Net Promoter Score comes from a single survey question: on a scale of 0 to 10, how likely are you to recommend this company to a friend or colleague? It was introduced by Fred Reichheld of Bain & Company in a 2003 Harvard Business Review article, and its appeal was that one question and one number could stand in for long satisfaction surveys. Two decades later it appears in most CX contracts, board decks and client reviews. It is worth understanding exactly what the number does and does not capture.

The calculation

Responses are put into three groups:

  • Promoters: scores of 9 or 10.
  • Passives: scores of 7 or 8.
  • Detractors: scores of 0 to 6.

NPS is the percentage of promoters minus the percentage of detractors. Passives count in the total but in neither group. The result runs from −100 (everyone a detractor) to +100 (everyone a promoter), and it is written as a plain number, not a percentage.

Say a telecom helpline gets 400 survey responses in a month: 180 promoters, 140 passives and 80 detractors. Promoters are 45%, detractors 20%, so NPS is 45 − 20 = 25.

What the subtraction hides

Two very different customer bases can share a score. A business with 30% promoters and 10% detractors has an NPS of 20. So does one with 60% promoters and 40% detractors. The first is mostly lukewarm; the second is polarised, perhaps delighting one customer segment and failing another badly. Always look at the three percentages, not only the net.

The cut-offs are steep. A 6 counts as a detractor, the same as a 0. In some cultures and markets people rarely give 9s and 10s to anyone, so a company can be well liked and still show a modest NPS. Comparing NPS across countries, or between an Indian domestic process and a US process, is risky for this reason.

Moving people between buckets has uneven effects. In the 400-response example, converting 40 passives into promoters lifts NPS from 25 to 35. Converting 40 detractors into passives also lifts it to 35. But moving a customer from a 3 to a 6, which might reflect a real improvement in how angry they are, does not change NPS at all. The metric is blind to movement within a bucket.

Sample size is where most NPS reporting goes wrong

Because NPS combines two proportions, it is noisier than a single percentage. For the example above, a standard approximation puts the 95% margin of error at about ±7.5 points with 400 responses. So a month-on-month move from 25 to 29 is well within normal variation and not evidence of improvement.

At team or agent level it gets worse. With 50 responses and the same mix, the margin is around ±21 points. An agent whose NPS falls from 30 to 10 between two months may have changed nothing at all. Ranking agents on monthly NPS, or tying incentives to it, often rewards luck. If NPS is used for individuals at all, it needs longer windows, minimum response counts, and should sit alongside quality monitoring rather than replace it.

Relationship NPS and transactional NPS

The original NPS question asks about the company as a whole, and is usually sent periodically to a sample of customers. That is relationship NPS. It reflects price, product, network quality, billing, brand and every past interaction, so a contact centre controls only part of it.

Transactional NPS asks the same question straight after an interaction such as a support call. It is closer to what the operation controls, but customers still answer based on the whole relationship. A customer whose broadband has been down for three days may give a 2 to a perfectly handled call. This is why many CX teams pair transactional NPS with CSAT (satisfaction with this interaction) and a resolution question, and why BPO contracts that put NPS in the SLA usually need a clear agreement on which drivers the vendor owns.

Getting value from it

The score matters less than the follow-up question, usually “What is the main reason for your score?” Coded verbatims from detractors are where the operational insight lives. A useful practice is to tag detractor comments into drivers such as “issue not resolved”, “had to call multiple times”, “agent attitude”, “policy or fee” and “product fault”, then separate what the operation can fix from what the client must fix.

Some programmes also run a closed-loop process: a supervisor calls back a sample of detractors within a day or two to understand and, where possible, repair the issue. This turns the survey from a measurement exercise into a recovery mechanism, and the callbacks often reveal process faults that call recordings alone miss.

Things that quietly distort it

  • Survey selection. If only resolved calls get a survey, or agents can choose who to transfer to the survey, the score is inflated.
  • Score begging. “Please give me a 10” at the end of a call raises scores without raising loyalty.
  • Low response rates. Very happy and very unhappy customers respond more often, so the people who answer may not represent the rest.
  • Changing the question or channel. Moving from an IVR survey to SMS, or changing the wording, can shift scores. Treat such changes as a break in the trend line.

NPS is a reasonable headline for the direction of customer sentiment over quarters, on decent sample sizes. It is a poor instrument for judging an individual agent in a single month. Knowing which of those two jobs you are asking it to do will save a lot of misplaced celebration and blame.

Written by the RP Strativa Team. Survey counts in the examples are illustrative.

Leave a Reply

Trending