NPS: how to calculate, interpret, and investigate what's behind it

NPS summarizes responses about the willingness to recommend a company or product. It can help track public perception, but an isolated number doesn't explain the reason for the evaluation nor directly point to which screen needs to change.

To use the metric effectively, preserve three elements: who responded, when they responded, and what experience they were evaluating. Without this context, comparing periods can hide changes in the audience or data collection.

How to calculate NPS

Use a 0 to 10 scale: scores 9 and 10 form the group of promoters; 7 and 8, the passives; 0 to 6, the detractors.

The calculation is percentage of promoters minus percentage of detractors. Passive respondents are included in the total used to calculate the percentages, although they are not directly added or subtracted.

In a hypothetical example with 50 responses, there are 25 promoters, 15 passives, and 10 detractors. The percentages are 50% and 20%, respectively. The NPS is 30 points. This does not mean that the average score was 3, nor that 30% of customers are satisfied.

Define the evaluated experience

A person might recommend the company and have encountered difficulty in the last task. They might also complete the task easily and not recommend the product for another reason.

Write the question consistently and identify the object being evaluated. If the study concerns an incomplete prototype, assess whether asking for a recommendation makes sense at that stage.

To investigate a specific step, it might be more appropriate to measure satisfaction or ease after the task. Choose the metric based on the question you need to answer, avoiding using NPS just because it already appears in other reports.

Preserve the distribution of scores

In UXTap, scales configured as NPS feed the metric's reading and the distribution of responses. Consult the histogram alongside the aggregated value.

Two groups can have the same result with different distributions. One might concentrate responses at the extremes; another might gather many intermediate scores. The research action does not need to be the same for both.

Also check the base. A small sample can change significantly with a few additional responses. Show the number of participants and avoid presenting small fluctuations as a consolidated trend.

Ask for the reason without leading the answer

After the score, offer an open-ended question, such as “what experience most influenced your evaluation?”. The goal is to identify episodes and expectations, not to defend the company.

Imagine management software where users assign low scores because they need to redo an import. The next investigation can observe this task and check which steps generate rework.

Leave room for other reasons. If you only ask about import, you might overlook support issues, billing clarity, or suitability for the job.

Compare periods with visible conditions

Record invitation channel, journey moment, profile, and period. A round with old customers and another with newly registered ones does not offer the same basis for comparison.

In UXTap, use the available filters and segments to explore coherent subsets. Treat differences found after data collection as clues, especially when there are many groups and few responses in each.

Also check which responses are included in the distributions. Changing the inclusion of partial sessions modifies the base and should be communicated.

Perform the calculation preserving passives in the denominator

A common mistake is to remove scores 7 and 8 from the total before calculating proportions. In the previous example, this would reduce the base from 50 to 35 responses and change the result. Passives are not included in the difference between groups, but they still represent people who answered the question.

Another confusion is using the average of scores as NPS. An average can be an additional description, but it corresponds to a different calculation. To allow verification, present the valid total, the number of promoters, passives, and detractors, in addition to the final value in points. The reader should be able to redo the calculation without guessing which responses were considered.

If you have different units or segments, preserve the composition of each. A simple average of NPS from groups of different sizes does not necessarily represent the combined result. To obtain the total, revisit the counts of promoters and detractors and the valid base of all included groups.

Separate relationship evaluation and episode evaluation

Imagine a fictional company that measures recommendation after resolving a difficult support ticket. This collection can help investigate that moment, but it does not automatically represent the relationship of all customers with the company. People without tickets, people who gave up contact, and inactive accounts may not appear.

Define who receives the invitation, in what situation, and after how long. If one round invites all active users and another only reaches those who completed a purchase, the change in audience can influence the result. Record these conditions in the history instead of showing an unexplained trend line.

In enterprise products, consider who responds. The person who buys the tool, the person who administers it, and the person who performs daily tasks may evaluate different experiences. Do not transform one person's response into an automatic representation of the entire account. The recruitment planning should reflect the question you intend to answer.

Use reasons to open specific investigations

After the score, a question like “what was the main reason for your evaluation?” allows gathering an explanation without anticipating a list of problems. Preserve responses that mention price, service, reliability, and interface as different subjects. A drop in recommendation does not necessarily mean that the menu worsened.

Also look for favorable reasons. They help understand what the audience values and what a revision should not compromise. If customers praise the predictability of the process, a change that reduces steps but hides confirmations could remove something important for that experience.

Choose a topic with sufficient evidence and turn it into an observable question. If the recurring reason is difficulty tracking a request, plan a status inquiry task. The next round can investigate whether the information is found and understood, without expecting NPS alone to explain the solution.

Turn the score into an investigation agenda

Separate reasons by theme and select a problem that can be directly observed. Relate the metric's result to usage evidence before prioritizing a solution.

Your next step is to set up an NPS question accompanied by an open-ended one in UXTap, define the audience, and record the collection moment. Then, choose a recurring reason and turn it into a research task. It is this continuity that gives utility to the measurement.

The reasons behind the scores deserve their own interpretation: see how to analyze open-ended responses with AI and consult the results resources to combine distribution and reports.

Frequently asked questions about NPS

What is a good NPS for my product?

There is no universal goal that dispenses with context. Compare equivalent conditions and examine the experience of the evaluated audience. External references need to have population, period, and method identified. A number above a reference does not eliminate important problems; a value below does not, by itself, explain where to act.

Can I combine responses from several months?

You can build an accumulated result if you make the period and rule explicit, using the counts and the total base. Do not present it as if it were the evaluation of the most recent month. Changes in product, recruitment, or audience composition can be hidden in the aggregate.

Does a low score allow identifying the cause of cancellation?

Not alone. The score records an evaluation, and the justification can indicate topics for investigation. Verifying cancellation requires information about that decision and its context. Avoid asserting a causal relationship based on a coincidence between an unfavorable evaluation and subsequent behavior.

Create your UXTap account to prepare your first study.