Alignment Score vs NPS: Why the Industry Standard Metric Is Measuring the Wrong Thing

TL;DR

• NPS measures how users feel about your product (sentiment). Alignment scores measure how well your product understands each user (understanding quality). These are fundamentally different things.
• Research shows NPS is not a reliable predictor of customer retention [1]. A product can have high NPS and high churn simultaneously because users like the concept but leave when the experience does not adapt to them.
• Alignment scores offer a per-user, diagnostic alternative that measures the cause (understanding quality) rather than the symptom (satisfaction sentiment).

Since Fred Reichheld introduced the Net Promoter Score in his 2003 Harvard Business Review article [2], NPS has become the dominant customer success metric, used by two-thirds of the Fortune 1000 [3]. But two decades of academic scrutiny have revealed serious limitations in what NPS can actually predict. This post covers the structural blind spots in NPS, the research questioning its link to retention, and why per-user alignment scoring offers a more diagnostic alternative.

The NPS Blind Spot

NPS has a structural blind spot: it asks about the product in aggregate, not about the relationship between the product and the individual user.

When someone scores a 9 on NPS, they might mean: “This product is great and fits my workflow perfectly.” Or they might mean: “This product has a great reputation and I believe in its potential, even though it has not clicked for me personally yet.” Both responses produce a 9. But one user is retained and the other is at risk. As Nielsen Norman Group has noted [4], NPS measures overall loyalty but cannot assess satisfaction with specific elements of the user experience, and the binning method (promoter/passive/detractor) discards nuanced data about what is actually happening.

NPS treats all promoters the same. It cannot distinguish between a user who promotes because the product is deeply aligned with their needs and a user who promotes because they like the brand. The first user stays. The second user is one bad experience away from leaving. Research confirms [5] that promoters churn for reasons NPS never captures: a champion leaves the company, budgets shift, the product is well-loved but duplicative of entrenched tools, or the company gets acquired.

This is the high-NPS, high-churn paradox. Typical NPS response rates sit around 4.5% [6], meaning NPS captures sentiment from a small, self-selected group of engaged users while churn happens among the silent majority. Companies with ostensibly healthy NPS scores lose customers at rates their survey data never predicted.

Why NPS Struggles to Predict Retention

The evidence that NPS is a weak retention predictor comes from multiple independent studies.

ProfitWell studied over two thousand subscription companies and found that NPS does not correlate with renewal or churn [7] as an aggregate measure. The only exception: companies with NPS in the top 25% of their vertical saw a modest 5 to 10% bump in retention. For the other 75% of companies, the correlation was absent.

A 2023 study by Customer Cross, analyzing 265,000 subscription businesses over 10 years [8], found no statistical correlation between NPS and churn risk. The researchers attributed this to low response rates creating biased datasets, with responses skewing to extremes that poorly reflect the full customer base.

Academic replication attempts have also been mixed. Jeff Sauro at MeasuringU found only a weak correlation (r = 0.35) between NPS and company growth [9], and noted it functions better as a trailing indicator than a predictor. Multiple academic studies, including work by Keiningham et al. (2007) and Morgan and Rego (2006), have failed to confirm Reichheld’s original claim [10] that NPS is a superior predictor of growth.

Why alignment scores offer a diagnostic alternative. The core problem with NPS is that when it drops, you know users are less satisfied, but you do not know why. Was it a feature change? A pricing adjustment? A competitor launch? NPS gives you the temperature, not the diagnosis.

When an alignment score drops, you know exactly what changed. Belief coherence dropped? Your model of the user has contradictions. Convergence stalled? Your interactions are not generating new signal. Depth decreased? Beliefs are going stale. Each component points to a specific operational response.

The Anatomy of Each Metric

Here is what each metric actually measures under the hood:

NPS is a single number derived from a survey question. It is collected periodically (monthly or quarterly). It reflects the user’s conscious, stated opinion about the product at the moment they answer the survey. Research has documented that it is subject to sampling bias, non-response bias, cultural bias, and questionnaire bias [11]. It aggregates to a company-wide score. And as Itamar Gilad has detailed [12], different response distributions can produce identical NPS scores, masking important shifts in customer sentiment.

Alignment score is a composite metric computed continuously from the user’s self-model. It has three components: belief coherence (how internally consistent is the product’s understanding of this user), directional convergence (is understanding improving over time), and context depth (how rich is the understanding). It is computed per user, updated on every interaction, and reflects the actual quality of the product-user relationship rather than the user’s stated opinion about it.

The deepest difference: NPS asks the user to evaluate the product. Alignment scores evaluate the product’s understanding of the user. One is subjective sentiment. The other is a structural measurement of how well the system models each individual.

1// NPS: periodic survey, aggregate score← sentiment
2const nps = await survey.collect({
3  question: 'How likely to recommend? (0-10)',
4  frequency: 'quarterly',
5  scope: 'all_users'
6});
7// Returns: { score: 72, promoters: 61, passives: 28, detractors: 11 }
8// Diagnostic value: none
9
10// Alignment: continuous, per-user, diagnostic← understanding
11const alignment = await clarity.getAlignment(userId);
12// Returns: {
13//   overall: 0.84,
14//   components: {
15//     belief_coherence: 0.91,← is our model consistent?
16//     directional_convergence: 0.82,← are we getting better?
17//     context_depth: 0.73← how much do we know?
18//   },
19//   trend: 'converging',
20//   risk: 'low'
21// }
22// Diagnostic value: pinpoints exactly what is failing

When NPS Still Matters

NPS is not useless. It measures something real: brand sentiment and likelihood of organic referral. These matter for growth, and NPS does correlate with expansion revenue [13] even when its link to retention is weak.

NPS is valuable for:

• Benchmarking against competitors. It provides a standardized comparison that alignment scores cannot yet offer.
• Tracking brand health over time. Long-term NPS trends reveal brand trajectory.
• Identifying detractors. Users scoring 0-6 need immediate attention regardless of alignment.
• Board-level communication. Investors and board members understand NPS.

But NPS should not be the primary metric for product quality, retention prediction, or personalization effectiveness. It was not designed for those purposes [14].

Trade-offs

Alignment scores are not a perfect replacement for NPS. Here are the limitations:

Infrastructure requirements. NPS requires a survey tool. Alignment scores require a self-model layer with belief tracking, confidence calibration, and real-time computation. The infrastructure investment is orders of magnitude higher.

Cold start. NPS works from day one. Alignment scores need enough interactions to build a meaningful self-model. For new users, alignment scores are thin and potentially unreliable.

External benchmarking. NPS is universal. Any company in any industry can compare scores. Alignment scores are internal, with no industry benchmark yet. This makes it harder to communicate alignment to external stakeholders.

Complementary, not replacement. In practice, both metrics serve different purposes. NPS for brand health and external benchmarking. Alignment for product quality and retention prediction. Running both adds complexity.

Organizational adoption. Everyone knows NPS. Board members understand it. Executives report on it. Introducing alignment scores requires education, change management, and proving the metric’s value before the organization trusts it.

What to Do Next

If NPS is the primary product quality metric on your dashboard, here is how to evaluate whether alignment scores could complement it:

1. Run a correlation analysis on your NPS and churn data. Calculate the correlation between NPS scores and 90-day retention for your last 500 users. If the correlation is below 0.5, NPS is not doing its job as a retention predictor, and a complementary metric is warranted. (Given that large-scale studies have found little to no correlation [15], this test is worth running.)

2. Build alignment scores for a test cohort. Pick 100 active users and build self-models for them. Compute alignment scores monthly for 90 days. Compare alignment-retention correlation to NPS-retention correlation for the same cohort. The difference will tell you whether the investment in alignment scoring is justified for your specific product.

3. Use alignment for operations, NPS for communication. Do not try to kill NPS. Instead, use alignment scores internally for product decisions (what to build, what to fix, which users need attention) and NPS externally for benchmarking and board communication. Over time, as alignment proves its value, it will naturally become the primary internal metric.

NPS tells you how users feel. Alignment tells you how well you understand them. Measure what matters with Clarity.

References

1. not a reliable predictor of customer retention
2. 2003 Harvard Business Review article
3. used by two-thirds of the Fortune 1000
4. Nielsen Norman Group has noted
5. Research confirms
6. sit around 4.5%
7. NPS does not correlate with renewal or churn
8. 265,000 subscription businesses over 10 years
9. weak correlation (r = 0.35) between NPS and company growth
10. failed to confirm Reichheld’s original claim
11. sampling bias, non-response bias, cultural bias, and questionnaire bias
12. Itamar Gilad has detailed
13. correlate with expansion revenue
14. not designed for those purposes
15. large-scale studies have found little to no correlation

Related

• The Alignment Score, Explained: Why It Matters More Than Engagement
• The Missing Metric in Your AI Dashboard
• Evaluation Beyond BLEU: Measuring AI Alignment