Net Promoter Score (NPS) can provoke strong opinions, in favour and against.
Critics will point to its volatility (and it's true that it has wider margins of error than other metrics), and ask whether the recommend question makes sense to customers and staff in all industries (it doesn't).
What I want to focus on in this post is a different criticism - that the cutpoints between Detractors, Passives, and Promoters are arbitrary.
In case you've been living in some kind of NPS-proof bunker, the premise of NPS is that customers are asked, on a 0-10 scale, how likely they would be to recommend you. Those who score 9 or 10 are deemed to be "Promoters" and those scoring 0-6 are "Detractors", and the difference between the percentages of Promoters and Detractors is your NPS.
Baked into that methodology is the idea that the relationship between the way customers score the recommend question and their future behaviour is non-linear. There is a step change in behaviour between 6 and 7, and another between 8 and 9.
This is not arbitrary. Fred Reichheld based the NPS methodology on a large dataset which showed exactly this step change in behaviour at those points in the scale.
So are the critics wrong? Yes and no. The cutpoints are not arbitrary, they're data-derived. But can we be sure they hold true for all customers, in all industries, in all countries, forever? Absolutely not.
Just as some can be too quick to dismiss NPS, advocates can be too uncritical in their acceptance of one standard approach. It's right to challenge the assumptions on which NPS is built, and to make sure they are sensible for your own market and customer base.
Specifically, you should use your own data to find out what the nature of the relationship is between the score customers give and their subsequent behaviour. If you do that you may well find, as some of our clients have, that entirely different cutpoints make sense in your situation.
Reichheld is not a prophet, and NPS is not graven in stone. Using your own data to revisit the basis for NPS is not inappropriate, far from it. I think it honours the spirit of NPS to base your use of it on what the data shows you will be most effective in terms of predicting behaviour.