The Money Metrics: Churn, NRR, GRR, and Friends

Sooner or later every conversation about Customer Success lands on a number. Usually it's NRR, usually on a slide, usually with no definition attached. Someone says “we're at 118” and the room nods along, because asking what went into it feels like admitting you don't know.

So let's go through the money metrics properly. There aren't many of them and none are difficult. The difficulty is that four companies can report the same metric having measured four genuinely different things, and all four can be honest about it.

The leaky bucket

The simplest question you can ask about a customer base is how many customers are still here. Logo retention answers it: take the number of customers at the end of the period, subtract the new ones you won during that period, and divide by the number you started with. Logo churn is the same picture from the other side, customers lost during the period divided by customers at the start. Stripe's published definitions use exactly this beginning-and-end-of-period basis, which is worth knowing, because plenty of internal dashboards quietly use a different one.

A logo is a logo. The account paying you $1,200 a year and the account paying $1.2 million each count as one. That's the metric's strength in high-volume subscription businesses, where the pattern of who leaves matters more than any single departure, and its weakness everywhere else. Logo churn tells you the bucket has a hole. It says nothing about how much is running out of it.

GRR, the number that can't be rescued

Gross Revenue Retention measures money instead of logos, and only the money you already had. Start with recurring revenue from existing customers at the beginning of the period. Recurring revenue here means contracted subscription or consumption revenue you expect to repeat, not one-time services or implementation fees. Subtract churn, the revenue from customers who left entirely. Subtract contraction, the revenue lost from customers who stayed but bought less: dropped seats, downgraded tiers, reduced usage commitments. Divide by where you started.

GRR cannot go above 100%. That ceiling is the whole point. No amount of upsell can paper over it, which makes it the cleanest read available on whether the revenue base is actually holding. If I could carry one number into a conversation about customer health, it would be this one.

NRR, the number everyone quotes

Net Revenue Retention adds expansion back in: starting recurring revenue, plus expansion, minus contraction, minus churn, divided by starting recurring revenue. You'll also see it called net dollar retention, NDR, or dollar-based net revenue retention. Stripe treats those as closely related terms for the same idea, and in practice people use them interchangeably.

Above 100% means your existing customers, as a group, pay you more than they did a year ago, with zero new logos. Snowflake reported 125% as of January 31, 2026, and ties that existing-customer growth to customers migrating additional workloads onto the platform and consuming more. A business like that grows without selling to anyone new. Investors are right to care.

Here's the strong opinion for this chapter: NRR is the most quoted metric in SaaS and the one that explains the least. It's a scoreboard, not a game tape. A handful of accounts expanding hard can carry a cohort that is bleeding underneath. A price increase can lift it without a single customer receiving more value. NRR at 118 with GRR at 84 is a different company from NRR at 118 with GRR at 96, and only one of those two numbers tells you which company you're running. So when someone shows you NRR, ask for GRR. Neither number is wrong. They answer different questions, and the mistake I see most often is letting one impersonate the other.

The fine print is part of the metric

Snowflake's SEC filing is the best public illustration of why the definition matters more than the figure. Its NRR calculation names a specific cohort of capacity-contract customers, compares their product revenue across a trailing two-year measurement period, and leaves customers that later stop using the platform inside that cohort with zero second-year revenue. That last clause is a company choosing to make its own number harder on itself. Plenty of internal calculations quietly drop those customers out instead, and the result looks better for reasons that have nothing to do with customers.

The same care applies to the two metrics that usually sit beside these. Renewal rate is renewed contracts or ARR divided by contracts or ARR actually due for renewal in the period, so you have to state whether you're counting logos or dollars, which accounts are eligible, how early renewals are handled, and what a multi-year contract does to the denominator. Expansion rate is expansion recurring revenue over starting recurring revenue, which tells you whether the installed-base growth motion works but not whether that growth came from delivered value or from a price change. Each of these numbers compresses a messy reality into one figure, and compression always drops something. A metric with an unstated denominator is not a specification. It's a vibe with decimals.

Instruments, not verdicts

Then there are the survey metrics, which measure something real and get asked to do far more than they can.

NPS is the share of respondents scoring 9 or 10 minus the share scoring 0 through 6, with 7s and 8s counted as neither. It's a relationship-level signal, useful for watching a trend inside a consistent program. It is not a universal predictor of growth, whatever the deck implies. A 2007 study in the Journal of Marketing by Keiningham and colleagues tested that claim longitudinally and found the evidence for NPS's superiority didn't hold up.

CSAT asks about one specific interaction, a support ticket, an onboarding milestone, a training session, and is usually reported as top-box responses divided by total responses. Qualtrics draws the line well: CSAT is a here-and-now measure, NPS a relationship-level one. Customer Effort Score, or CES, asks how hard something was to do, and works best pointed at a single place in the journey where you suspect friction. It has no single calculation convention, so whichever one you pick, write it down and stop changing it.

All three tell you where to look. None of them tell you what's true.

Where the money metrics actually sit

Churn, GRR, NRR, renewal and expansion all live at the far end of the chain that runs from a promise through adoption to a verified outcome. They're consequences. By the time one of them moves, whatever caused it happened months earlier: an adoption curve that flattened, an outcome nobody verified, a champion who changed jobs in March. That's why a quarter spent trying to improve GRR directly tends to produce save plays and discounts rather than retention.

Read these numbers as the receipt for last year's work. They confirm whether value showed up and whether the customer paid for more of it. What to do about next year is written further up the chain, in the links this series has been working through one at a time.