Why NPS Is Overrated (And What to Measure Instead)

NPS Measures Intentions, Not Behavior

Net Promoter Score asks one question: "How likely are you to recommend this product to a friend?" The answer is supposed to predict growth. In practice, it predicts almost nothing.

I have seen NPS fail as a useful signal in three different environments. As a consultant working with 20+ SMBs and startups through Gradient Growth, where I have watched teams celebrate high NPS scores while their retention curves cratered. At a financial social media startup before its acquisition, where NPS told us users liked the product while behavioral data told us they were leaving. And now on a GenAI squad at a major finance technology company, where we have enough data volume to rigorously test what predicts retention and revenue. NPS consistently ranks near the bottom.

The core problem with NPS is the gap between stated preference and revealed preference. Stated preference is what people say they will do. Revealed preference is what they actually do. Decades of behavioral economics research confirm these are often completely different things. NPS lives entirely in the stated preference world. It captures sentiment. It does not capture behavior. And sentiment is a terrible predictor of business outcomes.

This is not a contrarian take for the sake of being contrarian. NPS has real, structural problems that make it unreliable as a decision-making tool. If you are using it as your primary customer health metric, you are steering with a broken compass.

The Three Problems with NPS

Problem 1: Survey Bias Distorts the Signal

NPS response rates in most SaaS products sit between 10% and 30%. That means you are hearing from a self-selected minority of your user base. The users who respond to NPS surveys are disproportionately either very happy or very frustrated. The silent majority in the middle, the users whose retention decisions will actually determine your revenue trajectory, rarely fills out the survey.

At one consulting client (a B2B SaaS tool with around 2,000 active users), NPS response rate was 14%. Their score was 52, which looks strong. But when we segmented responders against non-responders using behavioral data, we found that non-responders had 23% lower feature adoption and 18% worse retention at 90 days. The NPS score reflected the enthusiasm of their most engaged users and completely missed the quiet churn happening in the rest of the base.

This is not a fixable problem. You cannot force a representative sample in a voluntary survey. The bias is structural.

Problem 2: The Score Does Not Correlate with Retention or Revenue

At the startup I worked at before the acquisition, we ran NPS quarterly for 18 months. Our score ranged from 38 to 56 across that period. During the same window, monthly retention fluctuated between 61% and 79%. The correlation between NPS and next-quarter retention was 0.12. Statistically, that is noise. We could have rolled dice and gotten a similar predictive signal.

Meanwhile, three behavioral metrics correlated with retention above 0.60: weekly active usage frequency (0.71), feature adoption depth (0.64), and time to first core action (0.67). None of these required a survey. All of them were available in our event data, updated in real time, and decomposable by cohort, channel, and segment.

The disconnect is not unique to that company. A 2019 study in the Journal of Marketing found that NPS was a weaker predictor of revenue growth than simple customer satisfaction scores or, more importantly, behavioral metrics like repeat purchase rate. Bain and Company (the consultancy that created NPS) has published case studies showing NPS correlating with growth, but independent research consistently finds the relationship is weak and often confounded by other variables.

Problem 3: It Is Not Actionable

Suppose your NPS drops 10 points this quarter. What do you do? The score tells you that fewer users would recommend you, but it does not tell you why, which users, or what specifically changed. To diagnose the drop, you need to dig into verbatim comments (if you collected them), cross-reference against product changes, and segment by cohort. By the time you have done all that work, you have essentially rebuilt the behavioral analysis you should have done in the first place.

I have sat in quarterly business reviews where teams spent 30 minutes debating why NPS moved 5 points. Five points is well within the margin of error for most sample sizes. The conversation was noise analyzing noise. That same 30 minutes spent reviewing [retention curves](/insights/retention-curve-analysis-guide) or feature adoption funnels would have surfaced three actionable findings.

NPS is not useless as a qualitative signal. If you already have strong behavioral metrics in place and want to layer on a sentiment check, fine. But as a primary metric for understanding customer health, it fails on reliability, validity, and actionability.

What to Measure Instead

The three metrics below are not theoretical alternatives. They are the metrics I have used across consulting engagements and internal roles that consistently predict retention and revenue. Each one is behavioral (measuring what users do, not what they say), available from event data (no survey required), and directly actionable (a change in the metric points to a specific area to investigate).

Metric 1: Retention Rate (Cohort-Based)

Retention rate measures the percentage of users who come back to your product over time. Not whether they would recommend it. Whether they use it.

Cohort-based retention is the version that matters. A blended retention number averaged across all users hides critical trends. You need to see retention curves for each weekly or monthly cohort, because that is the only way to determine whether your product is getting better or worse over time.

At the startup before the acquisition, retention analysis was the centerpiece of the entire analytics function. When we redesigned onboarding and saw D30 retention move from 18% to 31% in subsequent cohorts, that was the single most important data point in the acquisition diligence process. No acquirer asked about our NPS. Every acquirer asked about retention curves.

The implementation is straightforward. Track user activity by cohort (signup week or month), calculate the percentage of each cohort that returns at D1, D7, D14, D30, and beyond, and monitor whether the curves are flattening (healthy) or decaying to zero (not). I wrote a full guide on how to build and interpret this analysis: [Retention Curve Analysis](/insights/retention-curve-analysis-guide).

The predictive power is not subtle. Across my consulting engagements, the correlation between D30 cohort retention and 12-month revenue retention averages 0.68. NPS has never come close to that in any data set I have worked with.

Metric 2: Feature Adoption Depth

Feature adoption depth measures how much of your product's value surface a user engages with. It distinguishes between users who try one feature and users who integrate multiple features into their workflow.

The metric is simple: count the number of distinct core features each user engages with at least twice within a measurement window. "At least twice" filters out accidental clicks and one-time exploration. Repeat usage of a feature is the minimum bar for adoption.

At one consulting client (a project management SaaS), we segmented users by the number of features adopted (defined as used 2+ times in a 30-day window). Users who adopted 1 feature had 34% 6-month retention. Users who adopted 3+ features had 78% 6-month retention. That 44-point gap is the difference between a churning user and a retained one, and it was invisible in their NPS data. Promoters and detractors had nearly identical feature adoption distributions, meaning NPS was not even capturing the dimension that mattered most.

Feature adoption depth also tells you where to invest. If a feature has high trial rates but low repeat usage, the feature is discoverable but not delivering sustained value. If a feature has low trial but high repeat usage among those who find it, you have a distribution problem, not a product problem. NPS tells you none of this. For the full framework on measuring and acting on feature adoption, see [Feature Adoption Rate Guide](/insights/feature-adoption-rate-guide).

Metric 3: Time to Value

Time to value (TTV) measures the elapsed time between a user's first interaction with your product and the moment they experience its core value. It is the leading indicator that sits upstream of both retention and feature adoption.

At a B2B analytics client, we measured TTV as the time from account creation to the user generating their first automated insight. The median was 4.2 days. Users who reached value in under 24 hours retained at 67% at D90. Users who took more than 72 hours retained at 19%. That is a 3.5x difference in retention explained by a single metric, measured in the user's first few days.

TTV is actionable in a way NPS never is. If TTV is too long, the diagnosis is clear: something in the onboarding path is creating friction between signup and value delivery. You can instrument each step, identify the bottleneck, and run experiments to reduce it. At the same client, we cut TTV from 4.2 days to 1.8 days by eliminating three setup steps and auto-configuring the data connection. D90 retention for subsequent cohorts improved by 14 percentage points.

I have written a detailed guide on how to define the value moment, measure TTV, and reduce it: [Time to Value](/insights/time-to-value-onboarding-metric).

The Hierarchy of Customer Health Metrics

If I had to rank customer health metrics by predictive power and actionability, the ordering would be:

1. Retention rate (cohort-based) is the definitive measure of whether users come back. If retention is healthy, the business is viable. If it is not, nothing else matters. 2. Feature adoption depth explains why retention is what it is. It decomposes the black box of "do they come back" into "which parts of the product are they using." 3. Time to value predicts whether new users will become retained users. It is the earliest signal you can act on to improve the downstream metrics. 4. NPS is a lagging, noisy, undecomposable sentiment indicator. It belongs at the bottom of the stack, not the top.

Most teams invert this hierarchy. They run NPS because it is easy to implement (one survey question), produces a clean number (score from -100 to 100), and has executive name recognition. Ease of measurement is not the same as quality of measurement. The three behavioral metrics above require event tracking and cohort analysis, which takes more work upfront. But they are available in real time, represent the full user base (not just survey responders), and directly connect to the levers your team can pull.

When NPS Is Acceptable

I am not arguing you should never run an NPS survey. If you have behavioral metrics in place and want a periodic sentiment check, NPS can surface qualitative themes through the open-ended follow-up question. The verbatim responses ("I love the reporting but the export is broken") are more useful than the score itself.

What I am arguing is that NPS should never be your primary customer health metric, your board-level KPI, or the basis for product prioritization decisions. It does not have the reliability, granularity, or predictive power to serve any of those functions. Behavioral metrics do.

Replace NPS at the top of your dashboard with retention, feature adoption, and time to value. You will make better decisions, faster, with more confidence. The data will not just tell you how customers feel. It will tell you what they are actually doing.

---

I help teams replace vanity surveys with behavioral metrics that predict business outcomes. [lester@gradientgrowth.com](mailto:lester@gradientgrowth.com)