What Is Net Promoter Score? Read Yours Properly Before You Act on It

Two things are true about this measure at the same time. It is the most widely adopted customer loyalty framework in existence, used by roughly two-thirds of the Fortune 1000. It is also the metric its own creator has publicly warned is being turned into a vanity statistic by the firms reporting it. Holding both […]
Predictive Dialer Vs Progressive Dialer

Two things are true about this measure at the same time. It is the most widely adopted customer loyalty framework in existence, used by roughly two-thirds of the Fortune 1000. It is also the metric its own creator has publicly warned is being turned into a vanity statistic by the firms reporting it.

Holding both ideas at once is, I think, the only sensible way to use it.

What Is Net Promoter Score?

Net Promoter Score, almost always shortened to NPS, measures customer loyalty using responses to a single question: on a zero to ten scale, how likely are you to recommend this company to a friend or colleague? Respondents are sorted into three groups, and the score equals the share of enthusiasts minus the share of critics. Results run from negative 100 to positive 100. Bain & Company, which developed the framework, groups ratings of nine and ten as promoters, seven and eight as passives, and six or below as detractors (Bain & Company).

Three characteristics distinguish it from other loyalty measures:

  • Single question. No battery of items, no weighting, no composite index.
  • Behavioral intent rather than reported feeling. Willingness to recommend carries social risk in a way that “were you satisfied?” does not.
  • A net calculation. Passives count toward the denominator but contribute nothing to the result, which is where most of the confusion starts.

Worth noting for anyone writing about it commercially: Net Promoter and NPS are registered trademarks co-owned by Bain & Company, Fred Reichheld, and a technology partner, so the terminology isn’t public domain.

How to Calculate Net Promoter Score

NPS = % promoters − % detractors

Percentages are calculated against total responses received, not against customers surveyed. Passive ratings dilute both percentages without adding to either side of the subtraction.

Promoters, Passives and Detractors Explained

Group Rating Given What Bain Says They Represent
Promoters 9 to 10 Customers pleased enough to recommend the brand to others
Passives 7 to 8 Customers who got what they paid for and nothing beyond it
Detractors 0 to 6 Customers disappointed enough to damage growth and reputation

That middle band surprises people. Somebody rating you eight out of ten sounds fairly happy, yet the framework treats them as carrying no loyalty value at all. Harsh, arguably. The reasoning is that a seven or eight signals a transaction completed rather than a relationship formed.

A Worked Example

Say 400 responses come back: 180 rate nine or ten, 140 rate seven or eight, and 80 rate six or lower.

  • Advocates: 180 ÷ 400 = 45%
  • Critics: 80 ÷ 400 = 20%
  • Result: 45 − 20 = NPS of 25

Simple arithmetic. The complications all live in what that number conceals.

One convention worth settling internally before anyone builds a dashboard: decide whether partial responses count. Somebody who opens the survey, gives a rating, then abandons the free-text field has supplied a valid data point, yet plenty of tools discard the whole record. Decide also whether to round. A result of 24.6 becomes 25 in most reporting, which is harmless until two teams round differently and spend a meeting arguing about a difference that doesn’t exist. Write both rules down once. Revisit them never.

Where the Framework Came From

Fred Reichheld, a Bain fellow, introduced the concept in a December 2003 Harvard Business Review article titled “The One Number You Need to Grow” (Harvard Business Review). His argument was that companies were spending heavily on elaborate satisfaction research while measuring the wrong thing, and that a single recommendation question predicted top-line growth better than the alternatives.

Adoption followed quickly. Bain points to eleven public firms highlighted in The Ultimate Question 2.0 whose median total shareholder return over the following decade ran five times the US median for comparable companies.

Impressive figures. Also, and this matters, figures selected and published by the framework’s owners. Which brings us to the part most explainer articles skip entirely.

Why did adoption move so fast? Partly because the framework solved a genuine reporting problem. Executives had spent years receiving satisfaction research they couldn’t act on, arriving as thick decks with dozens of weighted dimensions and no obvious priority. A single integer that fits in a board pack, moves visibly quarter to quarter, and requires no statistical training to read has enormous organizational appeal. That appeal is a strength and a weakness in equal measure: what makes it easy to communicate also makes it easy to misread, because the simplification that got it into the boardroom is the same simplification that hides the detail underneath.

Why the Same Score Describes Two Different Businesses

Here’s the structural weakness nobody mentions in the vendor glossaries. A single result can be produced by completely different customer bases.

Customer Mix Advocates Passives Critics Result
Company A 40% 45% 15% 25
Company B 55% 15% 30% 25
Company C 30% 65% 5% 25

Identical readings. Three businesses with almost nothing in common. Company B is polarizing: strong advocacy alongside a substantial group actively working against it. Company C has almost no critics but almost no enthusiasm either, which describes a utility more than a brand. Company A sits between them.

Reporting the headline figure alone throws away the distinction. My habit is to publish the three-way split next to the number, always, because the split is what tells you which problem you have.

What Is a Good Net Promoter Score?

Anything above zero technically means advocates outnumber critics. Common practice treats results above 50 as strong and above 70 as unusual. Those thresholds circulate widely without much underlying evidence, so treat them as convention rather than science.

Why Cross-Industry Comparison Breaks

Sector norms diverge enormously, and for reasons that have little to do with service quality. People rarely feel warmly toward a utility or an insurer regardless of how well either performs. Grocery, hospitality and consumer technology start from a friendlier baseline.

Response culture matters too. Some markets produce systematically higher ratings than others on identical scales, which makes multinational comparison a trap for anyone running a single global target. A regional team hitting 30 in one country and 45 in another might be delivering identical service.

Compare against your own trend line first, your direct competitors second, and published cross-sector tables somewhere a distant third.

Establishing that trend line takes longer than most teams expect. Four quarters gives you a shape; two gives you a guess. Until enough periods have accumulated to show normal seasonal variation, any target you set is essentially arbitrary, and setting an arbitrary target early is how programs acquire the pressure that later corrupts them. My preference is to run measurement for a year with no target attached at all, publishing results internally as observation rather than performance. Teams resist this, understandably, since an unmeasured year feels wasted. It isn’t. It’s the year that tells you what your baseline actually is.

Relational and Transactional Surveys Are Not the Same Thing

Two distinct implementations get reported under one name, which causes a lot of avoidable argument in quarterly reviews.

  • Relational surveys ask about the overall relationship, typically quarterly or twice yearly, sampled across the customer base.
  • Transactional surveys fire after a specific interaction: a purchase, a delivery, a support conversation.

When Each One Belongs

Relational measurement suits board reporting and year-over-year tracking. Transactional measurement suits operational improvement, because it attaches feedback to something you can actually locate and fix.

Mixing the two in one number produces nonsense. A relational figure of 40 and a post-interaction figure of 12 are not in conflict; they are answering different questions. Report them separately or don’t report them at all.

Sampling design differs between the two as well. Relationship surveys should draw a representative cross-section of your base, weighted to match it on whatever dimensions matter commercially: tenure, product held, spend band, region. Interaction surveys sample events instead of people, which means heavy users of your support line appear repeatedly and light users barely at all. Left uncorrected, your feedback ends up describing the experience of your most demanding customers rather than your typical one.

What NPS Actually Measures After a Support Call

This is where contact center teams get burned most often.

A survey triggered immediately after a conversation rarely measures loyalty toward the brand. It measures how that conversation went, coloured by whatever problem prompted the call in the first place. Somebody phoning about a billing error may rate you four out of ten while remaining a perfectly loyal customer who was simply annoyed on Tuesday afternoon.

Post-Interaction Surveys and Their Blind Spots

  • The problem contaminates the rating. Callers with expensive problems rate lower irrespective of agent performance.
  • Resolution timing distorts everything. Surveying before an issue is fully closed captures frustration rather than outcome.
  • Agents get judged on things outside their control. Product faults, pricing decisions and delivery failures all land in the same rating.
  • Channel changes the number. SMS, email and IVR-delivered surveys produce different results from identical experiences.

None of this makes post-call measurement useless. It makes it a measure of interaction quality that happens to use loyalty phrasing, which is a different thing from what the board thinks it’s reading. TabaTalk’s overview of inbound contact center metrics treats it accordingly.

The Academic Case Against the Metric

Reichheld’s central claim was superiority: that this question predicts growth better than satisfaction or retention measures. That claim has been tested independently, and it did not hold up.

A 2007 study published in the Journal of Marketing used longitudinal data covering 21 firms and more than 15,500 interviews from the Norwegian Customer Satisfaction Barometer, replicating the analysis method on industries Reichheld himself had cited as exemplars. The authors reported that they could not reproduce the finding of clear superiority over other measures, including the American Customer Satisfaction Index (Keiningham, Cooil, Andreassen and Aksoy, Journal of Marketing, 2007).

Nearly two decades later that paper remains the most cited counterweight in the literature, and no subsequent replication has settled the argument decisively in either direction. Defenders point out that the Norwegian dataset covers a small economy with unusual competitive dynamics. Critics respond that a claim advertised as universal ought to survive a test in Norway. Both positions have merit, which is roughly where the academic conversation has stalled since.

What follows from this? Not abandonment. The measure remains useful as a directional indicator and as a conversation starter with customers. What doesn’t follow is treating a movement of three points as evidence of anything, or presenting the figure to investors as a growth forecast.

What the Inventor Himself Says Goes Wrong

The most useful criticism comes from Bain rather than from academia, and it’s remarkably blunt.

Reichheld and his co-authors have written that self-reported figures and misapplication of the framework have created confusion and damaged its credibility. Their specific complaints, published in Harvard Business Review and reproduced on Bain’s site:

  1. Bonus linkage: Tying results to frontline pay made employees care about the rating rather than about serving people better.
  2. Reporting without method: Firms publish figures to investors with no explanation of how they were produced.
  3. Pleading: Staff asking customers directly for top marks because their job depends on it.
  4. Bribery: Offering something of value in exchange for a ten.
  5. Selective sampling: Not surveying customers whose claim was denied, for instance.
  6. Missing disclosure: No mention of who was surveyed, how many, what response rate resulted, or whether an independent party ran it under double-blind conditions.

Their conclusion was that some organizations had turned the measure into a vanity statistic. Bain’s response was to develop Earned Growth, an accounting-based companion drawn from audited revenue rather than survey samples, specifically because revenue data resists gaming, coaching and response bias in ways that surveys cannot.

Sit with the implication for a moment. The people who created the most widely used loyalty framework in business concluded that survey-based measurement alone had become unreliable enough to need an accounting backstop. That is a far stronger criticism than anything the academic literature has produced, and it comes from the one source nobody can accuse of commercial motive to undermine the metric.

Score Inflation and How It Starts

Nobody sets out to corrupt a measurement program. It happens gradually, and the sequence is fairly predictable: a target gets set, the target attaches to a bonus, agents begin mentioning the survey at the end of calls, the wording of that mention drifts from neutral toward suggestive, and eighteen months later the figure has climbed nine points while nothing about the service has changed.

I’ve watched that sequence play out more than once. The tell is usually a rising result alongside flat resolution rates and flat repeat-contact volume. Real improvement moves several numbers together.

Response Rates and Who Actually Answers

Survey results describe respondents, not customers. Those two populations differ systematically.

People with strong feelings answer at higher rates than the indifferent, which pushes both tails of your distribution outward. Customers who have already left rarely respond at all, removing exactly the population whose opinion would be most instructive. Response rates in the low single digits are common for email delivery, and a figure built on 2% of your base carries a margin of error most reporting quietly ignores.

Practical guidance: publish the response count and the response rate next to every figure you circulate. If somebody objects that this makes the reporting look weak, the reporting was already weak; you’ve just made it visible.

There’s a partial remedy, though it takes effort. Compare the profile of your respondents against the profile of your whole base on a few known attributes, tenure and product mix being the easiest, and check whether the two match. Where they diverge sharply, weight the results or state the skew openly in your commentary. Some organizations also run a small outbound telephone sample alongside their digital surveys, precisely to reach the population that ignores email entirely. Expensive, yes, and the phone sample often reads several points lower, which is uncomfortable and informative in roughly equal measure.

Running Loyalty Measurement in the UAE and Wider Gulf

Regional operations face a few wrinkles that global playbooks tend to skip over.

Language, Scales and Local Practice

Survey language changes results. A respondent answering in their second language interprets the middle of a scale differently than a native speaker, and translated instruments rarely preserve the exact connotation of “recommend to a friend or colleague.” For an operation serving Arabic, English, Hindi, Urdu and Tagalog speakers in the same queue, running one language version and treating the aggregate as clean data is optimistic at best.

Timing matters as well. Surveys landing during Ramadan, or across a weekend structure that differs from the one your survey tool assumes, produce response patterns that look like a service change but aren’t. Segment by language and by nationality where volumes allow, and expect the segments to differ by more than you’d like.

Business-to-business measurement in the region carries its own complication. Where a single account involves several stakeholders, whoever happens to answer determines the rating, and the procurement contact and the daily operational user frequently hold opposite views of the same supplier. Surveying one named contact per account and calling the output an account-level reading is common practice and fairly weak practice. Sampling multiple roles within larger accounts costs more and produces something you can defend when a client disputes their number, which in enterprise relationships tends to happen eventually.

What to Pair It With

The measure works as one input. Alone it invites overinterpretation.

Sensible companions:

  • Resolution rate, because unresolved problems drive ratings down independently of service quality
  • Repeat contact volume, which catches failures the survey misses entirely
  • Abandonment and answer speed, since people who never reached an agent never got surveyed
  • Quality assurance scoring, which measures what agents did rather than how customers felt about their week

TabaTalk’s guides to contact center reporting and quality assurance cover how these fit together in a single reporting layer.

The Follow-Up Question Doing the Real Work

Ask why. “What is the main reason for your rating?” attached to the numeric question produces the only genuinely actionable output of the entire exercise.

Free-text answers at volume are hard to read manually, which is where speech and text analytics earns its place: theme extraction across thousands of comments finds the three recurring problems that a spreadsheet of ratings never will. Honestly, if I had to choose between the numeric result and the verbatim comments, I’d take the comments without much hesitation.

How to Raise Your Score, Legitimately

Fixing the underlying experience is slow and unglamorous. It also works.

  1. Close the loop on critics within 48 hours: Somebody who rates you three and then receives a genuine call back frequently converts. Nothing else in this list has a comparable effect.
  2. Find the two or three drivers behind most low ratings: They cluster far more tightly than people expect.
  3. Reduce effort rather than adding delight: Removing friction outperforms surprise-and-delight programs in most measurement I’ve seen.
  4. Fix repeat contacts: Second and third calls about one issue produce disproportionately low ratings.
  5. Watch operational leading indicators: Real-time dashboards surface queue problems weeks before survey data reflects them.
  6. Keep the survey ask neutral: Scripted requests for high marks corrupt the instrument permanently.

Expect the timeline to be long. Experience changes reach the measurement layer slowly, because customers who formed an opinion last quarter carry it into this quarter’s responses regardless of what you fixed in between. Two full quarters before a genuine operational improvement shows up in the figures would not be unusual, and that lag creates its own hazard: teams abandon working initiatives at the ten-week mark because nothing moved. Hold the course longer than feels comfortable, and judge early progress on operational indicators instead, which respond within weeks rather than months.

Mistakes Worth Avoiding

  • Tying results to individual agent bonuses, which Bain identifies as the most damaging misuse of all
  • Comparing your figure against a cross-industry table and drawing conclusions from the gap
  • Reading movements smaller than your confidence interval as trends
  • Surveying the same customers repeatedly until they stop responding
  • Reporting a single organization-wide number when your product lines differ substantially
  • Treating passives as nearly-promoters, when the framework deliberately treats them as neutral
  • Changing survey wording, channel or timing mid-year, then comparing across the change as though nothing happened
  • Excluding customers who churned, which removes the most informative respondents you will ever have access to

Frequently Asked Questions

Is NPS a percentage?

No, though it’s frequently written as though it were. The result is a whole number between negative 100 and positive 100, produced by subtracting one percentage from another. Writing it as “NPS of 42%” is technically incorrect and causes confusion when non-specialists read it. Bain’s convention is a plain integer with no percentage sign. Keeping to that convention makes it easier to distinguish loyalty results from satisfaction percentages in the same report.

How many responses do you need for a reliable reading?

Reliability depends on your customer base size and how much variation exists within it. As a rough working figure, a few hundred responses gives a usable read for a mid-sized operation, while fewer than a hundred produces swings that reflect sampling noise rather than change. Calculate a confidence interval and publish it. Teams reacting to two-point movements on ninety responses are, statistically speaking, responding to randomness.

What is the difference between NPS and CSAT?

CSAT asks how satisfied someone was with a specific interaction, usually on a one-to-five scale, and reports the percentage giving positive answers. Loyalty measurement asks about willingness to recommend and reports a net figure. CSAT works better for evaluating individual interactions and agents; the recommendation question suits relationship-level tracking. Neither replaces the other, and organizations running both typically find they move independently more often than expected.

Can NPS be used for employee feedback?

Yes. Employee Net Promoter Score, or eNPS, applies the same arithmetic to a workplace recommendation question, usually phrased around recommending the organization as a place to work. It carries the same limitations plus one more: anonymity is harder to guarantee in small teams, which suppresses honest responses. Treat eNPS as a directional signal within a wider listening program rather than as a standalone measure of workplace health.

Does a negative score mean the business is failing?

Not necessarily. Several sectors run persistently negative averages, particularly regulated utilities, telecommunications and insurance, where customers interact mainly when something has gone wrong. Context determines the reading. A negative figure that improves steadily against a sector where competitors sit lower can indicate genuine progress. What warrants concern is a negative result paired with rising complaints, falling retention and worsening resolution rates.

How often should surveys be sent?

Relationship surveys typically run quarterly or twice yearly; interaction surveys fire after individual contacts, usually sampled rather than sent to everyone. Over-surveying produces fatigue, declining response rates and a sample increasingly skewed toward the small group willing to keep answering. Set a suppression rule so no individual receives more than one request in a defined window, commonly thirty to ninety days depending on interaction frequency.

What is Earned Growth and does it replace NPS?

Earned Growth is an accounting-based companion measure Bain introduced in 2021, calculated from audited revenue rather than survey samples. It tracks revenue from returning customers plus revenue from new customers who arrived through referral. Bain positions it as reinforcing rather than replacing the survey measure, specifically because revenue figures resist the gaming and response bias that affect self-reported ratings. Most organizations lack the customer-level revenue tracking it requires.

Should NPS be reported to investors?

Only with methodology attached. Bain has criticized firms that publish results with no disclosure of sample size, response rate, survey trigger or whether independent double-blind research produced them. Without those details the figure carries no comparability and invites the accusation that it was selected favorably. If you report externally, report the method alongside it, or expect analysts who understand the measure to discount whatever you’ve published.

Talk to the TabaTalk Team

Loyalty measurement only pays off when it connects to something operational: the queues, the resolution rates, the conversations that produced the ratings in the first place.

TabaTalk provides cloud contact center software built for Gulf operations, covering routing, real-time reporting, speech analytics and multilingual survey delivery, with deployment measured in hours. Contact our sales team to review your inbound setup, or ask us to show how post-interaction feedback links back to individual conversations and the themes running through them.

Read More:

30 Oct 2026
One of these concerns what your number says. The other concerns what it does. They get compared constantly, and comparing them is a little like choosing between a memorable shop name and a working front door. This guide covers both properly, including the part almost every article on the subject skips: the mechanism behind vanity […]
29 Oct 2026
Every verification code you have ever received, every delivery notification, every “your driver is arriving” message: something sent those, and it was almost certainly a CPaaS platform sitting behind somebody else’s application. The category is invisible by design, which makes it easy to buy for the wrong reasons. This guide covers what it is, what […]
28 Oct 2026
Changing your contact center software usually means changing your phone numbers, your carrier relationship and your invoices all at once. BYOC is the arrangement that lets you change one without touching the others. For a business in the Gulf, that separation is worth more than it is almost anywhere else, for reasons that have nothing […]

Smarter conversations,
straight to your inbox.

Subscribe for updates on features, trends, and stories shaping the future of customer connection.