NPS survey: asking one question in a way that produces action

An NPS survey asks customers how likely they are to recommend you on a scale of nought to ten, then subtracts the percentage scoring 0–6 (detractors) from the percentage scoring 9–10 (promoters). The result is a number between -100 and +100. It takes thirty seconds to answer, which is why it has spread so widely, and it produces a single figure, which is why it is so often misused.

Treated as a scoreboard it is close to useless for a small company: the sample is small, the number moves for reasons you cannot attribute, and comparing it to a published benchmark tells you nothing about your own customers. Treated as a way of generating conversations, it is one of the cheapest research tools available.

What the survey should contain

• The question itself, in the standard wording, with a 0–10 scale. Changing the wording makes your results incomparable with your own history, which is the only comparison that matters.
• One follow-up asking why they chose that number. This is the field that produces action; the score alone never does.
• Nothing else. Every additional question reduces the response rate, and the response rate is what makes the result meaningful.
• Context you already know — customer, plan, tenure, who they deal with — attached from your records rather than asked.
• A clear statement of what you will do with the answer, which improves both response rate and honesty.
• An identifiable respondent, unless you have a specific reason for anonymity. Anonymous NPS cannot be followed up, and following up is the point.
• Timing: after a meaningful interaction, not at a random moment or immediately after a support ticket, which measures the ticket rather than the relationship.

Read the comments, not the score. With a hundred responses your NPS has a margin of error wide enough to swallow most movements you will see, so quarter-on-quarter changes are usually noise. The comments are not noisy — thirty people telling you the same thing about onboarding is a finding regardless of what the number did.

Running it usefully

1. Decide the trigger: relationship NPS on a schedule, or transactional NPS after a specific event. Do not mix them in one dataset.
2. Survey a rolling sample rather than everyone at once, so you have a continuous read and nobody is asked too often.
3. Cap frequency at roughly twice a year per person. Surveying more often trains people to ignore you.
4. Send it from a person, not a system, and keep the message to two sentences.
5. Follow up on every detractor within a few days, by a human, asking what happened. This single practice recovers more customers than any product change you make from the data.
6. Follow up on promoters too — they are the population to ask for referrals, reviews and case studies, and almost nobody does it.
7. Tag the comments into a handful of themes and count them. The counts are what you act on.
8. Report the themes and the actions alongside the score, so the exercise is visibly worth its cost.

What NPS cannot tell you

It cannot tell you why the number moved without the comments, it cannot be meaningfully compared with a competitor's published figure — sampling and timing differ completely — and it is a poor early-warning system, because people often stop being satisfied long before they say so. If you need to know which accounts are at risk, usage and engagement data will tell you sooner than a survey will. NPS is a way of hearing what customers think in their own words, at scale, and that is a real thing to want; it is not a health metric.

In Ettex, the survey itself works as a form in Ettex Forms — a 0–10 scale question and one free-text follow-up, with submissions landing timestamped in a single searchable inbox, a live summary, and export to CSV or XLS when you want to score it. Because the inbox is shared, following up on detractors is a task anyone can pick up rather than something living in one person's mail. The score calculation and the trend belong in Ettex Sheets, the themes and the actions in Ettex Docs, the invitation goes out through Ettex Mail with reusable text and scheduled send, and the customer context sits in Ettex CRM.

Plainly: Ettex has no NPS or survey product. There is no score calculation, no in-app survey widget, no sampling engine, no benchmark data, no sentiment analysis of comments and no alerting when a detractor responds. It is a form, an inbox and a spreadsheet. For a company surveying hundreds of customers a year that is proportionate; large-scale experience programmes are a software category of their own.

Common mistakes

• Treating the score as a target, after which people start managing the score rather than the experience.
• Omitting the follow-up question, leaving a number with no explanation.
• Adding five more questions, which halves the response rate.
• Anonymous responses, which make following up impossible.
• Comparing your figure to a published industry benchmark collected in a completely different way.
• Surveying immediately after support tickets, which measures the ticket rather than the relationship.
• Detractors never contacted, which wastes the most valuable thing the survey produced.
• Reacting to quarter-on-quarter movements that are within the margin of error of a small sample.

NPS asks about the relationship; it does not explain a movement in it. A customer satisfaction score attached to a specific interaction tells you which moment went wrong, and a feedback form with one open question tells you why. Running all three is overkill; running only the first is common and rarely actionable.

A score is a prompt, not an answer. A feedback form with one open question usually explains a movement that the number only reports, and a customer satisfaction score tied to a specific interaction locates where the relationship went wrong.

A score prompts a question it cannot answer. A feedback form with one open question usually explains a movement, and customer success management is where the answer has to be acted on rather than reported.

An NPS survey is worth running if you read the comments and call the detractors. One question, one follow-up, a rolling sample twice a year — and report the themes, not just the number.