9–10 · Promoters
They would actually recommend you. Not “satisfied” — willing to put their own reputation behind you.
The method, honestly
What is NPS, and why does it work?
The arithmetic takes one sentence. Convincing your own company that the number means something takes considerably longer, and nobody writes that part down. This page is that part.
The whole method in four lines
Ask one question. Subtract the unhappy from the delighted.
You ask “how likely are you to recommend us?” on a scale of 0 to 10. Then you sort the answers into three groups, and the score is the percentage who love you minus the percentage who don’t.
7–8 · Passive
These count for nothing. Not a mistake — the point. A 7 is somebody who will leave for a better price and not tell you why.
0–6 · Detractors
They will actively cost you business. A 6 is not “a bit above half”. It is a complaint politely phrased.
So a score of zero is not failure
If half your customers say 10 and half say 0, you score zero. If everybody says 8, you also score zero. Those are wildly different businesses and the number cannot tell them apart — which is the first honest thing anyone should say about NPS, and the reason the comments matter more than the score.
The question that gets asked in every meeting
“Why is a 7 a bad score?”
Because the scale is not measuring satisfaction. It is measuring whether somebody will stake their own name on you, and almost nobody does that at 7.
When did you last recommend a supplier you would score a 7?
Nobody has. A 7 means “they were fine”, and fine is what you say about a business you would replace tomorrow if somebody was cheaper. The alternative to understanding this is a company that celebrates an average of 7.8 while quietly losing customers.
What it cannot do
Four honest limits.
Any page selling you NPS without this section is selling you something.
The four limits, in detail
It is one number for a complicated thing
It compresses everything — product, price, delivery, the person who answered the phone — into a single figure. That is useful for spotting a trend and useless for diagnosing it. The comment does the diagnosing.
It moves on tiny samples
At twelve responses a single answer swings the score by eight points. Most NPS “improvements” that get reported to a board are noise, and a tool that draws you a confident line through four data points is lying to you.
Benchmarks are mostly worthless
Published industry averages come from different populations, asked at different moments, with different response rates. Your score against your own score last quarter is the only comparison worth making.
Who answers is not who you asked
The people who reply are disproportionately the delighted and the furious. That is fine — those are the two groups worth hearing — but it means the score is not a survey of your customer base.
The part that actually works
The number is not the point. The phone call is.
Here is the finding that survives every criticism above: businesses that contact unhappy respondents keep more of them. Not because the score went up — because somebody rang a customer who was about to leave, and found out why, and fixed it.
That is called closing the loop, and it is the entire mechanism. Everything else — the scale, the arithmetic, the trend line — exists to find the person worth calling. A programme that produces a monthly number nobody acts on has skipped the working part and kept the reporting.
Which is a testable claim, and the test is uncomfortable: if you stopped calculating the score but kept calling every unhappy customer within a day, you would keep most of the benefit. If you kept the score and stopped calling, you would keep almost none.
What that means for how it should be built
The person waiting for a call goes above the score. A comment matters more than a rating. Ask after something actually happened, so the answer is specific enough to fix. And an unhappy answer should arrive as a task with a name on it, not a row in a report. Those are the decisions Avota is built on — they come out of the method, not a design meeting.
The section nobody writes
How to introduce it to a company that thinks it’s a marketing survey.
This is the hard part. Not the method — the internal argument. Some things that help:
Never present it as a grade
The moment a rep believes the number is about them, you get coaching-to-the-survey and a useless dataset. It is a measure of the experience, and the first score is a baseline, not a verdict.
Lead with a real comment, not the number
“We scored 41” starts an argument about methodology. “A customer wrote: shipped short twice and nobody called me” starts a different conversation entirely, and it is the one you wanted.
Give the follow-up to somebody who can fix things
If unhappy answers land with the person who owns the relationship, the programme survives. If they land in marketing, it becomes a report, and reports get cancelled in the second budget round.
Promise a small frequency and keep it
“No customer hears from us more than twice a year” is a promise that buys you internal support, and breaking it once costs more trust than the whole programme earns in a quarter. Make the tool enforce it rather than trusting a rule.
And the first result will be bad
Plan for it out loud, before you start. A first score arrives lower than anyone expects, because the people who answer first are the ones with something to say. If nobody has been warned, the programme spends its first month defending itself instead of calling customers back. If everyone has, the same number is a starting line.
If you want to try it
Two ways in, and neither takes a project.
- A counter, a van, a waiting room: print a code, ask one question, get a name attached to every problem. Nothing to connect.
- A distributor: ask after an order closes or a quote is lost, triggered from your own system, read-only.
- Either way the limits are enforced by the software rather than by a rule somebody remembers.