Why the System Persists

The Unread Research

None of the findings in this report are secret. The gross churn floor, the contract-length differential, the expansion-at-scale threshold, and the top-quartile performance gap have been visible in the survey series for years, and adjacent research from Bain, McKinsey, and ChartMogul has said much the same. Which raises the only question the data cannot answer on its own: if the evidence is this clear, why has the industry not closed the gap? Inference

The answer is not that operators lack intelligence or diligence. It is that the organizational systems most companies run produce a specific, predictable set of behaviors that make system-level change extraordinarily hard, not as a matter of will but as a matter of organizational physics. Understanding why the system persists is the first step of the path forward. A company that tries to rebuild its retention model without first understanding the forces that work against it, from inside its own measurement, its own budget process, and the rational self-interest of its own leaders, usually repeats the alignment initiative that failed last time. Inference

Measurement as Distortion

The first obstacle is not politics. It is the measurement architecture itself. Standard SaaS financial reporting was built to answer one question: how much revenue do we have, and how efficiently are we generating new revenue? That was the right question in the acquisition-led era, when new ARR was the growth engine and the metrics that mattered all traced back to acquisition efficiency: CAC, payback, pipeline velocity, quota attainment. Inference

The growth model has shifted and the reporting has not. Expansion crossed half of new ARR in 2024, 52% on a pooled basis, a crossing driven entirely by companies above roughly $25M, while below that line growth still comes mostly from new logos (see The Expansion Myth). Either way, the metrics that dominate board reporting, new ARR and pipeline and quota, are acquisition metrics. The inputs that actually determine net dollar retention, contract-length mix, professional-services attach, customer-success methodology investment, and ICP discipline, are absent from standard reporting entirely. Finance reports the output without reporting the inputs that control it. Inference

The problem compounds through Goodhart’s Law: when a measure becomes a target, it ceases to be a good measure. As soon as NDR becomes a customer-success performance target, the team optimizes for the number rather than for the customer outcomes a healthy number reflects. Renewals are managed, relationships preserved, customers who never achieved value retained through engagement rather than results. The number holds flat. It reads as an execution problem in the one function that does not control the inputs, and the real problem stays invisible in the same reporting system that produced it. Inference

The Attribution Gap

The second distortion is attribution. New-logo revenue is cleanly credited to Sales at the moment of close: the timestamp, the ACV, the rep, all captured with precision. Expansion falls into a contested zone between Sales and customer success, settled by definitions that get renegotiated whenever quota credit is at stake. Retained revenue is not counted as revenue at all. The customers who did not churn appear in no one’s credit column. A team that saves millions of at-risk ARR through systematic outcome delivery earns no revenue attribution for the work, because the loss it prevented simply never shows up. Inference

This asymmetry is a historical accounting convention, revenue recognized when earned, not when preserved, that made sense when retention was assumed rather than engineered. In a market where gross churn runs 14–15% a year and a new-logo dollar cost $1.78 of sales and marketing spend in 2021 data, the last year the survey split CAC by motion, treating retained revenue as a non-event systematically undervalues the function responsible for producing it. Capital then flows to the functions whose contribution is legible in the reporting system and away from the ones whose contribution is invisible. In this survey series customer-success spend was never even broken out as its own line, and most of its cost sat inside sales and marketing, the very line the margin correction cut hardest (see The Invisible Function). The investment disparity is not irrational given the architecture. It is the rational output of a system that counts one function’s contribution and ignores the other’s. Inference

The Physics of Silos

Decades of organizational research explain why cross-functional alignment fails at a rate practitioners treat as background noise. W. Edwards Deming argued, in effect, that local optimization is the enemy of system optimization. When an organization measures and rewards each function’s own performance, it creates structural incentives for every function to optimize itself even when doing so degrades the whole. That is not a failure of character. It is the rational response to the incentive structure. Inference

The map is exact. Marketing is measured on pipeline volume against its own ICP definition, so it optimizes toward whatever that definition rewards, even when the definition was set by sales targets rather than retention data. Sales is measured on closed ARR against a quota that does not adjust for downstream retention, so it closes deals, including marginal-fit deals, because the comp plan rewards the close and does not penalize the eventual churn. Customer success is measured on renewal rate and NDR, metrics it does not fully control and whose inputs were set in Sales and Finance, so it manages relationships and suppresses visible churn signals. Each function behaves rationally. The system produces irrational outcomes. And because each can point to legitimate performance inside its own metrics, no one accepts accountability for the whole. Inference

Defensive Routines

Chris Argyris named the pattern organizations use to protect themselves from examining their own assumptions: defensive routines. When NDR declines, the honest diagnosis is almost always “multiple functions, through multiple mechanisms, over the past twelve to twenty-four months.” But that answer is threatening to everyone in the room, because in an environment where budget is allocated on demonstrated performance, admitting contribution to a failure is equivalent to arguing for your own resource reduction. Inference

So the routine converts a system diagnosis into a bilateral negotiation, each function arriving with evidence of the others’ failures. Sales cites poor onboarding of the customers it qualified correctly. Customer success cites commitments on features that do not exist. Every accusation has evidence; every defense has evidence. The CEO mediates, agreements are made, the agreements are genuine in the moment and irrational to keep, and the system does not change. Argyris called this single-loop learning, solving problems inside existing assumptions. What this report describes is a double-loop problem: the assumptions that need questioning are the structure, the measurement, and the capital allocation model, not the tactics inside each function. Inference

The Political Economy

Jeffrey Pfeffer’s work on organizational power documents an uncomfortable finding: resource allocation tends to follow power rather than logic. Around retention the dynamics are acute because accountability is contested. Everyone agrees NDR matters; no one fully agrees who owns it; and in that ambiguity the most powerful function, historically Sales, tends to win the budget contest regardless of where the retention leverage actually sits. Inference

The implication is directional and it shapes the entire playbook that follows: the case for rebalancing toward retention cannot be made to the CRO or the CCO, because both are competitors in the allocation contest with incentives to preserve the split that favors their function. The case has to be made to the CFO and the CEO, who sit above the contest and whose interests, enterprise value, forecast accuracy, and capital efficiency, are best served by the rebalancing the data supports. Inference

Why the Report Flatters

There is a recursive quality to the measurement problem. The reporting architecture was built by and for the people who use it, and the metrics that reach board slides were selected, over years of iteration, to show the company in its best available light rather than to show the full composition of its retention economics. This is not deception. Leaders report what they can stand behind, understand, and control. A CFO who decomposes NDR into gross churn and downsell for the first time is voluntarily creating visibility into dynamics they have not previously been accountable for. Inference

Amy Edmondson’s research on psychological safety extends to leadership teams: when a culture treats problem identification as evidence of underperformance rather than as diagnostic information, leaders learn not to identify problems clearly. The result is board reporting that shows 101% NDR as a single reassuring line rather than its composition, roughly 14% gross churn offset by expansion. Reported honestly, that composition raises questions the team isn’t ready to answer. The fix is less a data-architecture problem than a leadership-culture one, and the authority to make honest problem identification safe sits with the CEO and CFO. The chapters that follow assume that authority is present, and show what to do with it: the fixes that only look like change in The Half-Measures Taxonomy, and the program that actually moves the floor in The Playbook. Inference