We will take one survey, run it through the eNPS calculation from start to finish, and then check how much faith the result deserves. No theory first; the theory will make sense after the sums.

The set-up

A company with 400 employees sends a pulse survey. One question is the standard: “On a scale of 0 to 10, how likely are you to recommend this company as a place to work to a friend or family member?” Employee Net Promoter Score, eNPS, borrows its method from the customer-side Net Promoter Score, a measure introduced by Fred Reichheld and popularised through Bain and Satmetrix in the early 2000s.

250 people respond. That is a response rate of 250 ÷ 400 = 62.5%.

Step 1: sort every answer into a bucket

The convention is fixed:

  • 9 or 10: promoters.
  • 7 or 8: passives.
  • 0 to 6: detractors.

Count them. Say the tally is 95 promoters, 90 passives and 65 detractors. Check that they add up: 95 + 90 + 65 = 250. Always check; a stray blank row in a spreadsheet is a common source of error.

Step 2: convert to percentages and subtract

Promoters: 95 ÷ 250 = 38%. Detractors: 65 ÷ 250 = 26%. Passives take no part in the subtraction.

eNPS = 38 − 26 = +12.

The result lies between −100 (everyone a detractor) and +100 (everyone a promoter). It is reported as a number, not a percentage, though it comes from percentages. Notice what has been thrown away: a person who scored 6 and a person who scored 0 count the same, and so do a 9 and a 10. That is a limitation of the method, not a mistake in your sums.

Step 3: cut it by group

A single company-wide +12 hides everything interesting. Suppose the 250 responses include 40 from the customer support team, with 10 promoters, 12 passives and 18 detractors. Promoters are 10 ÷ 40 = 25%, detractors 18 ÷ 40 = 45%, so support’s eNPS is 25 − 45 = −20. Now we have a lead: something about support is different from the rest.

Segments worth cutting by: department, location, tenure band, manager, grade, and work mode. Keep a minimum group size for reporting, often five to ten responses, so nobody can be identified from the numbers. Promised anonymity is the foundation of honest answers; breaking it once ends the survey programme.

Step 4: ask how much the number can move by chance

Here is the part that most dashboards skip. A survey is a sample, so the score has uncertainty. Treat each respondent as +1 (promoter), 0 (passive) or −1 (detractor). The eNPS is the average of those values times 100. For the company-wide result the mean is 0.12, the spread of individual values is about 0.79, and the standard error is 0.79 ÷ √250, roughly 0.05, or 5 points. An approximate 95% confidence range is therefore about ±10 points: +12 could plausibly reflect a true value anywhere from about +2 to +22.

For support’s 40 responses, the same working gives a standard error near 12.8 points, so the range is around ±25: from roughly −45 to +5. Support looks worse than the company, but the evidence is not as strong as the −20 suggests, and the safest reading is “worth investigating”, not “proven bad”.

Two consequences. First, if next quarter’s score is +9, do not tell anyone engagement fell. A three-point movement is well inside the noise for a sample this size. Second, differences between small teams are mostly uninterpretable. Look for patterns that persist across quarters or show up in other data such as attrition and exit reasons.

Step 5: read the comments before the number

Add a follow-up: “What is the main reason for your score?” The detractors’ comments in support might cluster around shift patterns, or around a manager, or around tools. Read two hundred words of real comments and you learn more than from the score itself. The score tells you where to look. The words tell you what you are looking at.

What eNPS cannot do

A few cautions. Nobody can say what a “good” eNPS is across companies; published benchmarks depend on industry, country and method, so compare your own score with your own history. A 62.5% response rate is decent, but non-responders may differ from responders, and an unhappy group that has stopped believing in surveys will not show up in the data at all. The question measures willingness to recommend, which is related to engagement but not identical; people can be loyal and still not recommend a workplace, or satisfied and quietly job-hunting.

Use it as one signal, collected the same way every time, with a visible follow-up action. The calculation takes ten minutes. Deciding what to change takes the rest of the quarter.

Figures and names in the examples are illustrative.

Leave a Reply

Trending