But there’s something you don't know: three months ago, they already decided they're churning. They haven't told you yet. They won't for another two months. And by the time they do, it's too late to save this account. 

This is the reality of customer retention today. And it's exactly why net revenue retention (NRR) is the metric every CS and CX leader needs to care about.

What NRR actually means

NRR measures the percentage of revenue you keep from existing customers over time, accounting for expansion, contraction, and churn. It's the clearest signal you have of how much value your customers actually think you're delivering.

Think about it this way: acquiring a new customer is expensive. Really expensive. So if you can provide more value to the customers you already have (and articulate that value clearly), your NRR goes up. If you're less relevant, or they're not getting what they came for, it shrinks.

Best-in-class SaaS companies target 120%+ NRR, and below 100% means you're losing ground even as you close new deals.

There are two elements to improving your NRR: reducing churn and identifying customers who are ready to expand. Great CS teams play defense, know which accounts are primed for upsell, and act on it.

How to calculate NRR

The formula is straightforward:

NRR = (Starting MRR + Expansion MRR − Contraction MRR − Churned MRR) ÷ Starting MRR × 100

If you start the month with $100k in MRR, gain $10k from upsells, lose $5k from downgrades, and lose $5k from churned accounts, your NRR is 100%. You're not growing, but you’re also not shrinking.

The goal is to be above 100%. The best SaaS companies in the world consistently hit 120%+, meaning their existing customer base is growing faster than it churns, even without net new logos.

NRR vs. ARR vs. GRR: what's the difference?

These three metrics often get used interchangeably. They shouldn't.

ARR (annual recurring revenue) is your total contracted revenue for the year. It tells you how big you are, but not how healthy you are. You can have strong ARR and terrible retention.

GRR (gross revenue retention) measures how much revenue you keep from existing customers, but only counts churn and contraction. It ignores expansion. GRR has a ceiling of 100%.

NRR (net revenue retention) is the full picture. It accounts for churn and expansion, which means it can exceed 100%. That's the goal. NRR above 100% means your existing customers are generating more revenue than you're losing: a compounding growth engine that doesn't require you to constantly refill a leaky bucket.

What is a good NRR benchmark?

It depends on your segment, but here's a rough guide to NRR benchmarks:

  • Below 90%: Red flag. You're shrinking from within.
  • 90–100%: You're staying afloat, but churn is eroding growth.
  • 100–110%: Solid. Expansion is offsetting losses.
  • 110–120%+: Best-in-class. Your existing customers are a growth engine.

Top-tier SaaS companies, like Snowflake and Twilio at their peaks, have posted NRR north of 130%. Below 100% means you're losing ground even as you close new deals. And that's a problem no amount of new business can outrun forever.

Why CS teams miss churn and growth signals

Here's the honest truth: the world is complicated, humans are complicated, and your CSMs are managing 40, 50, sometimes 60+ accounts at a time. They only get signals when they're actively engaging with a customer. And they can't engage with everyone, all the time. So things get missed.

Add to that: market consolidation, budget scrutiny, and the fact that every vendor relationship is under a microscope right now. Customers are walking into renewals already knowing their answer, and they're not telling you.

Even the best CSMs in the world aren't catching everything. You need a better system, but how do you build one that doesn't depend entirely on manual analysis? 

What "at risk" signals look like

Before a customer churns, they leave breadcrumbs. Signals are there, but it’s hard to understand which ones matter for your customer base, industry, region, and company size. A mid-market fintech account will behave differently than an enterprise healthcare customer. “Normal” behavior for one type of industry is often a red flag for others. 

Signs a customer is heading for the exit:

  • Login frequency declines, relative to their segment’s baseline
  • Feature adoption is narrowing, and they're using less of the product over time
  • Support tickets are spiking or remain unresolved
  • Your core champion left or disengaged, and nobody's stepped in to replace this role 
  • CSM outreach is getting slower responses, or none at all

Signs a customer is ready to grow:

  • Usage approaches the customer’s seat or plan limits
  • They're exploring advanced features and getting more sophisticated
  • New teams are being onboarded inside the account
  • NPS is strong and support sentiment is positive

And while most of us know these things, tracking all of this manually, across a full book of business, is where things break down. No matter how hard you try, you’re still missing what’s about to happen. 

Why DIY’ing your churn prediction model isn’t the answer

A lot of teams look at this problem and think, “we'll just vibe code our own churn model.” But dashboards and models aren’t the same thing: dashboards visualize past behavior, whereas predictive models ingest thousands of behavioral signals, trains on historical outcomes, continuously recalibrates itself, and generates forward-looking risk scores. 

To build that properly, you need:

  • Large volumes of product usage data
  • CRM, support, billing, and warehouse integrations
  • Feature engineering across thousands of variables
  • Ongoing model validation and retraining
  • Infrastructure to operationalize outputs

It often takes teams months of work, dedicated data science resources, and it’s still stuck within BI tools. And the biggest challenge of them all comes down to getting sellers and CSMs to change their behavior based on these insights.  

That’s why Pendo Predict ingests your entire product usage dataset, embeds account health signals directly into CRM workflows, flags changes the moment they happen, and triggers next-best actions for CS and sales teams. The model works because it’s operationalized, meets users where they’re already working, and is constantly re-tuning itself. 

You don’t need Pendo Analytics to use Predict

You can use Pendo Predict, even if you’re not using Pendo Analytics today. And if you’re already using Pendo, it’s even faster to get started.

Your behavioral data is already unified and structured, speeding up model deployment and insight generation. It’s also connected to other parts of Pendo, so you can:

  • Trigger automated retention and upsell workflows with Orchestrate
  • Build and segment in-app guides to drive adoption within at-risk accounts
  • Watch replays and drill into user feedback from at-risk accounts 

NRR is a reflection of whether your customers are getting more value over time.

To win today, use predictive signals to intervene early, expand intentionally, and operationalize growth across every account. Get a demo of Predict.