Data Integrity & Corrections Log

This page lists the tier system, the per-data-point registry, and the publishers’ own restatements and estimate misses between survey editions. For how each figure is verified, including the double-read procedure and how the certified dataset is stored and organized into a graph, see Methodology.

The Tier System

Four tiers label the confidence behind every claim: the answer to the weak-marketing-stats problem. Verified is taken directly from a primary survey. Cross-Survey is consistent across surveys but varies in exact value by respondent pool (the range is documented). Calculated is derived from sourced data with formula and inputs recorded. Inference is editorial interpretation: forward projections and causal claims, labeled so you can weigh them as such.

Most inference tags on this site share a cause: the survey instruments each function separately and rarely asks the connecting question directly, so a causal or comparative claim has to be built from adjacent, individually-verified figures rather than read off a single line. The Benchmark’s Blind Spots catalogs the specific missing questions behind the report’s largest inference calls.

Reading KBCM Across Editions

KBCM publishes a new SaaS survey every year, and each edition restates the prior years it covers. The same data year doesn’t carry the same value across editions. Because the respondent panel changes annually, a 2022 figure printed in the 2024 survey is a different measurement than the 2022 figure printed in the 2025 survey. Both are correct for their own edition; neither is a typo. Across the cross-edition differences we verified directly against the source PDFs, the later edition revised the earlier year downward about 79% of the time, and that pattern holds even on settled-history-vs-settled-history figures, so it is not just optimistic forecasts being trued up. Later editions consistently mark the past as less profitable and slower-growing while marking it as more expansion-efficient (EBITDA and ARR revised down; NDR and sales-efficiency revised up). Inference

The revisions are also easy to miss, because the survey sometimes renames a metric between editions while revising its history. The 2022 median contract value, for example, appears under one label in the 2024 edition and a slightly different label in the 2025 edition, $10K apart. A reader matching on names alone would never see the two prints as the same series. We maintain a register of every cross-edition restatement we have verified, including the renamed ones, and each one is pinned to its source page in both editions.

Two consequences we hold ourselves to: every figure in this report is pinned to a single named edition, and we never splice values from different editions into one trend line. Doing so would manufacture movement that is partly a measurement change. Where this report shows a multi-year trajectory, it uses the most recent edition’s restated series for the years that edition covers, because within a single edition the numbers are internally consistent. We can show that KBCM revises the past downward; we cannot prove why from published medians alone (no fixed-cohort series exists), so we treat the panel-composition explanation as a hypothesis, not a finding.

Framing built on verified figures (EBITDA −26%/−47%, ARR $22.5M/$17.0M, GDR 89%/86%, NDR 102%/106%, all PDF-confirmed); the downward-restatement interpretation is editorial synthesisInference

Why the ‘E’ Years Persist

The survey itself states this directly. The 2025 edition reads: “The survey was conducted from Q2’25 close through Q3’25 close, so executives had a clear line of sight into 2025E performance” (KBCM/Sapphire 2025 Private SaaS Survey, p.1). Fielding runs roughly April through September; the current year’s full-year financials do not exist yet at that point, so it is published as an estimate alongside the following year. The same edition pins its market-comparison exhibits to “as of September 30, 2025 unless otherwise noted” and, for public-market multiples, “Source: Capital IQ as of 11/7/2025.” That is consistent with a report finalized and distributed in the weeks around that date. The prior edition shows the identical pattern one year earlier: market data pinned to “as of September 30, 2024” and “market close on 10/4/2024.” This is a fixed annual cadence, not a one-off. Verified

The practical consequence: this report draws on every KBCM/Sapphire edition back to 2019, but no chart or table splices figures across editions; each is pinned to the edition that reported it, per the no-splicing rule above. The current scorecard and forecast tables are pinned to the 2025 edition, whose fielding window closed before 2025 had a full year of results. In that edition’s own tables, 2024 is the last settled (actual) year; 2025E and 2026E are the survey’s forward estimates, carried as such in every chart and table on this site that draws on the 2025 edition, and they will remain estimates until KBCM and Sapphire publish their next edition (historically each November) with settled 2025 actuals in hand. The year after that will not settle until the edition after that.

Read this report accordingly: where the prose refers to “the current state,” “where retention stands,” or similar, it means the survey’s last settled year (2024), not the calendar date on which the report is being read. The scorecard, forecast tables, and dashed chart segments carry the estimate/actual distinction explicitly for exactly this reason. See the Disclaimer’s point-in-time notice for how the report treats its own currency more generally.

Survey fielding language, market-data pin dates, and cadence pattern verified directly against the 2025 and 2024 KBCM/Sapphire Private SaaS Survey source PDFsVerified

Data Point Registry

Sort by any column. Each row links a headline figure to its primary source and tier.

Corrections Log

KBCM restates prior data years from one edition to the next, so the same data year can appear with different values across editions. The entries below are not this report’s corrections; they are the publishers’ own restatements. Comparing the 2024 and 2025 survey editions, 39 figures were restated, and 31 of them (79%) moved the past in a worse direction in the newer edition. The pattern is not a forecast artifact: on the figures where both editions report settled actuals (data years 2022 and 2023, no estimate involved), 78% still move down. The only systematic upward restatements are net retention and sales efficiency, and they concentrate in the oldest (2022) vintage.

The direction is consistent: later editions mark the recent past as less profitable and slower-growing (EBITDA, ARR, and Rule of 40 revised down) while marking it as more expansion-efficient (net retention and the 2022 magic numbers revised up). The mechanism is a hypothesis rather than a finding. The pattern is consistent with panel-composition and survivorship drift, since KBCM redraws its respondent panel each year and publishes no fixed-cohort series, and it cannot be cleanly separated from real market change using the published medians alone. Inference

What the panel limitation does and does not touch. The annual redraw weakens one class of claim and leaves two others intact, and the report is built on the two that survive:

• What it weakens: point-to-point level changes. When a median moves between editions, some unknown share of the move is sample turnover. The report treats single-step level changes as weak evidence and says so where they appear.
• What it cannot fake: structure inside an edition. Gradients measured within a single pool (churn by contract length, churn by services attach, expansion share by company size) compare companies that answered the same questionnaire in the same year. Turnover cannot manufacture a gradient, and these gradients carry the report’s strongest findings.
• What it strengthens: invariance. The gross-churn floor held at 11 to 15% across seven editions and materially different respondent pools. A constant that survives that much sample change is more credible for it, not less.

The bias direction is also knowable, and conservative: companies that decline, sell distressed, or shut down stop answering surveys, so survivorship tilts every median toward health. The structural decline this report documents is likelier understated than overstated. The reader rule that follows: trust the gradients and the directions, hold the levels loosely, and treat any single-year improvement as unproven until a second edition confirms it.

Who answers the survey, and since when we can know. The 2025 edition is the first in at least seven to publish a respondents-by-role breakdown: 57 of 71 respondents sit in the CFO or finance function (2025 Survey, p. 8). Every earlier edition in this report’s archive (2019 through 2024, checked page by page) describes its panel only in aggregate and cuts participants by ARR, sector, ownership, and growth, never by role; editions before 2019 were not available for examination. Two things follow. The survey’s richly instrumented financial metrics and thinly instrumented operational retention metrics are consistent with the panel that answers it. And for at least the six preceding editions, readers benchmarked against this instrument without knowing whose field of vision produced it; the disclosure is new transparency, and it deserves credit as such.

Estimate and Forecast Accuracy

KBCM prints forward estimates, marked E, alongside settled actuals, which lets a prior edition’s estimate be checked against the actual a later edition reports. Where a 2024 estimate in the 2024 edition could be measured against the 2024 actual in the 2025 edition, the actual came in worse than the estimate in 10 of 12 cases (83%), concentrated in profitability, ARR dollars, and growth. The estimates lean optimistic, and consistently so.

Forward estimates in this report are marked E and tiered accordingly. They should be read as optimistic-leaning, and none carries the weight of a settled figure. Calculated

Cross-edition restatements and estimate-vs-actual deltas derived from PDF-confirmed values in both the 2024 and 2025 survey editions; 39 divergences confirmed, direction and magnitude computed arithmeticallyVerified