NPS and CSAT: calculating scores and acting on complaints

Saltanat, the owner of the cosmetics shop from the loyalty programme lesson, opens a report: NPS (Net Promoter Score, a measure of willingness to recommend) is +15. She suggests giving you a target of +30. Meanwhile, in the comments on the same survey, customers write that they could not spend their promised points at the till. You could ask people for higher scores, but the till would keep refusing redemptions. The real job is different: find the step where the promise and the purchase came apart, and fix it. The numbers in this lesson are illustrative.

NPS and CSAT (Customer Satisfaction Score) collect feedback in the same format every time, so you can compare one month with the next. Read them together with comments, support requests and purchase history. First agree what experience people are rating, then count the answers correctly and hand the problem you found to someone who can fix it.

Two questions for different jobs

NPS asks how likely someone is to recommend a company, product or service to a friend or colleague, on a scale from 0 to 10. In Saltanat's shop the question is: "How likely are you to recommend our shop to someone you know?" Keep the object of the rating the same from survey to survey. An answer about the whole shop and one about the latest delivery do not belong in the same series: people were rating different things.

CSAT asks whether someone is satisfied with one episode: a purchase, a delivery, a reply from support. After a query about points, the question would be: "How satisfied are you with the way we resolved your points redemption query?" Ask it once the customer can see the outcome. A survey straight after an automated support reply captures the first minute of the conversation, before anything has been resolved.

CES measures effort: how much searching, repeating and asking for help it took. If you decide to collect it, describe its question and scale as carefully as for NPS and CSAT. A table of NPS answers gives you neither CSAT nor CES: people answered a different question.

NPS: passives stay in the calculation

Bain's NPS method divides answers into three groups: 9 and 10 are promoters, 7 and 8 are passives, and 0 to 6 are detractors. Subtract the share of detractors from the share of promoters.

The shop invited 100 customers to the survey and 20 responded:

NPS = 8 / 20 × 100 - 5 / 20 × 100 = +15.

The seven passive responses fall into neither share, but they sit in the denominator of both. If you take them out of the denominator, you get (8 - 5) / 13 × 100 ≈ +23.1. Not a single customer changed her mind; the score was simply calculated wrong.

Report NPS as a number from -100 to +100, without a percent sign. Here +15 is the difference between 40% and 25%, and the promoter share is 40%. The groups are for counting; when you talk to a person, their explanation matters more. A 6 may hide an awkward payment option, and a 9 may arrive with a serious complaint about delivery.

With 20 responses, one person moving from detractor to promoter shifts NPS by 10 points at once: one share rises by 5 points and the other falls by 5. So if next month you see +25 instead of +15, reread the responses first and only then credit the team with the increase.

CSAT: choose the scale before calculating the percentage

For points queries, the shop chose a scale from 1 to 5: 1 is very dissatisfied and 5 is very satisfied. CSAT counts 4 and 5 as positive, as in the version described by Qualtrics. Write the rule down before you start collecting responses and keep it for the next surveys, or the months will stop being comparable.

After support queries, 20 people answered and 15 of them chose 4 or 5. CSAT = 15 / 20 × 100% = 75%. In the report, the question, the scale and the number of respondents sit next to the percentage. You can also calculate the average score, but label its column differently, for example "average score", so nobody mistakes it for CSAT.

These 20 responses come from a different survey, run after support queries. They were not taken from the NPS table, and the 75% does not follow from NPS +15. The same person can rate the consultant who helped her highly and still not want to recommend a shop where she needed that help in the first place.

One answer out of 20 moves CSAT by 5 percentage points if it crosses the boundary between 3 and 4. So write "75%, 15 out of 20": that shows at once how many people the figure rests on.

Who reached the survey

Of the 100 customers invited, 20 responded, a response rate of 20%. The other 80 stayed silent and have no score. Giving them a 7 or 8 and filing them as passives would mean inventing data.

Who was invited matters too. If the manager sent the link only to people who praised them in chat, you have collected the opinions of a preselected group. If the link sits only in the customer account on the website, people who order through a manager on WhatsApp may never see it. The report should say who was invited and how.

Before the first survey, write down:

• who receives the question and which orders qualify;
• the delay after a purchase or support query;
• the channel and exact wording;
• how many times a person may respond during the period;
• when collection ends and who reviews the comments.

Keep these conditions stable if you want to compare months. If you switch from email to a phone call or change the wording, start a new series or mark the break in the report. Regular customers and newcomers can be shown separately, with the size of each group. Then you will notice when the overall score moved simply because different people answered this time.

A manager's bonus for 10s gives them a reason to coach customers towards the right answer. It is better to judge feedback work by what got done: the issue reached its owner, the owner found the cause, the customer received an outcome. And help should not depend on the score: the problem gets solved for someone who no longer wants to fill anything in, too.

Turn a comment into a task

After the numerical question, one open question is enough: "What most influenced your rating?" Keep the answers word for word and put the topic in a separate field. "They would not let me spend my points" may mean a till error, a product exclusion or an unclear promise in an ad, and one sentence will not tell you which.

The manager contacts people who left their details and agreed to discuss their feedback, and clarifies the order and what they expect. Then they look at the rules, the balance and the transaction log. An anonymous answer is not wasted either: you cannot write back to its author, but you can still fix the process.

Once the problem is solved, the manager tells the customer what was done. A closed task records the outcome: "Limit corrected, order with points redemption went through." "Passed to the developer" is only a handover; the work has not finished there.

Take recurring complaints to a meeting with whoever owns the process. Count them per purchase or per support query: ten complaints in a sale week and ten in a quiet week call for different investigations. Customer interviews fill in the details of the experience, and repeat orders show whether people actually came back.

Discussing the cost with the owner

Suppose fixing redemption and adding hints to the basket costs 12,000 ₸. A 10,000 ₸ order, after 500 points redeemed and 6,500 ₸ in variable costs, leaves the shop 3,000 ₸. So covering 12,000 ₸ takes 12,000 / 3,000 = 4 extra orders with the same economics.

That threshold is a good thing to take to Saltanat, but it is not an NPS forecast. Five low scores do not turn into four future orders: some people would have bought again without the fix, and some will not return for entirely different reasons. Multiplying an NPS gain in points by average order value makes no sense either. To estimate additional purchases, you need a comparison planned in advance.

A fault in a redemption the shop has already promised gets fixed regardless of how many people complained about it. The estimate helps choose the repair and explain its cost. Scores after the fix, comments and repeat support requests show how the experience changed; orders and contribution show what happened to the money.

Exercise: a report that produces work

Take one survey from your project. If your client has no surveys, draft a question and a collection plan, and practise the calculation on the numbers from Saltanat's shop.

0 of 6 done

• Record the question, scale, audience, invitation timing and collection period.
• State the numbers invited and responding, plus the response rate.
• Recalculate NPS or CSAT from the original scores, keeping every eligible response.
• Choose one recurring topic and attach anonymised quotes and the data needed to investigate it.
• Assign an owner, a deadline and the result that will close the task.
• Keep the experience rating, the process fix and money expectations apart in the conclusion.

The shop's report will show NPS +15 from 20 responses to 100 invitations. If a colleague came back with +23.1, they dropped the seven passive responses. CSAT 75% belongs to a different survey, run after support, and is labelled as 15 positive responses out of 20.

The work this report produces is to reproduce the redemption failure and assign someone to fix it. "Raise NPS to +30" names neither the fault nor an owner, so on Monday there is nothing to do with it. The exercise is done when it describes a change the customer will notice at the till and a date when the team will come back to the result.