What 15 years of NPS disclosures actually look like

After a parser fix, classifier overhaul and full-archive backfill, the corpus has grown roughly 25× — to 44,749 classified items, 10,606 unique companies and 3,763 disclosed-score rows on the live page. The patterns that survive the larger sample are the ones worth taking seriously.

Volume keeps rising. Scores plateau-ed half a decade ago.

Scores climbed from 2011 to a 2021 peak (avg 72), then drifted modestly lower (~67-71) and have stayed in that band. 2026 is on pace for ~750 disclosed scores at an avg of 70.8 — broadly consistent with the post-COVID plateau.

Mean continues to sit 6-8 points below median in every window — the 43 negative-NPS disclosures and the long tail of below-average scores pull the mean down. Median is the more honest "where companies actually are" measure.

5.7% of scores remain implausible-on-their-face.

216 scores (5.7%) at 95 or above. 70 scores (1.9%) exactly 100. With a 3,763-row sample these proportions are stable — the quality-bar pattern from the smaller corpus held up.

Composition of the ≥95 zone is unchanged: small B2B SaaS vendors, niche consultancies, and PR-driven announcements rather than consumer companies with meaningful sample sizes. The ≥95 tail almost never appears in earnings call coverage from the major investor wires.

Where the disclosures cluster.

3,570 source sectors collapse into ~20 top-level buckets. Ranked by disclosure volume:

Tech/SaaS still dominates volume (30%) but the gap has narrowed — Financial Services, Healthcare and Retail are each at 12-13%. The corpus is now genuinely cross-sector.

Highest performing: Healthcare & Pharma (77.5 avg, 84 median), HR & Staffing (73.8), Tech/SaaS (70.7), Professional Services (70.5). All sectors with high-engagement specialty buyers and small comparison sets.

Underperforming: Energy & Utilities (43.8 — captive customers, regulated, structural friction), Travel & Hospitality (56.6), Media & Advertising (59.3). Confirms long-running CX research about where NPS is structurally hard.

Source intelligence.

The 44,749-item corpus is sourced from ~3,000 distinct domains. A handful of channels do most of the heavy lifting; another handful punch above their weight on yield.

Top sources by volume of HIGH-quality disclosures

Three families do the bulk of the work:

• PR wires (PRN, BW, GlobeNewswire) — bedrock. ~700 unique companies between them. Mostly self-issued press releases.
• Earnings and investor research (Yahoo Finance, Motley Fool, StreetInsider, Morningstar, Seeking Alpha, Nasdaq) — paraphrases earnings calls where executives quote NPS. Best filter for forced or audited disclosures.
• Major media is misleading. Forbes is a striking outlier: 1,644 mentions but only 37 HQ scores (2.2% yield). Mostly explainer/marketing pieces, almost no actual disclosures.

Top sources by yield (≥10 articles, sorted % HQ)

Niche trade press and regional regulator wires have far higher hit rates than the big aggregators. UK financial trade press (InvestEgate, The Intermediary, FMJ, Sharecast), Australian and NZ business wires (BrokerNews.com.au, Voxy.co.nz, Appliance Retailer) and sector-specific outlets (HealthcareITNews, Collision Repair Mag, Security Systems News) consistently produce score-bearing pieces.

Five things worth flagging from the larger sample.

1. Negative-NPS disclosures jumped from 5 to 43. The parser fix surfaced a long tail of bad-news stories that were hidden under the old plumbing. Highlights: Reliance Jio (-46, 2021), Singapore-market financial services (-37, 2014), Club Monaco (-28, 2026), CapFed (-26, 2019). Negative disclosures are still rare (~1.1% of the set) but the survivorship signal is meaningfully weaker than it looked in the smaller corpus.
2. Tech/SaaS share of disclosures dropped from 35% to 30%. As the corpus broadened, Financial Services, Retail and Healthcare each grew faster. The pre-fix sample was tilted toward tech press releases; the actual disclosure landscape is more balanced than it appeared.
3. Energy & Utilities is the standout low performer. 47 disclosures averaging 43.8. Captive customer base, heavy regulation, billing friction — these are the structural barriers NPS theory predicts.
4. Programme announcements: 1,434 stories of named companies launching or expanding NPS programmes, on top of the 3,763 disclosed scores. Together: ~5,200 named companies actively engaged with NPS over 15 years. That's the SEO and prospect-list base for the broader CustomerGauge / TheCustomerTest combined dataset.
5. Scores haven't risen in 5 years. The 2021 peak (avg 72) hasn't been matched since. Either (a) post-COVID realism has set in, (b) growing sample size has diluted the cherry-pick effect, or (c) genuine stagnation in CX outcomes despite continued investment. The honest answer is probably "all three".

Programme announcements: 1,434 named companies, no score (yet).

Programme stories track the score-disclosure stream but ~1.5× lagged in time. New programmes launched in 2017-2019 (peak: 207 in 2019) feed the score-disclosure peak in 2020-2021. The relationship is consistent and the leading-indicator value of programme news is clearer at this scale.

How this was produced.

15 years of Google Alerts for "Net Promoter", "Customer Effort Score", "Customer retention" and the "Net Promoter S-1" filing-specific query, plus a 2015 HTML email archive. After a CRLF parser fix the raw item count moved from ~7,400 to 48,685. Items classified by claude-haiku-4-5 with metric tagging (NPS / CES / CSAT / RETENTION). Scores validated to range. Soft dedup at canonical level on (company-or-domain, score, ±5-day window) with quality-tiered selection. Companies missing a name on high-value rows are extracted via Sonnet for recent items and Haiku-with-Sonnet-escalation for older ones; remaining gaps land in data/needs_manual.csv. NPS Intelligence does not independently verify any reported figure.

Generated 2026-05-05 · Dataset: 44,749 classified entries, 3,570 source sectors, 10,606 unique companies, 3,763 disclosed-score rows on the live page.