Introduction

The Reckoning

Read in isolation, each annual edition of the KeyBanc SaaS survey reports retention as stable. Read together, across seven editions and reconciled against one another, the same editions describe a sustained decline.

This report is that longitudinal analysis: 247 private SaaS companies, drawn from seven annual editions of the KeyBanc and Sapphire Ventures survey (2019 through 2025), with each figure verified against the source materials and graded for confidence. Its subject is not a single weak year but a structural shift in how SaaS revenue compounds, and a multi-year decline in the measure that most determines durability: net dollar retention.

The thesis is straightforward. The retention-and-expansion engine that made SaaS unusually capital-efficient has weakened, and for several years cost reduction obscured the weakening. Margins improved; the underlying revenue engine did not. Net dollar retention fell from a 2021 peak of 109% to 101% even as EBITDA rose, because reductions in operating expense produce profit without producing retention. The 2021 figure coincided with the top of the cheap-capital cycle, and 101% sits near the level the metric held before it; the durable signal is the collapse of the cushion beneath the number, which the NDR Crisis section takes up in full. The survey itself projects a recovery in its 2025E and 2026E columns; the report treats those as the survey’s estimates to be read against its forecasting record, not as measurements, and the KPI Scorecard carries that calibration.

Part of why the decline drew little notice is methodological. Each edition of the survey restates prior years, generally revising them downward, so any single edition appears stable. The trend becomes visible only when each figure is held to its own edition, values are not spliced across editions, and the full series is read at once. That reconciliation is the work this report performs, which is why its conclusions differ from the headline any single edition prints.

How This Research Came About

A note on how this research came about, because it affects how to read it. We started with a theory, not a dataset. After years of working inside SaaS revenue organizations, our working model was that churn, retention, and expansion are not separate problems parceled out across four separate teams (Product, Marketing, Sales, and Customer Success); they are connected parts of one revenue system that spans those four functions plus support, services, and finance, and they move together. What we lacked was public data rich enough to test that model properly.

The KBCM survey series gave us that test. Before we looked at the results, the model made specific predictions: downsell should be hiding inside the netted retention metrics, expansion should stall below a certain company size, and department-level fixes should show little effect on system-level outcomes. Then we checked the predictions against the data. Most held. Some needed revising, and one (the decline of multi-year contracts) surprised us outright; the findings note which is which. This matters for you as a reader because a hypothesis-first analysis can be judged on whether its predictions held, and the evidence grade on each finding exists so you can do exactly that.

Reading the ‘E’ Years

This report draws on every KBCM/Sapphire edition back to 2019, not one edition in isolation, but the current scorecard and forecast tables are pinned to the latest one available: 2025, fielded from Q2 through Q3 of 2025 and published before 2025 had a full year of actuals behind it. That gap is not particular to this edition: KBCM closes its survey mid-year and publishes each November, so the two most recent years in every edition print as forecasts: 2025E and 2026E in the 2025 edition. The last settled year in that edition is 2024, and it stays the last settled year until the next edition, typically about a year later, replaces 2025E with an actual.

This matters for how the report is read. Where the prose refers to “the current state” or “where retention stands,” it means the survey’s last settled year, 2024, not the calendar date on which the report is read. Charts mark the estimate segment with a dashed line wherever it appears, for exactly this reason. The full mechanism, verified against the source survey PDFs, is documented in Data Integrity & Corrections Log.

The Two Halves

The eight findings in this report are not a list of separate problems. They describe one system, and they divide into two groups, which is why the report is ordered as it is.

The first group concerns the revenue engine itself: declining net retention, expansion that proves to be a function of scale rather than a general rule, structurally frozen churn, and downsell that drains roughly a third of all revenue loss without appearing on standard dashboards.

The second group concerns the operational response, and how it compounded the first. Companies withdrew from the multi-year contracts most associated with lower churn, reduced operating expense in the functions that support retention, allowed acquisition economics to stretch until expansion was subsidizing new-logo growth, and continued to price by seat as AI began to reduce seats.

Together the two groups form a reinforcing cycle: weakening retention prompts cost reduction, cost reduction falls on the functions that defend retention, and retention weakens further. The Integrated View section traces that cycle in full; the sections between here and there are the evidence that it is real.

How to Read This

Each quantitative claim carries a confidence tier, from Verified, taken directly from a survey edition, to Inference, the report’s interpretation of what a pattern indicates. Interpretation is labeled as such and is not presented as fact. Figures that could not be reconciled to the source are held back and marked rather than published. The Methodology and Data Integrity section documents what was verified and what was not.

The analysis is intended to be auditable. Each figure traces to a named source edition, and the Metric Explorer allows the underlying calculations to be reproduced. There is also a story in what we found when we lined the editions up, and the Proof section opens with it: The Story Behind the Story.

Source: KBCM/Sapphire Private SaaS Survey, seven editions 2019–2025; see Methodology & Data Integrity for the per-figure registryVerified