NPS survey best practices

The best practices for an NPS survey are short to state and easy to break. Keep the standard question and its eleven point scale, send it about the relationship on a fixed cadence rather than after every transaction, ask one open follow up and a permission to reply question, choose who receives it before you know how they feel, and report the score with its response count and spread for each segment. A four question survey that does all of that beats a twenty question survey that does half of it.

The number itself is well documented. Frederick Reichheld introduced it in the Harvard Business Review in December 2003, arguing that companies spent heavily on complex satisfaction tools while measuring the wrong thing. How the score is calculated, and why the wording is fixed, is set out on the NPS question page. This article is about everything around the question: the screen it sits on, the questions after it, when it goes out and what happens next.

Keep the exact question, and put it first

Change the wording and you lose the only thing the metric offers, which is comparison over time. The Ledgerly NPS survey asks how likely you are to recommend Ledgerly to a colleague, names the product rather than “us”, and puts it on the first screen with no welcome screen in front of it. It is a thirty second survey and says so.

Resist two common edits. The first is adding a softening clause such as “based on your recent experience”, which turns a relationship question into a question about one event. The second is putting demographic questions first. Qualtrics advises leaving them out when the answers already sit in a customer record, and on a survey this short every extra screen costs responses.

Show the scale as eleven choices

Bain’s explanation for the zero to ten scale is practical: a zero is never read as a good score, so people do not mix up the ends, and eleven points give enough room that a satisfied customer does not default to the top box. That only holds if the scale is on screen in full.

In the example, the score is a single choice field with eleven options laid out in three columns, and the field’s description, “0 means not at all likely, 10 means extremely likely”, sits under the question. Choice options can be set to one, two or three columns, and on a conversational form the description is shown to the respondent. Every option is a tap of the same size, and the export carries the chosen number as the answer, ready to count.

If you are deciding whether a satisfaction score would serve you better than a recommendation score, the difference is covered in NPS against CSAT.

Ask one follow up, and ask permission to reply

A score tells you direction. Nielsen Norman Group is blunt that quantitative metrics tell you how an experience is perceived but not why. The follow up is where the value sits, and one well worded open question beats a list of reason codes.

The Ledgerly form asks “What is the main reason for your score?” and leaves it optional. The customer feedback survey asks the same question and makes it required, which suits a six question survey whose whole purpose is the reason. Both are defensible. What is not defensible is a second, third and fourth open question, because Qualtrics notes that “why this score” and “how could we improve” often get the same answer.

Then ask whether a person may reply, as its own yes or no question, before asking for an email. Qualtrics recommends asking permission because not everyone wants to talk about their issue. The Ledgerly form labels its email field “Your email, if yes”, which tells the respondent the field is only for people who said yes. A longer bank of follow ups, including different wording for promoters, passives and detractors, is in NPS survey questions.

Timing: send it about the relationship

Qualtrics distinguishes relational surveys, sent quarterly or annually to understand how customers feel overall, from transactional surveys sent after a purchase or support call, and it recommends against using NPS on a transactional basis. Going too granular risks annoying customers by asking too soon or about too small an interaction.

So pick a cadence, send to the whole active base in the period, and hold it. The customer feedback survey is a useful contrast: it runs straight after a visit, so its satisfaction question is the one to trend visit by visit, and its recommend question is better read in aggregate across a quarter than as a verdict on one afternoon.

For a one off event, where there is no ongoing relationship to measure, a plain question works better. The event feedback form asks “Would you come next year?” instead, which is the decision the organisers actually face.

Segment before you report, and respect sample size

Bain describes tracking the score for each business, product, store or team, and for customer segments. That is where the number becomes useful, and it is also where it becomes fragile. Nielsen Norman Group warns that scores are rarely meaningful on small samples, and that binning answers into three groups discards information, so a group that moves from scores of 2 to scores of 5 has improved a lot while still counting entirely as detractors.

Three habits keep segments honest. Report the response count beside every score. Look at the spread of answers, not just the net figure. And set a floor below which a group is not reported on its own. The employee pulse survey records team as its only segment and never reports a team with fewer than five responses, a rule worth borrowing for customer segments too.

The export helps here. Each response is a row, the first column is the submission time, and a choice answer arrives as its label, so a spreadsheet can split a quarter by segment and count nines and tens without any reformatting.

The settings at a glance

The table at the end of this article lists twelve settings for a relationship NPS survey, what to choose for each and why. Most of them are small. Together they decide whether the number you report in March can be compared with the one you reported in December.

The row most teams skip is “who receives it”. Nielsen Norman Group describes a call centre agent mentioning the survey only after calls that went well, which produces sampling bias, and customers offered a coupon for a good rating, which inflates the score. Decide the send list from your records before any conversation happens, and never tie a reward to the answer.

Common mistakes

Tracking it after every interaction. The score then moves with whatever happened last, and nobody can trace why.

Adding questions. Every screen after the follow up costs responses, and detractors leave first.

Reporting the net figure alone. Two groups with the same score can have completely different spreads.

Letting staff choose who gets the link. It is the most common way the number gets gamed, usually without anyone meaning to.

Promising a reply and not sending one. A customer who said yes to a follow up and heard nothing will score lower next quarter, and they will be right to.

Build it

The conversational form guide walks through generating a form like the Ledgerly survey and editing it question by question. A conversational form suits NPS because each question gets its own screen and the description under the score is shown. Build the score as a single choice field with eleven options in three columns, the reason as long text, permission as a two option single choice, and the email as an optional field.

Of the 41 component types, you need four here. If you want a set of statements rated on one scale alongside the score, the rating matrix page shows how that field works in both formats. To close the loop, turn on the email alerts add-on so each response reaches the person who replies, or the webhooks add-on to send it to your own system; the webhook posts once per submission with no retry, so keep the export as your record. Finish with a thank-you screen that states when a reply will come. For the wording of the rest of any survey you add around the score, the rules in how to write survey questions apply.