Net Promoter Score
The most widely adopted customer metric in business, and the gap between how confidently it is used and how well it is evidenced.
NPS is the most successful customer metric ever published. It came out of a 2003 Harvard Business Review article, “The One Number You Need to Grow”, proposing a single question — how likely are you to recommend — converted into a score from -100 to +100 by subtracting the percentage of detractors from the percentage of promoters. By the 2010s roughly two thirds of Fortune 1000 companies had adopted it.
Its success is easy to explain. It is cheap, it produces one number, and one number can go in a board pack and on a bonus scheme. Those are real advantages and they are not methodological ones.
What the evidence supports
Word of mouth matters. Few researchers dispute that recommendation behaviour is associated with growth, and a metric that puts advocacy in front of management has done something useful.
What the evidence does not support
The strong claim — that this single measure accurately predicts future growth and is the one number a company needs — is where it comes apart.
The foundational research was never fully published. The pivotal studies were not set out in full detail, were not peer reviewed, and the underlying data was not made public. That is an unusual basis for a metric embedded in executive compensation.
Single items are less reliable than composites. This is basic measurement theory and it is not controversial: an index built from several items is more stable than any one item. NPS deliberately trades that stability for simplicity.
The three-box transformation discards information. Collapsing an eleven-point scale into promoters, passives and detractors throws away most of the variance, and does so asymmetrically. Two organisations with visibly different response distributions can return the same score.
The scoring is volatile in a misleading way. Because the middle of the scale is discarded, small movements across the 6/7 and 8/9 boundaries produce large swings in the headline number that look like change and are frequently noise.
Using it without being misled by it
Keep the question — it is a reasonable advocacy item and the benchmarking is genuinely useful. Then do three things the standard implementation does not.
Report the distribution, not only the net score. Analyse the verbatim follow-up, which is where the actionable content lives. And validate it locally: check, in your own data, whether the score actually relates to retention and revenue in your category. Where it does, use it with confidence. Where it does not, you have learned something considerably more valuable than the score.
Above all, do not let a single tracked number displace the question that the retention literature has always pointed at: why are the customers who left leaving?
