What public retention disclosures actually tell us

A separate corpus of 20,860 customer- and client-retention items spanning 4,928 unique companies. After a v2 reclassification pass that extracts numeric rates and directional signal explicitly, the rate-bearing population is now 1,069 (up from 373 under the v1 prompt) and the directional-only population is 960 (up from 489). Retention is structurally different from NPS — not a survey score, but a measure of customers (or revenue) carried from one period to the next. The data still tells two distinct stories: a gross-retention story below 100%, and a Net Revenue Retention story that can exceed it. Both matter, and they should not be benchmarked against each other.

Distinguishing gross retention from NRR.

Retention disclosures cluster into two distinct ranges:

The corpus contains roughly 6.5× more gross-retention disclosures than NRR disclosures, but NRR is rising faster — and NRR is the more analytically valuable number because it captures the expansion dimension that pure retention misses. NRR disclosures more than doubled from 2024 (15) to 2025 (31) and 2026 is already at 35 by early May.

20,860 items, 4,928 companies, 2018–2026.

Roughly 5× the size of the NPS dataset by raw items. The v2 reclassification pass roughly tripled the rate-bearing yield (373 → 1,069) and doubled the directional yield (489 → 960) — both because the v2 prompt knows how to read "achieved a 96% retention rate" and "client retention rose 20%" as structured fields rather than passing them through as off-topic prose. Most "customer retention" press writing remains generic (programmes launched, vendor partnerships, churn-reduction commentary) rather than rate-bearing, but the signal density of the rate-bearing half is now competitive with the NPS corpus.

Gross retention is bumpier than it looked. NRR is accelerating.

Two distinct movements are now visible. Gross retention disclosures dipped sharply in 2022–23 (from 201 in 2021 to a trough of 60 in 2023, before recovering to 131 in 2024 and 128 in 2025). The dip lines up with the broad SaaS valuation reset of 2022 and the post-ZIRP corporate disclosure pullback — companies stopped self-reporting retention figures publicly when the numbers were under pressure. NRR disclosures kept growing through that same period (13 in 2022, 11 in 2023, 15 in 2024) and inflected sharply upward in 2025 (31) and 2026 YTD (35 already, on pace for a record year). 2026 is the first year in the corpus where NRR disclosures will likely outnumber gross-rate disclosures.

Snowflake remains the single most disciplined NRR-disclosure cadence in the corpus — 10 rate-bearing rows spanning 2020 to 2025, all between 125% and 171%. Figma, GitLab, Showpad, ketteQ, Wallarm, PayZen and Monday.com are the next-tier consistent NRR disclosers.

The 90–99% gross-retention zone is the corpus's centre of gravity.

The 90–99% gross-retention zone now accounts for 57% of the corpus (608 of 1,069), up sharply as a share of the v2 yield. This is exactly the band you'd expect — it's the standard "we kept most of our business" disclosure that companies use to anchor narrative without committing to either a brag-worthy near-100% or an admit-defeat sub-90%. The 22% disclosing below 90% is more interesting: these are companies acknowledging real customer turnover, which (unlike NPS) doesn't seem to suffer the same survivorship-bias filter. Public retention numbers below 90% appear in earnings calls more freely than negative NPS scores do.

SaaS dominates by volume; Healthcare and Cybersecurity post the highest medians.

Gross retention (rates 1–100%)

Net Revenue Retention (rates >100%)

NRR remains essentially a SaaS-only disclosure norm. SaaS / Technology accounts for 124 of the 143 NRR rows (87%), at a median of 117.5%. The remaining 19 NRR disclosures scatter across fintech, cybersecurity, vertical-software firms and one outlier consumer-financial-services row at 119%. The pattern hasn't changed materially from the v1 reading: SaaS investors expect NRR; nobody else does, yet.

Within gross retention, the picture has shifted. Healthcare (98%, n=14) and Transportation/Logistics (97.5%, n=10) now post the highest medians — both above SaaS. These are sectors where the v2 prompt picked up disclosures the v1 prompt ignored, often from earnings transcripts where retention is reported as a renewal-rate metric. Retail (80%) and Retail/E-commerce (63.9%) and Automotive (69.5%) sit far below SaaS, which actually reflects industry reality. Physical retail and auto sales have genuinely tougher repeat-customer dynamics than subscription software. The honest read is still that 70% in retail is a defensible result; 95% in SaaS is table stakes.

The companies whose retention story we can actually track over time.

The 960 directional rows give the corpus a freshness layer the rate-bearing rows can't.

Recent directional claims from April–May 2026 alone include "+42%" (Burst), "down" (Spectrum, T-Mobile, ERIE Insurance, Asana), "up" (NRC Health, Charter Communications, CrowdStrike, Verizon, Pinterest, Yubico, Workiva, Backblaze, Netflix, Rubrik, Apple) and "+2pp" (Tradewindow). These are easier to take at face value precisely because they don't commit to a specific number with no methodology context. A telecom that admits retention is "down" without quantifying is being more candid than a SaaS vendor claiming a clean 100% gross.

For prospecting and outreach, both halves of the corpus are useful: rate-bearing rows are benchmarkable; directional rows surface intent and movement. Combined, the corpus covers 4,928 named companies that have publicly engaged with retention as a metric in the last 8 years.

Five things companies should do when they publicly disclose a retention number.

1. Name the metric exactly. Gross logo retention, gross dollar retention, net revenue retention. The single biggest credibility lift available, currently absent from ~95% of disclosures. NRR vs gross are not interchangeable.
2. Specify the cohort. "All customers" or "enterprise tier" or "FY2024 cohort" matters. A retention figure without a denominator is unverifiable.
3. State the time window. Quarterly, annual, trailing-12-month. "Retention rate of 94%" with no window can mean anything.
4. Disclose for at least three consecutive periods. One-shot announcements look like cherry-picks. Snowflake's 10 NRR disclosures and Trade Desk's 31 gross disclosures are the gold standard. Apple's range of 77–95 across 6 disclosures is more credible than any single 95% claim.
5. Include the directional context even when the number is good. "94%, up from 91% last year" is harder to fake and easier to trust than a standalone 94%. "NRR of 125%, up from 119% last year" is the SaaS-investor gold standard.

How this was produced.

15 years of Google Alerts for "Customer retention", "client retention", "net revenue retention" and "renewal rate". Items classified by claude-haiku-4-5 with the metric_type tag set to RETENTION. Numeric rates extracted into a dedicated retention_rate field (0–200% range, capturing both gross retention 0–100% and Net Revenue Retention >100%); directional language ("doubled", "+2pp", "increased", "down") into a retention_change field. The v2 reclassification pass re-ran every retention-relevant item with a refined prompt that knows to extract these fields explicitly; the rate-bearing population grew from 373 to 1,069 and the directional-only population from 489 to 960 as a result. Soft dedup at canonical level on (company-or-domain, value, ±5-day window). Retention Intelligence does not independently verify any reported figure.

Generated 2026-05-07 (v2 corpus) · Dataset: 20,860 RETENTION-tagged items · 3,599 canonical SCORE_ANNOUNCEMENT rows · 1,069 with valid 1–200% rate (926 gross + 143 NRR) · 960 with directional claim · 4,928 unique companies.