The one number that decides whether a SaaS business compounds — or quietly runs in reverse

Everyone calls SaaS a “compounding machine.” Few can point to the single dial that decides which direction the machine turns. It’s net revenue retention (NRR): how much recurring revenue this year’s existing customers generate a year later, after churn, downgrades, and expansion.

The math is almost hypnotic. Take a company at $100M of ARR with 120% NRR. Sign not a single new customer, and next year it’s $120M, then $144M, then $173M — the existing base alone throws off growth, year after year. That’s the dream the whole industry sells: predictable, cumulative, high-margin (multi-tenant cloud makes serving one more customer near-free), and self-financing revenue, stacking on itself like interest. It genuinely can be one of the finest business models ever built.

But flip the dial. Drop NRR below 100% and the very same machine runs in reverse — the installed base shrinks itself like a leaking bucket, and the company must pour in ever more new customers just to hold flat. The machine doesn’t automatically turn the right way. Which way it turns depends entirely on which side of 100% NRR sits.

The purest expression of this is usage-based software. Snowflake doesn’t ask how many employees you have; it charges for what you consume — queries, storage, compute. As a customer’s business grows, usage grows automatically, no new contract required, and NRR can run well above seat-based peers. Snowflake’s fiscal-year revenue crossed ~$4.68B (from ~$3.63B, ~29% growth). It looks like the perfect machine.

Contrast that with seat-based software: a seat contract is rigid within its term, so revenue holds steady but expansion needs a salesperson to negotiate more seats. Usage flips both sides — expansion needs no contract, and neither does contraction. That’s why, for the same headline “high NRR,” a usage-based number swings far more violently than a seat-based one’s. Neither is “better” in the abstract — but they demand completely different underwriting, and pretending a usage NRR is as durable as a seat NRR is exactly how investors get hurt.

Here’s the discipline the boom forgot: the same mechanism that makes expansion automatic makes contraction automatic. When budgets tighten, a usage-based customer needn’t cancel anything — they just use a little less, and revenue feels it immediately. Usage NRR is high and volatile; it swings harder than seat-based both ways. So when you see a gorgeous NRR, don’t extrapolate — interrogate. Three reasons a boom-time NRR overstates the future: it naturally recedes with scale (big customers saturate); it’s procyclical (flattered in booms); and its basis can be quietly “prettied” by redefining the cohort or window. A beautiful number is the start of the questions, not the end of them. No part of this machine comes with only upside.

This is Chapter 2 of “Compounding in the Cloud.”