in a small population?

Because each response carries a large share of the result. In a small team, or in any segment thinned by filtering, a couple of people moving between groups shifts the figure a long way, where the same movement would be invisible in a large population answering identically. The consequence is that most of what is seen at team level is noise: it reflects who happened to answer this time, or who had a bad week, rather than any change in the team. Reporting that movement back to the team as a change makes it worse, because people will look for a cause, find one, and act on it. Below the size at which the figure is stable, the responses are worth reading and the score is not worth quoting.

Why can an eNPS figure not be compared with another organization's?

Almost nothing that shapes the number is held constant between two employers. The populations differ in what work they do, how long people have been there and what they were promised on joining. The wording differs, and small changes to a question about recommendation move answers noticeably. The timing differs, and a survey sent the week after a reorganization is measuring the reorganization. Most of all, candour differs: a workforce that believes answers are read carefully and one that believes they are not will report different figures from identical experiences, and the second will report the better one. A published number from another company is the output of a process you cannot see, applied to a population you know nothing about.

The same objection applies inside a single employer, which is the part that gets missed. Engineering and field sales do not answer this question the same way; a country where open criticism of an employer is socially costly will not answer it like one where it is not; a team formed last quarter answers differently from one that has been together for years. Comparing functions or geographies and then acting on the difference usually surfaces a property of the population rather than of the management. Where the attributes used to segment results sit in the same [HRMS](/hrms) as tenure, location, function and manager history, it is at least possible to see which explanation the gap is consistent with before anyone is asked to account for it.

If there is no good score, what should the number be baselined against?

There is no threshold that is good, and the absence is not a gap waiting for better research to fill it. A figure carries meaning only relative to a population and a method, and since neither is comparable between employers, there is no external value it can be held against. What remains is your own series: the same question, the same wording, the same population definition, the same point in the cycle, wave after wave. Measured that way a movement is at least evidence about something. Where the series is kept alongside the rest of the workforce record rather than inside a survey product, an [HR analytics](/hr-analytics-software) view can show what else changed between waves, which is the only route to an honest reading of a movement.

The second baseline is internal difference rather than internal comparison. It is worth knowing that one segment sits far from the rest, not so that segment can be handed a target, but because the gap is somewhere to go and ask questions. The distinction sounds thin and it decides how the whole exercise is received: a gap used as a prompt produces a conversation, and a gap used as a verdict produces a manager who spends the next wave protecting a number. The third baseline is the one people skip, which is the response rate and who answered, because a figure that moved while the sample changed shape has not necessarily said anything about sentiment at all.

Which part of an eNPS survey tells you what to do?

The free-text comment, and it is the part most often discarded. The index registers that something is true of a population; it never names the thing. A downward movement is equally consistent with a pay round that landed badly, a well-liked leaver, a policy change, a manager who left, or nothing at all, and the figure is identical in every case. The comment is where the cause is stated in the respondent's own words, and it is the only part of the instrument that can be acted on directly. Reading comments costs time and does not aggregate neatly into a slide, which is why it gets skipped in favour of the number it exists to explain. Collecting them is easy; most [engagement software](/employee-engagement-software) will do it. Reading them is the work.

What follows is ordinary qualitative work. Group comments by what they are about rather than by whether they are positive, since praise and criticism of the same manager are evidence about the same subject. Count how often a theme appears rather than how strongly it is expressed, because one furious comment reads louder than a dozen mild ones describing the same problem, and the dozen is the more useful signal. Then pick from the themes rather than from the score. The figure can indicate that this wave is worth a closer look; only the comments can say what the look should be about, and anything done on the number alone is a guess dressed in arithmetic.

What decision can an eNPS number legitimately support?

It can support looking. It cannot support a conclusion about how well a team is managed, whether a policy worked, or whether anybody intends to stay, and it is regularly asked to carry all three. The difficulty is attribution: everything that happened between one wave and the next competes to explain a movement, including events with nothing to do with the workplace, and the index offers no way to separate them. A wave sent during a restructure, in the week of a pay decision, or shortly after a well-regarded colleague resigned reads differently from one sent in a quiet month, and the difference is real without being informative. Treat a movement as a prompt to go and find out, and the number keeps earning its place.

The reason to keep asking it despite all of this is cost. It is one question, it takes a moment to answer, and it can be run often enough to notice that something has shifted before a longer instrument would have caught it. That is a genuine use and a modest one. The failure is not running eNPS; it is promoting it from a tripwire to a verdict, at which point it acquires a target, the target acquires an owner, and the owner acquires an incentive to manage the number rather than the thing it stood in for. Kept in its place it is among the cheaper things an employer can measure. Given weight it cannot hold, it quietly stops being measurement.