SaaS Metrics Benchmarks

Greetings! Ben here.

Bookmark this page for future reference or download the report and markdown file. Feed the MD file into your favorite LLM for future use.

The download includes an HTML (better than a PDF report) and a MD file. The MD file is contained in ZIP folder b/c my download plugin did not like the MD file extension.

The 2026 SaaS Benchmarks Report | The SaaS CFO

The 2026 SaaS Benchmarks Report

Nine leading benchmark studies, reconciled into one set of numbers you can plan with.

The SaaS CFO · July 2026; revised October 6, 2026 · One honest read of the credible 2025 to 2026 SaaS benchmark studies

Data vintages include 2024 actuals, CY2025 results, Q2 2025 observations, pooled 2013 to Q2 2025 enterprise data, and ICONIQ’s year-cohorted 2026 data (through Q2 2026). Every table identifies its applicable period.

Every year the SaaS industry produces a dozen excellent benchmark reports. And every year operators compare numbers across them without realizing the reports do not measure the same things the same way. Reported net retention can differ materially because of denominator choice, cohort construction, company scale, and sample composition. I built this report to fix that. It does two things: it maps the credible 2026 sources into common ARR bands when the evidence supports it, and it keeps results separate when definitions, periods, or populations are not comparable. Every source is credited. Every modeled value is labeled.

1. Executive Summary

I analyzed nine credible SaaS benchmark studies, pulled the core operating metrics by ARR band, and mapped comparable results into common bands. Where formulas or populations could not be reconciled, I kept the results separate and said so. Four findings stand out:

  • We got more efficient, and we paid for it with retention. Rule of 40, CAC payback, and ARR per employee all improved in 2025. But R&D fell from 35% to 27% of revenue, and gross retention gave back about four points. The efficiency gains arrived alongside spending cuts and a retention decline; the data cannot say the cuts caused the gains, but they were not free. If your GRR slipped in 2025, that is the first number to fix in your 2026 plan.
  • Growth has structurally decelerated. Benchmarkit’s median growth fell from about 30% in 2022 to about 20% in 2025 — four straight annual declines, with top-quartile growth falling even faster. KeyBanc shows the same slide through 2024 (31% → 20% → 15%) but a projected rebound: 20% expected for 2025 and 25% for 2026. Growth is no longer something you can buy with spend. The companies still growing fast are the ones whose unit economics let them keep reinvesting. Pressure-test your 2026 growth target against your retention and CAC payback, not against last year’s plan.
  • Pricing model is tied to net retention — with caveats. Benchmarkit and ICONIQ both show higher NRR for usage-based models, but the signal has limits: ICONIQ’s gap holds only below $100M ARR (at $100M+ both sit at 117%), Benchmarkit’s usage-only cohort is just nine companies, and High Alpha’s pricing cut crowns hybrid models (detailed in Section 6). If you run seat-based pricing and NRR is stuck below 100%, pricing architecture belongs on your 2026 agenda, not just your sales motion.
  • “Median NRR” is a trap. The single most-cited SaaS metric is defined three incompatible ways across these reports. Use the wrong one and a healthy company looks broken, or a broken one looks healthy. Match the formula before you benchmark (Section 5).

How to use this report

  1. Start with Section 2. Use the modeled median as a planning baseline and P25/P75 as directional performance boundaries.
  2. Check the confidence badge. A/B cells have broader comparable support; C/D cells require more reliance on the source-specific tables.
  3. Match the definition. For NRR, CAC payback, and Rule of 40, use the provider spans and confirm that your formula matches before benchmarking.
  4. Use ACV when go-to-market structure matters. A company with $5K ACV should not rely only on the same ARR benchmark as a company with $100K ACV.

2. The SaaS CFO Normalized Benchmarks

This is the planning table. It translates the source-specific bands in Section 3 into six standard ARR bands, so you get one usable baseline even where no single provider covers your stage. The raw provider tables remain the authoritative record. Every value below is a normalized estimate or a clearly labeled reference span, not a newly observed survey statistic.

MetricStatistic<$1M$1–5M$5–20M$20–50M$50–100M$100M+
ARR GrowthNormalized estimatesModeled P2520%13%11%10%10%10%
Modeled median36% B29% B25% A19% A18% B15% C
Modeled P75100%50%50%31%30%29%
Gross Revenue RetentionNormalized estimatesModeled P2575%77%81%80%85%82%
Modeled median84% B84% B87% B87% B89% B89% B
Modeled P7595%93%94%93%91%94%
Net Revenue RetentionSource spans; mixed definitionsP25 source span75–78%75–91%93–95%91–98%97–100%95–97%
Median source span94–100% C94–104% C102–103% C101–103% C101–104% C101–103% C
P75 source span110–116%110%108–115%110–112%108%108–115%
CAC PaybackSource spans; months; lower is betterP25 source span2–105–108–118–1113–256–13
Median source span5–18 C8–18 C14–17 C10–20 C17–25 C10–17 C
P75 source span8–2314–2322–2416–2722–3422
Software Gross MarginNormalized estimatesModeled P2555%55%71%71%77%70%
Modeled median73% B75% B80% B78% C83% B79% C
Modeled P7580%83%86%84%87%83%
ARR per EmployeeNormalized estimatesModeled P25$56K$78K$120K$194K$154K$148K
Modeled median$85K B$100K B$155K A$268K A$253K B$253K C
Modeled P75$138K$188K$222K$332K$344K$397K
S&M% of revenue onlyModeled P2525%25%28%21%27%—
Modeled median38% C38% C41% C34% C38% C37% C
Modeled P7540%40%48%50%53%—
R&D% of revenue onlyModeled P2530%30%23%17%16%21%
Modeled median38% C38% C33% C27% C25% C23% C
Modeled P7550%50%47%27%30%39%
Rule of 40Source spans; mixed margin basisP25 source span7–25%7–10%−3%11–22%15–18%12–15%
Median source span30–46% D30–33% D8–20% D24–35% D25–30% D30–37% D
P75 source span44–98%44–80%32–35%41%38–52%38–83%
  • Modeled percentile rows: for sufficiently comparable metrics, P25, median, and P75 are estimated separately after mapping each provider’s native bands to the six standard bands. They are planning benchmarks, not pooled survey percentiles.
  • Source-span rows: NRR, CAC payback, and Rule of 40 show the low-to-high provider span at each percentile. They are deliberately not averaged because their formulas are not reconciled.
  • CAC direction: lower payback is better. Its P25 row therefore represents faster/top-quartile performance, while P75 represents slower/bottom-quartile performance.
  • Confidence: grades shown on the median row apply to the complete three-row cell group. A = strong multi-source band match; B = two or more usable sources with limited mapping; C = broad-band, single-source, or formula uncertainty; D = incompatible definitions, directional only.
  • Population: core values use broad-private surveys. ICONIQ remains an enterprise/top-quartile overlay and is excluded from modeled estimates and reference spans.
Use this as a planning guide, not a valuation formula. A company should still choose the raw provider cohort that best matches its ACV, go-to-market motion, customer mix, and metric definition. Appendix C documents the model.

3. Source Benchmarks by ARR Band

This section is the audit trail behind Section 2. Each metric is shown in one combined table. Every native ARR band published by a source has its own column header, even when bands overlap. Values appear only under the source’s exact published band; unreported intersections are blank. Tables scroll horizontally, with Provider and Period / Statistic pinned for readability.

Read the ICONIQ rows carefully. Each ICONIQ row names its edition. The State of Software 2025 rows are pooled longitudinally: every company-quarter from 2013 through Q2 2025 at that scale, whenever it occurred. They mix ZIRP-era and post-2022 observations, so they run hot versus the single-year surveys in the same tables. ICONIQ publishes no current-period quartile grid at this band granularity. For ICONIQ’s current-period (2026) values — year-cohorted medians and top quartiles split <$100M versus $100M+ — the State of Scaling 2026 Pacesetter Index rows now appear in the tables below under their native bands (<$10M, $10–25M, $25–100M, $100M+), and the broad-portfolio 2026 values (split only <$100M versus $100M+) are in Appendix A4. The two editions always sit in separate rows. Treat the Pacesetter rows as an aspirational reference, not a market benchmark: a filtered cohort of top-quartile-growth, AI-forward companies with very small N, rounded to the nearest multiple of 5.

ARR Growth (YoY %)

ProviderPeriod / statistic<$1M$1–3M$1–5M$3–5M<$5M$5–10M$5–20M<$10M$10–20M$10–25M$20–50M>$20M$25–50M$25–100M>$50M$50–100M>$100M$100–250M$250–500M>$500M
BenchmarkitCY2025
Median [25th–75th]
————21% [2–50%]—26% [11–50%]———17% [5–29%]————19% [10–30%]10% [5–25%]———
High Alpha2025 / Q2 2025
Median [25th–75th]
100% [29–300%]—50% [24–100%]———31% [15–72%]———30% [16–41%]———16% [10–29%]—————
SaaS Capital2024 actual
Median [25th–75th]
40% [20–100%]28% [14–50%]—24% [10–40%]—24% [13–42%]——20% [8–37%]——20% [10–31%]————————
KeyBanc2024 actual
Median
———————31%—17%——16%—14%—————
ICONIQState of Software 2025 ed.
Pooled 2013–Q2 2025 (longitudinal)
Top quartile
———————517%—173%——104%——91%—60%50%60%
ICONIQState of Scaling 2026 ed.
Pacesetter cohort, 2024–Q2 2026
Median / top quartile
———————900% / 2600%—430% / 575%———190% / 300%——115% / 165%———

Net Revenue Retention (NRR)

ProviderPeriod / statistic<$1M$1–5M<$5M$5–20M<$10M$10–25M$20–50M$25–50M$25–100M>$50M$50–100M>$100M$100–250M$250–500M>$500M
BenchmarkitCY2025
Median [25th–75th]
Formula undisclosed
——94% [75–110%]102% [93–108%]——101% [91–112%]———104% [100–108%]103% [95–115%]———
High Alpha2025 / Q2 2025
Cohort median [25th–75th]
100% [78–116%]104% [91–110%]—103% [95–115%]——103% [98–110%]——101% [97–108%]—————
ICONIQState of Software 2025 ed.
Pooled 2013–Q2 2025 (longitudinal)
Average-base NDR, top quartile
————124%123%—120%——122%—114%115%123%
ICONIQState of Scaling 2026 ed.
Pacesetter cohort, 2024–Q2 2026
Net $ retention; median / top quartile
————105% / 115%125% / 140%——130% / 140%——115% / 130%———

Gross Revenue Retention (GRR)

ProviderPeriod / statistic<$1M$1–5M<$5M$5–20M<$10M$10–25M$20–50M$25–50M$25–100M>$50M$50–100M>$100M$100–250M$250–500M>$500M
BenchmarkitCY2025
Median [25th–75th]
Formula undisclosed
——76% [70–90%]85% [79–92%]——83% [75–90%]———89% [86–91%]89% [79–98%]———
High Alpha2025 / Q2 2025
Median [25th–75th]
92% [80–100%]92% [83–95%]—88% [82–95%]——90% [85–95%]——88% [84–90%]—————
ICONIQState of Software 2025 ed.
Pooled 2013–Q2 2025 (longitudinal)
GDR, top quartile
————100%96%—95%——94%—94%93%97%
ICONIQState of Scaling 2026 ed.
Pacesetter cohort, 2024–Q2 2026
Gross $ retention; median / top quartile
————100% / 100%95% / 100%——95% / 95%——90% / 95%———

CAC Payback (months)

ProviderPeriod / statistic<$1M$1–5M<$5M$5–20M$20–50M<$50M>$50M$50–100M>$100M$100–200M$200M+
BenchmarkitCY2025
Median [25th–75th]
Formula undisclosed
——18 [10–23]17 [11–24]10 [8–16]——25 [25–34]10 [6–22]——
High Alpha2025 / Q2 2025
GM-adjusted median [25th–75th]
5 [2–8]8 [5–14]—14 [8–22]20 [11–27]—17 [13–22]————
ICONIQState of Software 2025 ed.
Pooled 2013–Q2 2025 (longitudinal)
Top quartile (= 25th percentile; faster is better)
—————12.9—12.3—13.314.8

Gross Margin (%): keep source bases separate

ProviderPeriod / statistic<$1M$1–5M<$5M$5–20M<$10M$10–25M$20–50M$25–50M$25–100M>$50M$50–100M>$100M$100–250M$250–500M>$500M
BenchmarkitCY2025
Software GM median [25th–75th]
——72% [50–80%]80% [72–86%]——————86% [84–90%]————
High Alpha2025 / Q2 2025
Software GM median [25th–75th]
74% [60–80%]77% [60–85%]—80% [70–86%]——78% [71–84%]——79% [70–83%]—————
ICONIQState of Software 2025 ed.
Pooled 2013–Q2 2025 (longitudinal)
Annual gross profit margin, basis undisclosed; top quartile
————86%84%—78%——80%—84%86%85%
ICONIQState of Scaling 2026 ed.
Pacesetter cohort, 2024–Q2 2026
Median / top quartile
————55% / 80%60% / 70%——80% / 85%——75% / 85%———

ARR per Employee

ProviderPeriod / statistic<$1M$1–5M<$5M$5–20M<$10M$10–25M$20–50M$25–50M$25–100M>$50M$50–100M>$100M$100–250M$250–500M>$500M
BenchmarkitCY2025
Median [25th–75th]
——$100K [$70–175K]$155K [$110–222K]——$282K [$231–313K]———$184K [$159–291K]~$206K*———
High Alpha2025 / Q2 2025
Median [25th–75th]
$56K [$42–100K]$136K [$86–200K]—$167K [$130–221K]——$268K [$157–350K]——$278K [$148–397K]—————
KeyBanc2024 actual
Median
————$85K$154K—$192K—$253K—————
ICONIQState of Software 2025 ed.
Pooled 2013–Q2 2025 (longitudinal)
Top quartile
————$90K$160K—$215K——$220K—$250K$280K$380K
ICONIQState of Scaling 2026 ed.
Pacesetter cohort, 2024–Q2 2026
ARR or revenue per FTE; median / top quartile
————$75K / $90K$115K / $145K——$225K / $355K——$655K / $890K———

*Benchmarkit’s >$100M figure is from the PDF narrative only; the Quartile-tab result is unavailable (n=4).

S&M % of Revenue / Ending ARR

ProviderPeriod / denominator<$1M$1–5M<$5M$5–20M<$10M$10–25M$20–50M$25–50M>$50M$50–100M>$100M$100–250M$250–500M>$500M
BenchmarkitCY2025
% of revenue; median [25th–75th]
——35% [25–40%]41% [28–48%]——30% [21–50%]——39% [27–53%]————
High Alpha2025 / Q2 2025
% of ending ARR; median [25th–75th]
28% [20–50%]30% [20–50%]—30% [22–53%]——29% [24–34%]—25% [20–40%]—————
KeyBanc2024 actual
% of revenue; median
————41%41%—38%37%—————
ICONIQState of Software 2025 ed.
Pooled 2013–Q2 2025 (longitudinal)
% of revenue; median
————106%73%—55%—57%—47%43%35%

R&D % of Revenue / Ending ARR

ProviderPeriod / denominator<$1M$1–5M<$5M$5–20M<$10M$10–25M$20–50M$25–50M>$50M$50–100M>$100M$100–250M$250–500M>$500M
BenchmarkitCY2025
% of revenue; median [25th–75th]
——32% [30–50%]30% [23–47%]——22% [17–27%]——26% [16–30%]22% [21–39%]———
High Alpha2025 / Q2 2025
% of ending ARR; median [25th–75th]
43% [23–76%]38% [24–50%]—31% [20–54%]——32% [23–40%]—30% [25–40%]—————
KeyBanc2024 actual
% of revenue; median
————43%29%—33%24%—————
ICONIQState of Software 2025 ed.
Pooled 2013–Q2 2025 (longitudinal)
% of revenue; median
————114%56%—41%—34%—27%25%28%

Rule of 40 (%)

ProviderPeriod / margin basis<$1M$1–5M<$5M$5–20M<$10M$10–25M$20–50M$25–50M>$50M$50–100M>$100M$100–250M$250–500M>$500M
BenchmarkitCY2025
Basis undisclosed; median [25th–75th]
——30% [7–44%]8% [−3–32%]——35% [22–41%]——25% [18–52%]37% [12–83%]———
High Alpha2025 / Q2 2025
FCF or EBITDA; median [25th–75th]
46% [25–98%]33% [10–80%]—20% [−3–35%]——24% [11–41%]—30% [15–38%]—————
ICONIQState of Software 2025 ed.
Pooled 2013–Q2 2025 (longitudinal)
FCF; top quartile
————n/an/a—70%—60%—55%50%70%
  • “—” means the source does not report that exact ARR-band intersection, or the source cell was blank/insufficient; it is not a zero.
  • Overlapping band definitions such as <$5M, <$10M, >$50M, and >$100M remain separate and are never consolidated.
  • S&M and R&D denominators differ: High Alpha uses ending ARR; Benchmarkit, KeyBanc, and ICONIQ use revenue.
  • NRR and Rule of 40 are source-specific references, not consensus calculations. Formula and sample differences prevent a defensible blended median.

4. Core Benchmarks by ACV Band

ARR band tells you how big a company is. ACV tells you how it sells. A $5K ACV product and a $100K ACV product at the same ARR run on different retention physics and different CAC math, so benchmark both ways. Benchmarkit is the only provider in the set that publishes full quartiles strictly by ACV (ICONIQ reports a top-quartile CAC-payback ACV cut, noted in Appendix A4). The first table is a modeled planning layer that fills sparse cells and smooths isolated reversals. The second preserves the published Benchmarkit quartiles as the audit trail.

Normalized ACV planning benchmarks

Each metric is modeled separately at P25, median, and P75. Values are not observed survey percentiles and are not cross-provider consensus figures. Use them to select a coherent planning benchmark when the published ACV cuts are sparse or noisy.

MetricStatistic<$5K$5–10K$10–25K$25–50K$50–100K>$100K
Gross Retention (GRR)Modeled P2574%74%74%79%79%81%
Modeled median C80%81%81%85%89%89%
Modeled P7585%88%88%94%95%95%
Net Retention (NRR)Modeled P2590%90%90%90%94%100%
Modeled median C96%96%96%104%104%106%
Modeled P75102%102%105%109%109%144%
CAC PaybackModeled P259 mo9 mo10 mo12 mo12 mo12 mo
Modeled median C11 mo12 mo17 mo17 mo20 mo20 mo
Modeled P7517 mo20 mo27 mo27 mo27 mo27 mo
Blended CAC RatioModeled P25$0.53$0.67$1.17$1.17$1.60$1.72†
Modeled median C$0.80$1.20$1.30$1.46$1.71$1.83†
Modeled P75$1.28$1.28$1.78$1.80$1.80$1.94†
New Name CAC RatioModeled P25$0.92$1.03†$1.11$1.26$1.76$1.76
Modeled median C$1.17$1.41†$1.60$1.62$2.11$2.11
Modeled P75$1.85$2.23†$2.53$2.53$2.60$2.60
LTV to CACModeled P253.1x2.9x2.9x3.1x3.3x3.6x
Modeled median C4.5x3.9x3.8x4.0x4.7x5.7x
Modeled P757.3x6.8x6.8x7.3x8.3x9.8x
Magic NumberModeled P250.280.43†0.570.710.881.10
Modeled median C0.380.61†0.841.081.341.67
Modeled P751.151.53†1.862.092.222.33
  • Interpretation: P25 and P75 are distribution positions, not universal performance labels. Lower is generally better for CAC payback and CAC ratios; higher is generally better for retention, LTV:CAC, and Magic Number.
  • Confidence: C reflects a single-provider modeled curve. A dagger (†) is confidence D because it replaces a blank or insufficient source cell rather than smoothing an observed value.
  • Method: GRR, NRR, CAC payback, and CAC ratios use sample-size-weighted monotonic smoothing. LTV:CAC and Magic Number use a weighted low-curvature trend because neither metric has a defensible one-way relationship with ACV. Appendix C provides the full rules.

Published Benchmarkit ACV quartiles (audit trail)

Values below are the published median [25th percentile–75th percentile] and are intentionally left unsmoothed.

Metric<$5K$5–10K$10–25K$25–50K$50–100K>$100K
Gross Retention (GRR)80% [74–85%]83% [75–89%]79% [73–87%]85% [79–94%]91% [79–96%]88% [81–95%]
Net Retention (NRR)98% [90–102%]95% [93–102%]92% [89–105%]105% [89–111%]103% [94–107%]106% [100–144%]
CAC Payback (months)11 [10–17]12 [8–20]18 [10–34]17 [13–24]22 [15–27]18 [9–24]
Blended CAC Ratio$0.80 [0.53–1.30]$1.20 [0.67–1.25]$1.30 [1.20–1.78]$1.46 [1.15–1.81]$1.71 [1.60–1.80]—
New Name CAC Ratio$1.17 [0.92–1.85]—$1.60 [1.11–3.00]$1.62 [1.26–2.18]$2.81 [2.22–3.21]$1.14 [1.12–1.75]
LTV to CAC6.0x [4.0–9.4]2.0x [1.2–3.1]3.1x [2.8–7.0]4.0x [3.5–8.0]3.9x [2.7–6.0]7.0x [4.1–12.0x]
Magic Number0.40 [0.26–0.84]—1.00 [0.85–2.23]1.13 [0.64–2.40]1.03 [0.67–2.13]1.81 [1.20–2.25]

Source: Benchmarkit 2026 Quartile tab. “—” means blank or insufficient observations, including the >$100K blended-CAC cut (n=2). Other >$100K sample sizes: GRR 24; NRR 23; CAC payback 17; new-name CAC 5; LTV:CAC 12; magic number 20.

5. Read Before You Compare: The Definition Traps

These are not errors in the source reports. Each methodology is legitimate. The problem is comparing across them. Here is how to translate.

NRR is defined three different ways

ICONIQ uses an average-base denominator, while KeyBanc uses beginning-of-period ARR and High Alpha uses a cohort approach. These formulas are not interchangeable, and the direction of the resulting difference depends on the company’s ARR movement. ICONIQ’s higher reported results also reflect its enterprise-skewed, top-quartile sample. Before you benchmark your NRR, confirm which formula your reference report uses. If you compute NRR on a beginning-of-period base, compare to KeyBanc, not ICONIQ.

Rule of 40 uses different margin definitions

ICONIQ uses free-cash-flow margin. KeyBanc uses EBITDA margin. High Alpha permits respondents to use either free-cash-flow margin or EBITDA margin. Benchmarkit does not disclose its margin definition. The gap may reflect margin definition, reporting period, sample composition, and company scale. Benchmarkit’s undisclosed formula prevents a clean reconciliation. Compare only within a single margin definition, and only for companies above about $25M ARR. Below that, the metric carries little signal.

Magic number: gross vs. net, and which quarter’s spend

KeyBanc reports both gross (about 0.64) and net (about 0.5) magic number using full-period S&M. ICONIQ reports net only, using prior-quarter S&M. Benchmarkit reports a single figure (1.37) without stating gross or net. These are not interchangeable; a “1.37” and a “0.64” can describe the same sales efficiency.

6. A Pricing-Model Signal in Two Independent Sources

Pricing model is associated with net retention — directionally. Benchmarkit and ICONIQ both show higher NRR for usage-based models, though the signal comes with real caveats.
  • Benchmarkit: usage-based NRR 108% vs. seat-based 98% and hybrid (subscription + usage) 102%. A 10-point usage-vs-seat gap, with seat-based below the break-even line — but the usage-only cohort is just n=9 (hybrid n=35, seat n=44), and usage-only gross retention (83% median) trails non-seat subscription (90%) and hybrid (85%).
  • ICONIQ: usage-based top-quartile net dollar retention 135% vs. subscription 121% — but only below $100M ARR. At $100M+ the gap disappears entirely: 117% for both.
  • High Alpha: its pricing cut points a different direction — hybrid models show the highest retention, not pure usage.
Because the direction shows up in two datasets but weakens at scale, rests on small usage-only samples, and a third source favors hybrid, I treat it as a directional signal worth testing, not an established relationship. Weigh pricing architecture alongside customer profile, ACV, product usage, and contract structure. But if I ran a seat-based product with NRR under 100% today, this is the chart I would bring to my next pricing review.

7. The 2026 Macro Story

Efficiency improved alongside tighter spending

Across sources, 2025 was the strongest efficiency year in recent memory. Benchmarkit shows CAC payback improving from 18 to 16 months, Rule of 40 rising from 15% to 25% (its biggest single-year gain in five years), and ARR per employee climbing 20 to 29% to about $190K. Over the same period, R&D fell from 35% to 27% of revenue, and gross retention gave back about 4 points to land in the mid-80s. Read those two trends together. The efficiency gains arrived alongside the spending cuts, and the data cannot prove the cuts caused them. What it can tell you: the gains were not free. I would not bank on a repeat in 2026 without a plan that protects retention.

Growth has structurally decelerated

Benchmarkit’s multi-year series shows a steady slide: median growth down from about 30% in 2022 to about 20% in 2025, and top-quartile growth falling even harder — its 75th percentile dropped from 75% to 42%. KeyBanc’s history matches through 2024 (31%, 20%, 15%) but its respondents expect a rebound: 20% in 2025 and 25% in 2026, which would break the streak. The pattern is consistent: companies with stronger unit economics have more capacity to reinvest in go-to-market and sustain growth. Efficiency and growth have converged into one question: do your underlying economics support reinvestment?

AI-native companies show a different early operating profile

ICONIQ shows AI-native companies scaling faster with leaner teams and more R&D-heavy headcount; Battery’s target ranges support the faster-growth part (5–10x ARR growth versus 2–3x for traditional SaaS) and the margin trade-off (20–40% gross margin targets), but say nothing about headcount or retention. The trade-offs: lower near-term margins, because inference compute sits in COGS, and in some cuts weaker gross retention as buyers evaluate incumbents and AI challengers in parallel. It is too early to call these durable benchmarks. But the divergence is real, and it deserves tracking as its own cohort.

8. Put These Numbers to Work

Benchmarks are only useful if they change a number in your plan. Here is how I would use this report to build or pressure-test a 2026 operating plan.

  • Pick your cohort first. Find your ARR band in Section 2 and your ACV band in Section 4. If the two tell different stories, trust the ACV cut for retention and CAC math; contract size often shapes those dynamics more than company size.
  • Plan to the modeled median. Manage to the quartiles. The modeled median is a defensible baseline for a board conversation. P25 and P75 are your guardrails. Watch the direction: for CAC payback and CAC ratios, lower is better.
  • Check the definition before you compare. For NRR, CAC payback, Rule of 40, and magic number, confirm your formula matches the source formula (Section 5). This is the step most teams skip, and it is the one that produces bad conclusions.
  • Sanity-check your growth target. Benchmarkit respondents planned 26% growth for CY-26 after delivering 20% in CY-25. That is four straight years of over-optimism. If your 2026 target sits more than a few points above your band’s median, your retention and CAC payback need to justify it.
  • Watch retention like revenue. The market’s 2025 efficiency gains coincided with roughly a four-point drop in gross retention — the data cannot prove one paid for the other, but they arrived together. GRR is the cheapest growth you have. Put a number on it in the plan, assign an owner, and review it monthly.

The appendices behind this page hold every source table, the report-by-report deep-dives, and the full normalization methodology. Nothing in this report is a black box. If a number matters to your plan, you can trace it to its source.

Appendix A: Per-Source Backing (No Black Boxes)

Section 3 is the canonical audit trail: it preserves each provider’s native ARR bands, period, statistic type, and metric basis. The table below summarizes which source can be used for each purpose without implying comparability where none exists.

SourceNative periodUse in this reportDo not combine blindly with
BenchmarkitCY2025Broad-private medians; full ARR and ACV quartilesDefinition-sensitive metrics where formulas are undisclosed
High Alpha2025 / Q2 2025Broad-private medians and quartiles by ARR bandRevenue-based S&M/R&D; its denominator is ending ARR
SaaS Capital2024 actualBroad-private ARR-growth anchorUnit-economics or margin metrics it does not publish
KeyBanc2024 actualPrivate-company medians; explicit metric definitions2025 estimates or differently defined retention metrics
ICONIQ (State of Software 2025)Pooled observations, 2013–Q2 2025Enterprise / top-quartile referenceBroad-private population medians
ICONIQ (State of Scaling 2026)Year-cohorted, 2022–Q2 2026Current-period enterprise reference; Pacesetter cohortBroad-private population medians; fine ARR-band quartiles
MaxioWinter 2026 editionDirectional, billings-based mean growthMedian or quartile distributions
Battery2025AI-native strategic context and target rangesObserved survey percentiles or population medians

No sample-weighted cross-provider average is calculated. Section 2 uses documented equal-provider medians only for sufficiently comparable metrics; otherwise results remain source-specific. Differences in periods, formulas, denominators, statistic types, and native bands are preserved rather than silently reconciled.

Appendix A2: Benchmarkit Deep-Dive (Full Cuts)

Benchmarkit is the most granular report in the set (n=342, cut six ways). Most of its data does not fit an ARR-band grid; it is cut by growth rate, ACV, GTM motion, pricing model, and solution type. I captured those cuts here so none of the report is lost. All figures are CY-25 unless a trend is shown. Note: Benchmarkit discloses no metric formulas (see the transparency note in Appendix B).

What to steal from Benchmarkit: the ACV and pricing-model cuts no one else publishes. The number I would carry into a pricing review: usage-based NRR of 108% versus seat-based at 98%. The caution: no formulas are disclosed, so treat cross-report comparisons of its NRR, Rule of 40, and CAC payback as directional.

Total-population medians & quartiles, with YoY trend

MetricNP25Median (CY-25)P75Trend (period as labeled)
ARR growth338~8%20%42%30%(CY-22)→20%
NRR23092%102%110%P75 110% (workbook CY-25; PDF narrative cites 108%); P25 95→92%
GRR22676% (min 33%)84%91%88%→84% (−4pts)
Blended CAC ratio122$1.08$1.30$1.78$1.40→$1.30
New-name CAC ratio128$1.14$1.63$2.22$2.00→$1.63 (−19%)
Expansion CAC ratio74$0.46$0.80$1.70$1.00→$0.80
Magic number1320.681.372.140.94→1.37
Rule of 40110+7%25%43%+15%→25% (+10pts)
Gross margin (total-rev)22868%76%84%+77%→76%
Gross margin (software)23272%80%86%+stable 79–81%
ARR / employee96$126K$193K ($175K in exec summary)$279K$150K→$193K (+29%)
CAC payback (mo)Published report: 19810162418→16; min 2, max 48

P25 and P75 are statistical positions, not universal performance labels. Lower values are generally better for CAC payback and CAC ratios; higher values are generally better for retention, growth, margins, magic number, Rule of 40, and ARR per employee.

Sample-size note: the N column reports the published report’s question-level sample sizes. The supplied Quartile-tab counts are consistently lower (e.g., ARR growth 338 vs 169; NRR 230 vs 111; GRR 226 vs 108; CAC payback 198 vs 81), and the difference is unresolved; all quartile values above were validated against the Quartile tab, which is treated as authoritative. S&M, R&D, and G&A spend ratios shown separately below.

S&M as % of revenue: total population

NP25Median (CY-25)P75YoY
19625%35%47%37%→35% (first YoY decrease; min 12%, max 121%)

R&D as % of revenue: total population

NP25Median (CY-25)P75YoY
19222%27%40%35%→27% (−8pts, AI productivity; floor 10%)

G&A as % of revenue: total population

NP25Median (CY-25)P75YoY
17611%17%22%CY-22 24%→CY-25 17% (−7pts over 3 yrs; min 5%, max 44%)

Benchmarkit by ARR BAND (median [25th–75th])

Validated against the supplied Benchmarkit workbook Quartile tab. “—” means the workbook was blank or had too few observations; it is not a zero. Sample sizes vary by metric and band.

Metric<$5M$5–20M$20–50M$50–100M>$100M
Growth (N=338)21% [2–50]26% [11–50]17% [5–29]19% [10–30]10% [5–25]
NRR (N=230)94% [75–110]102% [93–108]101% [91–112]104% [100–108]103% [95–115]
GRR (N=226)76% [70–90]85% [79–92]83% [75–90]89% [86–91]89% [79–98]
CAC payback mo18 [10–23]17 [11–24]10 [8–16]25 [25–34]10 [6–22]
Blended CAC ratio (N=122)$1.15 [1.06–1.81]$1.20 [0.86–1.69]$1.30 [1.25–2.00]——
New-name CAC ratio (N=128)$1.76 [1.30–3.00]$1.75 [1.40–2.22]$1.63 [1.12–2.04]——
LTV:CAC (N=146)4.6x [2.7–10.6]5.3x [3.8–7.7]3.2x [2.2–7.2]3.1x [2.0–4.9]8.0x [3.6–11.0]
Magic number (N=132)1.38 [0.90–1.58]1.37 [0.55–2.14]1.81 [0.85–2.67]0.80 [0.64–1.61]0.84 [0.68–1.20]
Software GM (N=232)72% [50–80]80% [72–86]—86% [84–90]—
Total-revenue GM (N=228)70% [56–80]77% [69–83]81% [66–85]80% [72–86]72% [71–79]
ARR/employee (N=96)$100K [$70–175K]$155K [$110–222K]$282K [$231–313K]$184K [$159–291K]~$206K*
S&M % rev (N=196)35% [25–40]41% [28–48]30% [21–50]39% [27–53]—
R&D % rev (N=192)32% [30–50]30% [23–47]22% [17–27]26% [16–30]22% [21–39]
G&A % rev (N=176)22% [20–28]20% [13–25]15% [7–20]16% [5–17]12% [11–18]
Rule of 40 (N=110; formula undisclosed)30% [7–44]8% [−3–32]35% [22–41]25% [18–52]37% [12–83]
Expansion ARR share (N=114)<$20M: 35–39%—62%

*The >$100M ARR-per-employee figure is from the PDF narrative only; the Quartile-tab result is unavailable (n=4).

Benchmarkit by ACV BAND (median [25th–75th])

Metric<$5K$5–10K$10–25K$25–50K$50–100K>$100K
NRR (N=230)98% [90–102]95% [93–102]92% [89–105]105% [89–111]103% [94–107]106% [100–144]
GRR (N=226)80% [74–85]83% [75–89]79% [73–87]85% [79–94]91% [79–96]88% [81–95]
CAC payback mo (N=198)11 [10–17]12 [8–20]18 [10–34]17 [13–24]22 [15–27]18 [9–24]
Blended CAC ratio$0.80 [0.53–1.30]$1.20 [0.67–1.25]$1.30 [1.20–1.78]$1.46 [1.15–1.81]$1.71 [1.60–1.80]—
New-name CAC ratio$1.17 [0.92–1.85]—$1.60 [1.11–3.00]$1.62 [1.26–2.18]$2.81 [2.22–3.21]$1.14 [1.12–1.75]
LTV to CAC6.0x [4.0–9.4]2.0x [1.2–3.1]3.1x [2.8–7.0]4.0x [3.5–8.0]3.9x [2.7–6.0]7.0x [4.1–12.0]
SaaS Magic Number0.40 [0.26–0.84]—1.00 [0.85–2.23]1.13 [0.64–2.40]1.03 [0.67–2.13]1.81 [1.20–2.25]

The >$100K blended-CAC cell is omitted because the workbook has only two observations. “—” elsewhere likewise denotes unavailable or insufficient data.

Benchmarkit by GROWTH-RATE cohort (median [25th–75th])

Metric>50% growth31–50%21–30%11–20%<10%
NRR (N=230)111% [95–155]105% [91–115]101% [100–107]103% [98–108]92% [83–101]
GRR (N=226)86% [76–94]79% [71–90]89% [82–95]89% [85–96]79% [72–86]
CAC payback mo (N=198)10 [7–18]14 [12–18]22 [15–25]18 [13–33]14 [8–23]
Rule of 40 (N=110)57 [23–125]8 [-13–30]35 [12–43]18 [-3–31]19 [12–55]
Magic number (N=132)2.40————
Blended CAC ratio (N=122)$1.20————
New-name CAC ratio (N=128)$1.40——$1.85$1.76
Gross margin (N=228)74%—78% (highest)——
ARR/employee (N=96)$235K$136K (lowest)———
G&A % (N=176)19%21% (highest)——12%

Benchmarkit S&M & R&D % by growth-rate cohort (median)

Metric>50% growth31–50%<10%
S&M % revenue35%44% (highest)—
R&D % revenue33% (highest)—26%

S&M and R&D are total-population quartile metrics in the source (shown above); only limited cohort points are charted.

Benchmarkit by GTM motion & solution type

MetricSales-ledPLGHybridVertical SaaSHorizontal B2B
GRR (median)89%79%80%——
CAC payback (median, months)———1814
CLTV:CAC (median)——6.8x5.6x4.1x
Blended CAC ratio (median)——$0.86——
New-name CAC ratio (median)———$2.22$1.62
Expansion ARR share (median)—36%54%——

Benchmarkit by PRICING model (the headline cut)

MetricUsage-basedSeat-basedSubscription+Usage (hybrid)
NRR (median)108%98%102% (n=35)
Gross margin (median)62% (lowest)subscription variants 76–84%
Rule of 40 (75th percentile)Non-seat subscription lowest median 18%43% (highest top quartile)
ARR/employee (median)$291K (highest)——

Benchmarkit: growth by solution type (N=338)

SegmentMedian growth (CY-25)
Vertical SaaS & AI24%
Horizontal B2B SaaS17–19%
AI-native applications17–19%

AI-native apps face stiffer competition than expected as LLM access lowers barriers to entry. The “build it ourselves” option and incumbent AI features compress growth. A vertical approach may be the strongest initial GTM strategy.

Benchmarkit: expansion ARR & G&A summary

MetricNMedianTop quartileYoY
Expansion ARR as % of New ARR11440%59%stable CY-24/25; 25th pct: 10%(CY-22)→29%(CY-25)
G&A as % of revenue17617%≤11%CY-22 24% → CY-25 17%
Burn multiple62Highest-burn medians sit in the early bands ($5–20M at 1.3x; sub-$5M at 1.0x median but 2.4x P75); scale drives improvement. Aligns with Ro40/S&M/R&D trends.

Expansion ARR share runs 62% above $100M ARR versus 35 to 39% below $20M. G&A target for scaled SaaS: 8 to 12%. When expansion crosses 40% of new ARR, it may signal substitution for new-logo weakness. Respondents planned 26% median growth for CY-26 against 20% actual in CY-25. That is four straight years of over-optimism.

Benchmarkit: AI monetization & pricing architecture

MetricValueN
Companies charging for AI29% (67% do not charge; 4% other)156
AI introduced in last 12 months35%254
Planning to add AI in 202624%254
No plans for AI6%254
AI expenses as % of R&D10% median (top quartile: 30%+)46 (directional)

Among the 29% who charge, AI pricing model split (N=156):

AI Pricing ModelShare
Subscription + Usage (hybrid)39%
Subscription only (seat-based)26%
Subscription only (non-seat)23%
Usage-only13%

By GTM: PLG companies strongly prefer usage-only AI pricing (60%); Sales-led favor Subscription + Usage (36%); Hybrid split more evenly with Sub+Usage at 50%. Despite market buzz, pure usage-based AI pricing remains a minority approach. Revenue predictability concerns and buyer resistance to open-ended costs are the primary restraints.

Appendix A3: High Alpha Deep-Dive (Full Cuts)

n=800+. Native ARR bands: <$1M, $1–5M, $5–20M, $20–50M, >$50M. All figures are median [25th–75th]. High Alpha is the only source that publishes retention-tier and NRR by CAC-payback matrix cuts.

What to steal from High Alpha: the matrix below. Companies with NRR above 106% and CAC payback under 10 months grew 71% at the median. Companies weak on both grew 10%. Retention plus efficiency is the whole game, and this is the one table that shows it.

High Alpha: full metric by ARR band (median [P25–P75])

Metric<$1M$1–5M$5–20M$20–50M>$50M
Growth100 [29–300]50 [24–100]31 [15–72]30 [16–41]16 [10–29]
NRR100 [78–116]104 [91–110]103 [95–115]103 [98–110]101 [97–108]
GRR92 [80–100]92 [83–95]88 [82–95]90 [85–95]88 [84–90]
CAC payback (mo)5 [2–8]8 [5–14]14 [8–22]20 [11–27]17 [13–22]
Rule of 4046 [25–98]33 [10–80]20 [-3–35]24 [11–41]30 [15–38]
Software GM74 [60–80]77 [60–85]80 [70–86]78 [71–84]79 [70–83]
ARR/employee$56K [42–100]$136K [86–200]$167K [130–221]$268K [157–350]$278K [148–397]
S&M % ARR28 [20–50]30 [20–50]30 [22–53]29 [24–34]25 [20–40]
R&D % ARR43 [23–76]38 [24–50]31 [20–54]32 [23–40]30 [25–40]
Employees (median)92266131361

High Alpha: growth by retention tier (unique cut)

TierLowMediumHigh
Growth by GRR tier30% (<85%)25% (85–93%)60% (>93%)
Growth by NRR tier20% (<98%)30% (98–106%)54% (>106%)

High Alpha: median growth by the NRR by CAC-payback matrix (the standout cut)

NRR tierLow CAC (<10mo)Med CAC (10–15mo)High CAC (>15mo)
High NRR (>106%)71%53%40%
Med NRR (98–106%)45%50%25%
Low NRR (<98%)30%20%10%

Rule of 40 by the same matrix: high-NRR + low-CAC = 47%; low-NRR + high-CAC = 5%. In this cut, the combination of efficiency and retention is associated with stronger outcomes than either dimension alone. Monthly burn by band also reported (<$1M $50K median → >$50M $0K).

Appendix A4: ICONIQ Deep-Dive (Full Cuts, Top Quartile, Enterprise-Skewed)

This appendix covers two ICONIQ editions. The tables immediately below come from State of Software 2025 (September 2025), and the State of Scaling 2026 update at the end of the appendix carries ICONIQ’s current-period values (through Q2 2026). The 2025 edition: n=127 (117 private/portfolio + 10 public), pooled 2013 to Q2 2025 — a longitudinal reference, not a current-period snapshot. Native 7-band ARR schema. All figures are top quartile (75th percentile) unless noted. Read them as best-in-class, not medians. N-sizes per cell are in the source file.

What to steal from ICONIQ: the enterprise ceiling. Top-quartile gross dollar retention holds between 93% and 100% at every scale. That is what durable enterprise retention looks like. Do not benchmark your median against these figures; they are the 75th percentile of an enterprise-skewed sample.

ICONIQ: top quartile by 7-band ARR schema

Metric<$10M$10–25M$25–50M$50–100M$100–250M$250–500M>$500M
YoY growth517%173%104%91%60%50%60%
NDR (avg-base)124%123%120%122%114%115%123%
GDR100%96%95%94%94%93%97%
Net magic number2.3x1.5x1.2x1.5x1.1x1.0x1.6x
Rule of 40 (FCF)n/a*n/a*70%60%55%50%70%
Gross margin86%84%78%80%84%86%85%
ARR/FTE$90K$160K$215K$220K$250K$280K$380K

*ICONIQ deems Rule of 40 “less relevant” below about $25M ARR. CAC payback (top quartile = 25th percentile, since faster is better): <$50M 12.9 months, $50–100M 12.3, $100–200M 13.3, $200M+ 14.8. By ACV: about 9.7 months under $10K, rising to 17 months at $150K+.

ICONIQ: OpEx by function, % of revenue (median, by ARR band)

Function<$10M$10–25M$25–50M$50–100M$100–250M$250–500M>$500M
S&M106%73%55%57%47%43%35%
R&D114%56%41%34%27%25%28%
G&A71%33%26%23%19%14%13%
Total OpEx291%162%122%114%93%82%76%

These are OpEx as a percent of revenue (median), not top quartile. The <$10M figures reflect deeply pre-scale, high-burn companies (total OpEx at 291% of revenue), so they are not comparable to broad-private percent-of-revenue at the low end. NDR by pricing: usage 135% top quartile versus subscription 121% top quartile (under $100M ARR).

ICONIQ 2026 update: State of Scaling, “The Great Sorting” (September 2026)

n=137 software companies, including 11 public companies selected on IPO performance criteria (ICONIQ also reports the mix as 76% private / 24% public), quarterly data through Q2 2026. This edition fixes the biggest limitation of the 2025 appendix: its performance charts are year-cohorted (2022–2023, 2024, 2025, 2026) rather than pooled back to 2013, so the 2026 column is a true current-period read. The trade-offs: scale splits are coarse (<$100M versus $100M+ ARR, companies under $10M excluded), values are medians and top quartiles rather than full P25/median/P75 grids, and 2026 cell sizes are small (roughly 30–100 company-quarters). N-sizes remain company-quarters, and the sample keeps its enterprise, top-quartile skew, so ICONIQ stays out of the Section 2 modeled centers.

What to steal from the 2026 edition: the sorting itself. Median NDR at $100M+ slipped to 99% in 2026 while ICONIQ’s AI-forward “Pacesetter” cohort holds a 130% median in 2026 (the Pacesetter Index figure below, 115%, is the 2024–Q2 2026 window). Public 20%+ growers trade at 9.2x NTM revenue versus 2.7x for everyone else, and FCF now makes up roughly 60% of the median public Rule of 40. Efficiency is no longer optional at scale, and the growth premium has concentrated in a small AI-forward group.

ICONIQ 2026: medians by year (broad portfolio, companies <$10M ARR excluded)

MetricCohort2022–2023202420252026 (thru Q2)
Net dollar retention (median)<$100M105%100%106%104%
$100M+107%102%102%99%
Gross margin (median)<$100M72%74%70%70%
$100M+78%79%79%78%
FCF margin (median)<$100M(81%)(51%)(51%)(51%)
$100M+3%8%11%13%
Total OpEx as % of revenue (median)<$100M159%133%130%128%
$100M+86%78%77%73%
Burn multiple (median, unprofitable cos.)<$100M2.5x1.9x1.7x2.2x
$100M+1.1x0.8x0.9x1.2x
New-logo share of gross new ARR (avg)<$100M60%56%55%57%
$100M+44%45%47%49%

2026 top-quartile NDR: 112% under $100M, 109% at $100M+. ARR per FTE, top quartile by range and year: $10–25M $139K → $196K → $203K → $149K; $25–100M $201K → $229K → $265K → $290K; $100M+ $315K → $398K → $488K → $478K (2022–2023 → 2024 → 2025 → 2026). OpEx mix in 2026 (S&M/R&D/G&A, average): 46/33/21 under $100M; 43/38/19 at $100M+. The 2026 edition does not republish CAC payback; the 2025 edition’s by-year series (top quartile: 11.9 months pre-2020, 17.3 in 2024, 15.9 in 2025) remains the latest ICONIQ read on that metric.

The ICONIQ Pacesetter Index (2024–Q2 2026, Pacesetters only)

Pacesetters are ICONIQ’s living cohort of top-quartile-growth, AI-Native or AI-Driven companies (60% AI-Native as of Q2 2026; median age six years; roughly 2–4 years to $100M ARR). Median / top quartile by ARR or revenue range; top quartile means 75th percentile except burn multiple, which uses the 25th:

Metric<$10M$10–25M$25–100M$100M+
YoY ARR or revenue growth900% / 2600%430% / 575%190% / 300%115% / 165%
Gross $ retention100% / 100%95% / 100%95% / 95%90% / 95%
Net $ retention105% / 115%125% / 140%130% / 140%115% / 130%
Gross margin55% / 80%60% / 70%80% / 85%75% / 85%
Net magic number3.4x / 7.7x1.1x / 1.5x1.2x / 2.7x2.2x / 6.5x
Burn multiple1.3x / 0.8x1.8x / 1.6x0.9x / 0.5x0.3x / 0.1x
ARR or revenue per FTE$75K / $90K$115K / $145K$225K / $355K$655K / $890K

Treat the Pacesetter Index as an aspirational reference, not a market benchmark: it is a filtered cohort with very small N (roughly 4–12 company-quarters in most 2026 cells), figures are rounded to the nearest multiple of 5 where applicable, and $100M+ Pacesetters run gross margins about 5 points below peers (73% versus 78%), which ICONIQ attributes to inference, infrastructure, and deployment costs. Pacesetters’ growth is new-logo-led: 70%+ of gross new ARR versus roughly half for everyone else.

Appendix A5: KeyBanc/Sapphire Deep-Dive (Full Cuts)

n=71 (59 for the ARR-band cuts). The only source reporting full 25th, median, and 75th percentiles for every year from 2022 to 2026E, plus cuts by ARR segment and growth rate. ARR segments: <$10M, $10–25M, $25–50M, >$50M.

What to steal from KeyBanc/Sapphire: the formulas and the trend lines. Its arithmetic shows the Rule of 40 recovery coming from EBITDA margin, not growth. The forward signal: respondents expect CAC payback to improve from 22 months in 2025E to 18 in 2026E.

KeyBanc: master KPI grid, median by year (25th–75th in source)

Metric2022202320242025E2026E
ARR growth31%20%15%20%25%
Gross retention86%85%86%89%91%
Net retention106%102%101%102%104%
Gross magic number0.650.660.640.700.82
CAC payback (mo)2425242218
Gross margin75%77%78%78%79%
EBITDA margin-47%-24%-12%-4%1%
Rule of 40 (EBITDA)-14%-3%6%13%27%
ARR/employee$119K$165K$182K$201K$236K

Share of companies meeting the Rule of 40 threshold: 11% in 2022, 5% in 2024, 20% in 2026E. In the reported arithmetic, the recovery comes primarily from higher EBITDA margin rather than higher growth.

KeyBanc: OpEx % of revenue, by ARR segment and by growth rate (median)

CutS&MR&DG&ATotal
<$10M41%43%31%115%
$10–25M41%29%25%94%
$25–50M38%33%23%93%
>$50M37%24%17%78%
Growth <10%35%29%21%85%
Growth 20–30%42%30%28%100%
Growth >30%56%41%30%128%

ARR per FTE by segment (2024): $85K under $10M, rising to $253K above $50M. By growth rate it is non-monotonic ($199K under 10% growth, $156K at 20 to 30%), flagged in the source as genuine but counterintuitive. Magic number formulas: gross = (new logo + expansion) / S&M; net deducts churn.

Appendix B: Sources & Methodology

SourceEditionnTypeRole in this report
Benchmarkit2026 SaaS & AI-Native342Broad-private surveyCore median; pricing-model cuts
KeyBanc / Sapphire16th Annual, Nov 202571Broad-private surveyCore median; full quartiles
High Alpha2025 (9th)800+Broad-private surveyCore median by ARR band
SaaS Capital2025 (14th)1,000+Broad-private surveyGrowth anchor (largest n)
ICONIQ GrowthState of Software 2025127Enterprise, top-quartileTop-quartile / late-stage view
ICONIQ GrowthState of GTM 2025205GTM subsetGTM funnel context
ICONIQ GrowthState of Scaling 2026 (Sep 2026)137Enterprise, top-quartile; year-cohortedCurrent-period (2026) ICONIQ overlay; Pacesetter reference
Maxio InstituteB2B Growth, Winter 20262,000+ (billings)Mean, billings-basedDirectional growth only
Battery VenturesState of AI 2025estimatesTarget rangesAI-native directional only

Normalization rules applied: (1) Segment before comparing. Always compare within ARR band. (2) Never average NRR, CAC payback, or Rule of 40 across unreconciled definitions; report labeled P25, median, and P75 provider spans. (3) Keep the source-specific values in Section 3 authoritative. The modeled table in Section 2 estimates each percentile separately only when definitions and denominators are sufficiently aligned; it is never sample-weighted. (4) Hold enterprise/top-quartile-only sources (ICONIQ) out of population estimates. (5) Keep mean/billings (Maxio) and target-range (Battery) sources directional, never in modeled percentiles. (6) Label every modeled percentile and confidence grade rather than presenting a synthetic distribution as an observed survey result.

Known source inconsistencies (reported, not silently resolved): Benchmarkit lists ARR-per-employee as both $175K (executive summary) and $193K (detail section), and seat-based NRR as both 95% and 98% in the same report; the workbook confirms 98%, so this report uses 98% and treats 95% as a PDF typo. Both ARR-per-employee values are shown. Maxio states AI-enhanced growth as 16% and 17% in different sections. KeyBanc’s total sample is n=71, but its ARR-band breakdown draws on the n=59 that answered that demographic (bands: <$10M n=8, $10–25M n=20, $25–50M n=13, >$50M n=18).

A note on transparency. Definition disclosure varies widely across these reports. High Alpha and KeyBanc/Sapphire publish full metric glossaries with explicit formulas. That is the gold standard. Benchmarkit, despite offering the richest segmentation, publishes results with no stated methodology for NRR, Rule of 40, magic number, or CAC payback; ICONIQ defines net dollar retention precisely but not gross. This is not a criticism of the data. It is a caution for anyone comparing across reports: where a formula is undisclosed, treat cross-report comparison of that metric as directional only.

Appendix C: Normalized Benchmark Methodology

Purpose. The normalized tables in Sections 2 and 4 are modeled planning layers built from the source-specific data. They do not replace the raw tables and do not claim to be pooled surveys.

ARR-band normalization (Section 2)

StepMethod
1. Standard bandsTranslate evidence into <$1M, $1–5M, $5–20M, $20–50M, $50–100M, and $100M+ ARR. No two target bands are combined into one table cell.
2. Eligible sourcesUse broad-private providers only. Benchmarkit and High Alpha contribute across most metrics; SaaS Capital contributes growth; KeyBanc contributes growth, ARR per employee, S&M, and R&D. ICONIQ, Maxio, and Battery are excluded from modeled centers.
3. Band mappingPrefer an exact native band. If absent, combine overlapping native bands by their logarithmic ARR width or use the narrowest published band that fully covers the target. Open-ended bands may support a narrower higher-ARR target but reduce confidence. KeyBanc’s native bands (<$10M, $10–25M, $25–50M, >$50M) only partially overlap several standard bands; where a KeyBanc band covers part of a target band, its value enters the modeled median as a partial-overlap vote rather than being dropped, which shifts seven modeled medians slightly versus an exact-band-only rule.
4. Modeled medianFor definition-aligned metrics, take the unweighted median of the eligible provider medians after band mapping. Each provider gets one vote; reported sample size affects confidence, not the center.
5. Modeled quartilesFor definition-aligned metrics, map each provider’s published 25th and 75th percentiles separately, then take the unweighted median at each percentile. A median-only provider may support the modeled center but cannot create a P25 or P75 estimate. Every percentile is labeled as modeled rather than pooled or observed.
6. ExtrapolationPrefer broader-band mapping to numerical extrapolation. If a future refresh requires extrapolation, limit it to one adjacent standard band, shrink the source slope by 50%, and cap the result within the observed cross-source envelope. No multi-band extrapolation is allowed.
7. Definition controlsDo not pool NRR, CAC payback, or Rule of 40 while formulas remain unreconciled. Show the low-to-high provider span separately for P25, median, and P75. Pool S&M and R&D only on the same denominator; therefore High Alpha’s ending-ARR cuts are excluded from the revenue-denominator model.

ACV-band normalization (Section 4)

StepMethod
1. Source and bandsUse the six native ACV bands and the published P25, median, P75, and subgroup sample sizes from the Benchmarkit 2026 Quartile tab. No other provider publishes a sufficiently complete, comparable ACV grid for these metrics, so this is a single-provider model.
2. Directional metricsFor GRR, NRR, CAC payback, blended CAC ratio, and new-name CAC ratio, fit P25, median, and P75 separately with sample-size-weighted isotonic regression (subgroup sample sizes normalized to unit mean). This preserves a non-decreasing ACV progression while moving high-sample bands less than sparse bands.
3. Non-directional metricsFor LTV:CAC and Magic Number, do not impose monotonicity. Fit a sample-size-weighted low-curvature trend across ordered ACV bands, penalizing abrupt second differences (curvature penalty λ = 4 on second differences, with subgroup sample sizes normalized to unit mean) while retaining the observed overall shape.
4. Missing internal cellsFill an internal blank from the fitted curve on a logarithmic ACV scale. A filled value is marked † and receives confidence D; it is never represented as an observed survey statistic.
5. Terminal extrapolationIf the final ACV band is blank, extrapolate no more than one adjacent band using 50% of the preceding log-band slope. Carry forward the latest observed P25-to-median and P75-to-median relationships so the modeled distribution remains ordered. This rule is used only for blended CAC above $100K.
6. GuardrailsEnforce P25 ≤ median ≤ P75 after fitting. Retention and payback are rounded to whole percentage points or months; CAC ratios and Magic Number to two decimals; LTV:CAC to one decimal. Published values remain visible in the audit table immediately below the normalized table.

The ACV model removes reversals that conflict with a stated shape constraint; it does not automatically cap a high value that remains directionally consistent. Accordingly, the published 144% NRR P75 in the >$100K band remains in the normalized table. Where an extrapolated value lands on an exact rounding boundary, the display reflects that tie: the >$100K blended-CAC median fits at $1.835 and is shown as $1.83.

Source eligibility by metric

MetricP25/P75 contributorsMedian-only contributorsExcluded from modeled core
ARR growthBenchmarkit, High Alpha, SaaS CapitalKeyBancICONIQ (enterprise/top-quartile); Maxio (mean); Battery (targets)
GRRBenchmarkit, High Alpha—ICONIQ (enterprise/top-quartile)
NRRBenchmarkit and High Alpha, retained as source spans—KeyBanc band data unavailable; ICONIQ definition and sample differ
CAC paybackBenchmarkit and High Alpha, retained as source spans—ICONIQ (top-quartile enterprise sample)
Software gross marginBenchmarkit, High Alpha—ICONIQ (basis undisclosed; enterprise/top-quartile)
ARR per employeeBenchmarkit, High AlphaKeyBancICONIQ (enterprise/top-quartile)
S&M / R&DBenchmarkitKeyBancHigh Alpha (ending-ARR denominator); ICONIQ (enterprise skew)
Rule of 40Benchmarkit and High Alpha, retained as source spans—KeyBanc band data unavailable; ICONIQ uses FCF and top-quartile sample

The provider set can differ between the modeled median and modeled quartiles. Confidence C applies when quartiles rely on one provider, the median uses a broader or median-only source, or formula comparability is uncertain.

Confidence grades. A requires at least three eligible providers with strong target-band alignment and compatible definitions. B requires at least two usable providers with only limited band mapping. C indicates a broad/open-ended mapping, quartiles supported by only one provider, a median-only gap fill, or material formula uncertainty. D indicates incompatible definitions and is directional only. The badge on the median row applies to the corresponding three-row ARR-band group.

Rounding and blank handling. Percentages are rounded to the nearest whole point, ARR per employee to the nearest $1K, and CAC payback to whole months. A source blank or zero placeholder for an insufficient subgroup is never treated as an observed zero. If only a median is available, P25 and P75 remain blank. Overlapping native bands are weighted by logarithmic ARR width before the final provider median is calculated.

Audit trail. Every modeled cell can be traced to the native provider rows in Section 3. The model introduces no external benchmark sources and does not use ICONIQ’s enterprise/top-quartile figures to fill broad-private gaps.

Appendix D: Revision History

DateRevisionChanges
October 6, 2026Current (v13)Removed Section 8, “Benchmark Report Card,” from the distributed report; the grades are preserved in an internal archive file. “Put These Numbers to Work” renumbered from Section 9 to Section 8.
October 6, 2026Audit correctionsApplied independent-audit corrections: fixed the Appendix A2 G&A 31–50%-growth cohort median (25% → 21%, per the Benchmarkit workbook); populated hybrid (subscription + usage) NRR at 102% (n=35) and resolved the seat-based NRR 95%/98% inconsistency to the workbook’s 98%; re-attributed the growth-deceleration narrative to Benchmarkit and added KeyBanc’s projected 2025–2026 rebound; removed Battery from headcount/retention claims it does not support; recast causal efficiency-vs-retention language as associative; caveated the pricing-model retention signal (ICONIQ $100M+ tie at 117%, usage-only n=9, High Alpha’s hybrid finding); refreshed stale ICONIQ edition references after the State of Scaling 2026 addition; documented the KeyBanc partial-overlap mapping rule in Appendix C; and clarified Appendix A2 trend-period labels. Regenerated the companion Markdown file, fixing three corrupted table rows introduced by the previous HTML-to-Markdown conversion.
October 6, 2026ICONIQ 2026 updateAdded ICONIQ’s State of Scaling 2026 (September 2026, n=137) as a ninth source. Labeled the Section 3 ICONIQ rows as longitudinal pooled data (2013–Q2 2025) with a callout explaining the period semantics, and added a current-period subsection to Appendix A4 with ICONIQ’s year-cohorted 2026 medians, top-quartile NDR and ARR-per-FTE values, and the ICONIQ Pacesetter Index. Added Pacesetter Index rows (median / top quartile, 2024–Q2 2026) to the Section 3 ARR-growth, NRR, GRR, gross-margin, and ARR-per-employee tables under their native bands, adding a $25–100M column where needed. No Section 2 modeled values changed; ICONIQ remains excluded from modeled centers.
October 3, 2026Narrative revisionRetitled the report “The 2026 SaaS Benchmarks Report.” Rewrote the executive summary, section introductions, and macro narrative in a direct first-person voice with explicit takeaways. Added a “what to steal” callout to each source deep-dive and a new Section 9, “Put These Numbers to Work.” Replaced remaining em dashes and shorthand (“(M)”, “TQ”, “exec-summ”) in prose. No data values were changed.
October 3, 2026Audit correctionsCorrected the Appendix A2 total-population table against the Benchmarkit Quartile tab (software gross-margin P25 to 72%; populated new-name CAC P75, expansion-CAC quartiles, and total-revenue GM P25; CAC payback minimum to 2 months); added a sample-size note distinguishing published-report N from Quartile-tab counts; restated three LTV:CAC modeled cells at the recalculated rounding; documented the ACV model’s weight normalization and curvature penalty (λ = 4) and the blended-CAC rounding tie. Replaced section-symbol (§) cross-references with plain “Section N” references for readability.
October 3, 2026ACV normalization revisionAdded a normalized ACV-band planning table with modeled P25, median, and P75 values; filled sparse cells; smoothed isolated reversals using sample-size-weighted shape constraints; preserved the raw Benchmarkit table; and documented the methodology and confidence limits.
September 25, 2026Normalized ARR revisionMoved the normalized ARR table forward; added P25/median/P75 rows, use guidance, source-eligibility disclosure, data-vintage labeling, mechanical report-card scoring, navigation, and methodology clarifications.
September 24, 2026Validation revisionValidated Benchmarkit workbook quartiles; corrected ACV and source values; removed duplicate or unsupported figures; combined provider tables by metric.
July 2026Initial publicationPublished the eight-report meta-benchmark and source deep-dives.

© 2026 The SaaS CFO. Figures extracted directly from each report’s 2025–2026 edition; cross-source ranges include only figures judged sufficiently comparable; definition-sensitive metrics are reported by source. All sources are independently credible and are credited above. This report reconciles their published data for like-for-like comparison; it does not reproduce any source’s proprietary tables in full. Companion materials: one-page executive card and full comparability matrix.