पाठशाला Pathshala · ग्राहक Grāhak, The customer · Lesson 12 · Build

NPS done honestly: the number and the verbatims

Net Promoter Score is the most quoted customer number and the easiest to game. Collect it so nobody can, read it with its margin of error, and mine the comments for the three things to fix this quarter.

Pathshala, The Founder Library · 11 October 2026 · 7 min read

A classic mechanical balance scale seen close up. — Photograph: Paloma Gil · Pexels

Net Promoter Score is the customer number most often quoted to boards and investors and the one most often quietly managed. Collected honestly and read with its comments, it is a useful early warning. Collected carelessly, it measures how hard your team asked for a ten.

This lesson covers what the score is, how it gets gamed, a collection method that resists gaming, how much noise sits in a small company’s score, how to mine the comments, a worked example and a quarterly routine.

What the number is

Fred Reichheld of Bain introduced the idea in Harvard Business Review in December 2003 as the one number you need to grow, arguing from two years of research that, in most industries studied, the share of customers willing to recommend a company tracked differences in growth between competitors. The standard question, as Bain sets it out, is how likely are you to recommend us to a friend or colleague, on a scale from zero to ten. Nines and tens are promoters, sevens and eights passives, zero to six detractors. The score is the percentage of promoters minus the percentage of detractors, so it runs from −100 to +100.

The claim that it predicts growth better than anything else has not held up cleanly. Keiningham and colleagues tested it on seventeen Norwegian firms in five industries in a 2007 paper; as MeasuringU’s summary puts it, NPS was the best or second-best predictor of revenue growth in two of the five industries and did not clearly beat a satisfaction measure. The fair reading is that NPS is a reasonable loyalty measure among several, not a magic one. Its real value for a founder is the comment box beside it.

Why it gets gamed

Reichheld himself, writing with Darci Darnell and Maureen Burns in 2021, listed the abuses. Companies linked scores to frontline bonuses, and staff began to care more about the score than about the customer. They pleaded: “I’ll lose my job if you don’t rate me a 10.” They bribed: a free service for a ten. They manipulated the sample: never sending surveys to customers whose claim had been denied. And firms began to report scores to investors without saying who was surveyed, how many, at what response rate, or what triggered the survey. Their verdict was that some firms had turned NPS into a vanity statistic.

A polished metal service bell on a wooden counter. — A survey handed over at the counter with the person who served you watching measures courtesy more than loyalty. Photograph: Şahin Doğdu · Pexels

Every one of these has an Indian startup version. The delivery partner who says “sir, please give five stars” at the door. The support agent who sends the survey only after the ticket is resolved happily. The app that asks for a rating just after a refund has been credited. The deck that says NPS 72 with no denominator.

Collecting it so you cannot game it

Survey the relationship, not the moment. Send to a random sample of all active customers on a fixed calendar, quarterly for most businesses, rather than after a transaction someone on your team chose. Include customers who complained, asked for refunds or reduced their plan. Send from a neutral sender. The survey comes from the company, not from the person who served the customer, and the person who served them never sees an individual score with a name attached. Never pay on it. No individual’s salary or incentive moves with the score. Team-level review is fine; personal bonuses are not. Publish the method with the number. Every time the score is reported, report beside it the sample, the number sent, the response rate and the dates.

Ask one open question after the score and only one: what is the main reason for your score? Make it optional and keep the box large. In India offer the survey in the language the customer uses with you, not only in English, and accept comments in any script; some of the best verbatims arrive in Hinglish on WhatsApp.

Watch the response rate as closely as the score. When it falls, the customers who stop answering are usually the least engaged, and the score can rise for the wrong reason. Compare the people who answered with the sample on plan, city and tenure, as you would for [any survey](/library/surveys-that-do-not-lie-to-you), and say so in the report when they differ.

How much noise is in your score

Because NPS is a difference between two percentages, it is noisier than either. Jeff Sauro and Jim Lewis at MeasuringU show how to put a confidence interval on it, and their worked example is sobering: an NPS of 19 from 36 responses has a 90% interval running from about −2 to +38. A score from two hundred responses still carries a margin of around nine points either way.

The practical consequence: most quarter-on-quarter moves in a young company’s NPS are not news. A drop from 31 to 24 on two hundred responses may be nothing. Before anyone writes a plan to fix a falling score, check whether the fall is outside the interval. The figure below does the arithmetic.

Mining the verbatims

The comments are where the money is. Read every comment from detractors and passives, and a sample from promoters. Code each one exactly as you would an interview note, with a short label for what it is about: late technician, price rise, app crash on payment, rude call, refund delay. The method is in [From twenty interviews to one decision](/library/twenty-interviews-to-one-decision). Then count each code by group.

Three counts matter. The codes most common among detractors tell you what is driving people away. The codes most common among passives tell you what is stopping people who are nearly happy from recommending you, and passives are cheaper to convert than detractors. The codes most common among promoters tell you what to protect and what to say in marketing, in your customers’ own words.

Choose three fixes for the quarter: the biggest detractor code you can fix, the biggest passive code you can fix and one promoter strength to protect. Write each with an owner and the number of comments it came from. Close the loop with the customers who left those comments: a short message saying what you changed is the cheapest retention programme there is.

The score tells you whether to worry. The comments tell you what to fix. A company that reports the first and does not read the second has bought a thermometer and thrown away the diagnosis.

A worked example: a home-services app in Hyderabad

A home-services company in Hyderabad, plumbers and electricians booked by app, runs its quarterly survey to a random sample of customers who booked in the last ninety days, from a company sender, in English and Telugu. Two hundred and twenty answer: 46% promoters, 22% detractors, an NPS of 24. Last quarter it was 31. The operations head prepares a plan to retrain technicians.

The interval says wait. At two hundred and twenty responses the score is 24 give or take about nine, so 31 sits inside the range. The drop may be noise. The comments say something sharper. Forty-one detractors left a comment. Nineteen mention the technician arriving outside the booked slot with no call, ten mention the visiting charge being added when no repair was done, and only five mention the quality of work. Among passives the top code is the same late arrival, followed by parts sourced at a mark-up.

Technician skill was not the problem; the schedule was. The three fixes are an automatic call when a technician will miss the slot by more than twenty minutes, a waived visiting charge when no repair is possible, and protecting what promoters praised most, which was the technician explaining the fault before starting work. Next quarter the late-arrival code falls by half among detractors. The score rises too, but the company reports the code first.

The quarterly NPS routine

Send on a fixed date each quarter to a random sample of active customers, from a neutral sender, with the score question and one open question. Within a week, code every detractor and passive comment and a sample of promoter comments. Report the score with its interval, response rate and sample beside it, and the top five codes per group under it. Pick three fixes with owners. Message the customers whose comments produced them. Next quarter, check the codes before the score.

Keep NPS beside retention and revenue, not above them. Bain’s own answer to gaming was to pair the score with growth measured from audited revenue. A founder can do the same with a cohort table: if NPS rises and retention does not, trust retention.

The method here is one honest way to collect and read the score, not the only one. Whatever method you choose, publish it beside the number.