What a 0-to-10 score is actually worth in a medical practice

Start with the mechanics, because most people who quote a Net Promoter Score have never seen them written down.

Fred Reichheld published The One Number You Need to Grow in the December 2003 issue of Harvard Business Review. The claim was that a single survey question predicted a company’s top-line growth better than the long satisfaction questionnaires firms were already paying for.^[1] Bain, where Reichheld is a fellow, still publishes the wording: “How likely is it that you would recommend [product, service or company/brand] to a friend or colleague?” Answers run 0 to 10. Bain classifies 9 and 10 as promoters, 7 and 8 as passives, 0 through 6 as detractors. The score is the percentage of promoters minus the percentage of detractors. It runs from −100 to +100.^[2]

Two things about that arithmetic are worth sitting with before you put the number on a wall. Passives are discarded entirely: a patient who gives you an 8 has no effect on your score in either direction. And the detractor band is enormous. A 6 and a 0 subtract identically. In practice this means the score is mostly a measure of how many people gave you a 9 or a 10, with a penalty attached to a bucket that lumps mild reservation in with genuine anger.

The honest version of the evidence

I would rather you heard the criticism from us than found it later and wondered what else we had left out. The academic record on NPS is mixed, and in places it is bad.

The most cited rebuttal is Keiningham and colleagues in the Journal of Marketing in 2007. They took longitudinal data from 21 firms and more than 15,500 interviews in the Norwegian Customer Satisfaction Barometer, replicated Reichheld’s analysis, and could not reproduce the claimed “clear superiority” of Net Promoter over other loyalty measures — and they tested it in the very industries Reichheld had held up as exemplars.^[3] That paper won the Marketing Science Institute / H. Paul Root Award for 2007, which is at least a sign the field thought the objection worth having.

Fifteen years later Baehre and colleagues revisited the question in the Journal of the Academy of Marketing Science, using seven US sportswear brands measured over five years. Their summary is the fairest I have read, so here it is in their words: the original premise of NPS is “reasonable”, the methodological concerns raised by academics are “valid”, and only the more recently developed brand health measure — measured across everyone who might buy, not just recent buyers — is effective at predicting future sales growth.^[4]

Healthcare has its own literature, and it is less kind. Krol and colleagues tested the NPS against three summary measures already in use, on Dutch hospital patient surveys covering 6,018 inpatient and 10,902 outpatient responses across six hospitals. The patient experiences reported in those surveys showed weaker associations with the NPS than with a plain global rating or an overall score computed from the experience items. Their verdict was that it is still unclear what the NPS specifically adds to a patient experience survey.^[5]

Then the systematic review. Adams, Walpola, Schembri and Harrison searched five databases for anything published between 2005 and September 2020 — worth knowing, because that is where the systematic evidence stops. Twelve studies met the inclusion criteria. Four identified real advantages: ease of use, high completion rates, and being well understood by a range of patients. The free-text comment box was viewed positively by patients and staff in four of the twelve. But four studies concluded that NPS adds minimal value to healthcare improvement, and three questioned whether the recommendation question is useful in healthcare at all — particularly where respondents are unable to select their service provider. The review’s overall conclusion is that NPS may not be sufficient as a stand-alone metric and is better used alongside a larger survey.^[6]

The criticism is setting-specific, and the setting matters

That last objection is not a complaint about how people answer. It is that in a great many healthcare settings the question describes a decision the patient never got to make. Ask an inpatient on a ward whether they would recommend the hospital their ambulance took them to, or a patient in a single-payer system with one regional service, and you are asking them to rate a choice that was made for them. Adams found exactly this in GP clinics: patients were frequently confused by the recommendation question, because they had limited or no say in which GP they saw.

The Adams review says so in its own conclusion: NPS may be better suited to healthcare settings where patients have a choice of provider. Krol’s data is hospital care. The three studies raising the relevance objection are pointing at the same structural fact.

Now hold that against an outpatient dental practice, a dermatology clinic, a medspa, a private family practice. The patient picked you off a map. They compared you to two or three others. They can leave next quarter without asking anyone’s permission, and a colleague at work will at some point say who do you see for that? The recommend question is not a hypothetical in that setting. It is a description of something the patient will actually be asked to do.

So the criticism that guts NPS for an inpatient ward is the same criticism that legitimises it for a private practice. That is a real argument and I will not stretch it further than it goes. Relevance is not validation. None of the work above demonstrates that NPS predicts growth in an outpatient medical practice, and I am not aware of anything that does. What the evidence supports is narrower and still useful: in a setting where patients choose, the recommend question is asking about the behaviour you care about rather than standing in for it.

Stop reporting one practice-wide score

This costs nothing but a change to how the report is built.

A single practice-wide NPS, reported quarter-to-date, is the least informative way to hold this number. It averages across providers with genuinely different chairside manner, across locations with different front desks, and across a window that resets to almost no data every ninety days. On the first Monday of a new quarter your headline number is built from whatever came back over the weekend, and by the last week of it the number is so heavily damped that nothing short of a catastrophe moves it.

Three changes:

1. Use a rolling 30-day window rather than quarter-to-date. The window should be the same length every time you look at it. Quarter-to-date is a different instrument in April than it is in June.
2. Split it per provider and per location. Compute it where the decision actually lives. A patient does not choose a practice-wide average; they choose a person, at a place, on a Tuesday. If one provider is carrying the score and another is dragging it, the blended number is built to hide exactly that. This is not only my preference: the Adams review concludes that the research does not support NPS for widespread benchmarking, and that it suits assessing localised performance. A practice-wide quarterly average is benchmarking. A per-provider rolling window is localised performance.
3. Set your operational alarm on the raw count of detractors in the last seven days, not on the score. The score is for the quarterly review. The count is what you act on this week.

That third one is the specific idea, and it is worth the arithmetic. NPS is a ratio. A ratio has no idea how many people are in it. If your response volume rises — more visits, a new provider, a survey that finally reaches everyone instead of the handful the front desk remembered to ask — then the number of unhappy patients per week can climb steeply while the score sits perfectly still.

Table 1. A worked arithmetic example, not data from any practice. Week 1: 36 promoters and 12 detractors out of 60 responses is 60% minus 20%, so +40. Week 8: 90 and 30 out of 150 is also 60% minus 20%, so +40. The score is identical. The number of people who had a bad time went from twelve a week to thirty.

The sharper version of the same trap is that the score can improve while the count rises. Give that Week 8 column 96 promoters, 24 passives and the same 30 detractors out of 150, and the score reads +44. Four points up on the dashboard, eighteen more unhappy patients a week walking around your city. Whoever presents that slide gets congratulated.

So put a threshold on the count and let it page someone. I do not have a study that gives you the right threshold and I am not going to invent one. My own rule of thumb is two alarms, not one. Fire the first when a single week’s detractor count passes twice the median of the trailing eight weeks; that catches a bad week. Fire the second when the last four weeks’ detractor total is up by half on the four weeks before it, even though no single week looked bad. That second one is what catches the drift in the table above, which a rolling median will quietly absorb as it re-baselines. Floor both at three in a week, and for most small practices the floor is what actually governs. Review them once a quarter. The point is not the number you pick. The point is that something in your practice notices a bad week while the week is still happening.

This is the part applaud is built around, so read it knowing I have an interest. Our survey goes to every patient who visited, not to a list someone picked, and the same check-in template goes to everyone. Robin, our agent, texts and calls in the practice’s own name and number, and the exchange is a conversation rather than a form — patients answer in their own words, and a model reads sentiment, intent, risk and themes out of what they said. The score in your panel is assembled from those answers.

The free-text comment box is the part of the Adams review I would point at. Patients use it heavily, and four of the twelve studies found it the most useful component of the instrument. Staff were harder to please: in one study only a tenth of GP practices had anything good to say about NPS, and several practitioners complained the comments lacked the detail to act on. That is a fair criticism of a free-text box, and it is the specific thing we built for. That matches what we see. The 0-to-10 tells you where to look. The sentence underneath tells you what happened. And the two instruments are different from your public reputation, which is a separate problem with separate numbers — we published ours in a report of its own. A private score and a public star rating should never be read as the same signal. The survey half of what we do is free. You pay when a review posts.

A single number is a poor summary of a practice

Here is what cuts against me hardest, and it is not the measurement theory. American patients are not free agents. Plenty arrive on a referral, inside a network, from a directory that listed three names within driving distance, and for them the recommend question sits closer to the ward version than I would like. The argument above is strongest where the patient paid attention to the choice — cash dentistry, dermatology, medspa, elective work — and it weakens the more the plan did the choosing. Adams draws the line at contexts that allow for the selection of treatment, naming primary care and elective surgery. That is roughly where I would draw it too.

And a single number is a poor summary of a medical practice. I think that is straightforwardly true, and we sell a product that produces one. Reichheld’s original claim was too strong, Keiningham was right to test it, Krol was right that a global rating summarises patient experience better, and Adams was right that this metric should not stand alone. If our panel ever becomes the thing a practice owner looks at instead of the sentences underneath it, we will have made that practice worse at listening, not better.

The number’s job is to route attention. It tells you which provider, which location and which week to go and read properly.

Sources

1. Reichheld FF. “The One Number You Need to Grow.” Harvard Business Review, December 2003. hbr.org/2003/12/the-one-number-you-need-to-grow. Introduces the recommend question as a single-item predictor of top-line growth.
2. Bain & Company. “Introducing: the Net Promoter System.” bain.com/insights/introducing-the-net-promoter-system-loyalty-insights. Source for the question wording, the promoter / passive / detractor bands (9–10, 7–8, 0–6) and the subtraction.
3. Keiningham TL, Cooil B, Andreassen TW, Aksoy L. “A Longitudinal Examination of Net Promoter and Firm Revenue Growth.” Journal of Marketing. 2007;71(3):39–51. doi:10.1509/jmkg.71.3.039. journals.sagepub.com/doi/10.1509/jmkg.71.3.039. 21 firms and 15,500-plus interviews from the Norwegian Customer Satisfaction Barometer; fails to replicate the “clear superiority” claim in the industries cited as exemplars. 2007 MSI / H. Paul Root Award.
4. Baehre S, O’Dwyer M, O’Malley L, Lee N. “The use of Net Promoter Score (NPS) to predict sales growth: insights from an empirical investigation.” Journal of the Academy of Marketing Science. 2022;50(1):67–84. doi:10.1007/s11747-021-00790-2. doi.org/10.1007/s11747-021-00790-2. Seven US sportswear brands over five years; only the brand-health measure of NPS predicted future sales growth.
5. Krol MW, de Boer D, Delnoij DM, Rademakers JJDJM. “The Net Promoter Score – an asset to patient experience surveys?” Health Expectations. 2015;18(6):3099–3109. doi:10.1111/hex.12297. onlinelibrary.wiley.com/doi/abs/10.1111/hex.12297. Six Dutch hospitals; 6,018 inpatient and 10,902 outpatient responses. Patient experiences correlated more weakly with the NPS than with a global rating or an overall score.
6. Adams C, Walpola R, Schembri AM, Harrison R. “The ultimate question? Evaluating the use of Net Promoter Score in healthcare: A systematic review.” Health Expectations. 2022;25(5):2328–2339. doi:10.1111/hex.13577. onlinelibrary.wiley.com/doi/full/10.1111/hex.13577 (open access at pmc.ncbi.nlm.nih.gov/articles/PMC9615049). Twelve studies met inclusion criteria. Four found minimal value for healthcare improvement; three questioned the recommendation question, particularly where respondents cannot select their provider; the conclusion notes NPS may suit settings where patients have a choice of provider.