A quarterly business review shows Net Promoter Score down four points in one region. The room spends twenty minutes debating whether the drop is significant, whether the survey sample changed and whether a competitor campaign is to blame. Nobody leaves with a decision about what to fix, because the score cannot tell them.
This is the first stall point in the VoC execution gap described in the pillar article of this series. Scores are good at signalling that experience has moved. They are poor at explaining why, and they say nothing about who owns the cause.
NPS, CSAT and CES are outcome indicators. They are useful for tracking direction and comparing units over time, but they are not diagnostic instruments. To decide what to fix, organizations need to pair every headline score with three things: the drivers behind it, the verbatim and operational context that explains those drivers, and a link from each driver to the team that controls it. Programs that treat the score as the product of the VoC program, rather than as its starting point, tend to generate debate instead of change.
It helps to separate two jobs a CX metric can do:
McKinsey's survey of more than 260 US CX leaders illustrates the gap: only 16% said surveys gave them data granular enough to address the root causes of CX performance (McKinsey, February 2021). Forrester's 2025 survey of feedback and CX measurement practitioners similarly found that root-cause analysis is rare in practice (Forrester, August 2025).
Forrester's 2026 predictions go further, warning that some CX teams will fall into a "death spiral" by feeding metrics obsession, because "narrative- and context-poor dashboards won't reveal which problems to solve, how to solve them, or why they matter to the business" (Forrester, October 2025).
Any honest discussion of scores has to acknowledge that the research community has argued about NPS for nearly two decades.
On one side, Fred Reichheld, who introduced the metric, and his Bain colleagues Darci Darnell and Maureen Burns revisited it in Harvard Business Review in 2021. They acknowledged that the score is often gamed and misused, including when it is tied to frontline bonuses, and proposed an accounting-based complement, Earned Growth Rate, to connect advocacy with actual revenue (HBR, November-December 2021).
On the other side, Timothy Keiningham and colleagues tested the claim that NPS is a superior predictor of company growth using Norwegian customer data across 21 firms and more than 15,500 interviews, and found no support for that claim compared with other satisfaction and loyalty measures (Journal of Marketing, 2007).
The practical conclusion is not that NPS is useless. It is that no single score deserves to carry the full weight of decision-making, whichever score an organization chooses.
A second limitation is that most relationship metrics measure stated intention. Qualtrics XM Institute compared consumers' likelihood to recommend after 5-star experiences with what they reported actually doing. While 86% said they were likely to recommend, only 50% reported having told friends or family after a very good experience, a 36-point gap (XM Institute, September 2025). The comparison relies on two self-reported questions, so it is indicative rather than definitive, but it supports a simple discipline: treat intention metrics as signals to explain, not as outcomes to celebrate.
Once a score is linked to targets or compensation, it changes behavior. Reichheld and colleagues themselves describe the pressure this creates. The symptoms are familiar: customers asked for a "nine or ten", surveys sent selectively after good interactions, and analysis effort spent explaining movements rather than understanding customers.
A useful test is whether a team could improve its score without improving the experience. If the answer is yes, the metric is being managed rather than the customer journey.
Three bodies of research suggest what a score should be paired with:
The following sequence turns a headline movement into a decision:
Software can shorten steps three to five considerably. In Pisano, AI-based text analytics categorizes open-ended feedback into themes and sentiment so that qualitative data can be reported and trended. Key driver analysis visualizes the themes respondents mention most and links directly to the underlying feedback in the inbox. The Quadrant Chart in Analytics & Insights plots attributes by importance and satisfaction to show where attention is most needed, and the Impact Simulator lets teams model how improving specific areas might affect future scores. These tools support prioritization. The decision about what to change, and who changes it, still belongs to the organization.
The most common misuse of NPS is not a flaw in the metric. It is asking a single number to do a job it was never designed for. A score is a smoke alarm: it tells you something is happening and roughly where. Finding the fire requires drivers, verbatims and operational data, and putting it out requires an owner.
The next article in this series looks at a less visible problem with scores: the customers whose experiences never reach the survey at all.