"Net Dollar Retention Benchmarks: Where'd All the 130s Go?" — @cjgustafson222
Why this is in the vault
The promised quantitative follow-up to the NDR disclosure piece: original dataset of 95 public software companies across 1,500+ quarterly disclosures, proving the 130% NDR club effectively collapsed and that the published median has ~6-8 points of survivorship bias baked in.
⚠️ Sponsorship
This issue is sponsored by Abacum (FP&A/AI-native planning platform). The sponsor block appears at the top ("Mostly metrics is proudly powered by Abacum") and promotes a CJ-moderated Abacum webinar on FP&A tooling vs. AI-native platforms. Clearly demarcated; no bleed into the NDR analysis. Abacum is a repeat sponsor on this newsletter.
The core argument
Gustafson built an original dataset: 95 public software companies, 1,500+ quarterly NDR disclosures from early 2020 onward. The headline: the 130% club had 18 members in 2020-2021 and now has exactly two (Palantir and Figma).
Key benchmark data:
- P75 NDR: compressed from ~130% to 116% (-14 pts)
- P50 NDR: compressed from ~123% to 110% (-13 pts)
- P25 NDR: compressed from ~113% to 103% (-10 pts)
- Interquartile spread narrowed from 18-20 pts at peak to 10-13 pts today
The expected narrative was a K-shaped bifurcation — winners escaping while losers got crushed. The data rejected that. All quartiles slid in parallel, roughly "everyone moved down a floor."
Survivor bias correction: Companies that stopped disclosing (Fastly, CrowdStrike, DocuSign) had NDRs near or below median at the time of exit. The disclosed median is a median of the survivors. Gustafson estimates the real aggregate compression is 19-21 points, not the headline 13.
Peak-to-trough standout: Snowflake dropped 53 points (177% → 124%) — described as "the most extreme revenue retention compression in public software history." Still top-quartile by current standards, which puts current benchmarks in perspective.
What drove the bleeders:
- Consumption-priced companies (Snowflake, Twilio): customers throttled usage faster than license renegotiations
- Productivity tools tied to dev/product teams (Asana, Monday): tech layoffs 2022-2023
- SMB-exposed marketplaces (BILL, Toast): macro depression in customer base
Valuation translation: Each 10 pts of NDR ≈ 1x forward revenue multiple (0.07x per point at <20% growth, 0.18x per point at >30% growth). For a $250M NTM revenue business at 8x, a 10-pt NDR improvement adds ~$250M EV immediately; if it also bumps analyst growth assumptions, closer to $500M over 18 months. This is the quantitative reason companies hide declining NDR — it's a direct valuation lever, not a vanity metric.
Mapping against Ray Data Co
Mapping strength: medium, with one sharp consulting-context angle.
- Pricing and retention conversations with clients. Ben is a Deal Solutions Architect at phData — scoping and discovery conversations often include a client's current data/AI ROI story. NDR benchmarks give a hard anchor: if a client's AI or data platform spend isn't generating measurable expansion, the floor is ~103% (P25 today) and the ceiling for best-in-class is barely above 130%. Useful calibration when a client asks "is our retention normal?" or when building a business case for a data platform renewal.
- "The median is not the median" as a client-facing line. The survivorship-bias corrected median (19-21 pts of real compression vs. 13 pts published) is a sharp practitioner insight RDCO can deploy when clients benchmark themselves against public SaaS comps. Most comp sets pull from published disclosures, which systematically over-state retention health.
- Sanity Check content angle. The valuation math section (10 pts NDR = 1x multiple = $250M EV delta on a mid-size SaaS) is exactly the kind of "here's what the number means in dollars" framing that Sanity Check can adapt — not restate, but use as evidence for an original argument about how companies weaponize metric opacity against investors.
- Investing diligence reflex. The consumption-pricing vs. seat-based split (consumption had highest highs, deepest drops) is a reusable variable in the Markov capital-cycle thesis — when evaluating data/AI infrastructure names like Snowflake in the chip-fab/memory cycle, consumption-model NDR volatility is the revenue durability risk to price in.
Honest read: core mechanics (B2B SaaS NDR) are one step removed from RDCO's own surfaces, but the client-facing utility and the Sanity Check content angle both have real legs. The Snowflake data point alone is load-bearing for any phData client conversation touching Snowflake spend justification.
Related
- [[2026-05-31-mostly-metrics-ndr-net-dollar-retention-decline]]
- [[Sanity Check]]
- [[project_investing_markov_capital_cycle]]
- [[2026-06-04-mostly-metrics-consumption-based-arr]]