Most growth teams track more metrics than they act on. Monthly active users, feature adoption rate, login frequency, support ticket volume, NPS quartile, email open rate: the list compounds with every tool added to the stack. None of that is wrong to measure. But if you want to know whether the business is getting healthier or sicker, you need three specific numbers, and most teams are not computing all three correctly.
This is not an argument for ignoring everything else. It is an argument for knowing which numbers make the other numbers interpretable, and for getting those right first before adding complexity.
Metric 1: Net Revenue Retention
Net revenue retention (NRR) answers the question: if we signed no new customers starting today, would our revenue grow, shrink, or hold flat? It captures expansion revenue from existing accounts minus contraction and churn, expressed as a percentage of the starting MRR from that same cohort.
An NRR above 100% means your existing customer base is growing on its own. Below 100% means you are losing ground inside accounts you already won. For a SaaS company or subscription business, NRR is the single best predictor of compounding revenue growth, because it tells you whether your product creates more value over time for the customers who stay.
The calculation most teams get wrong: they include new logo revenue in the numerator. NRR should be computed only on the cohort that existed at the start of the period. New customers are a separate question. Mixing the two inflates the number and hides contraction in existing accounts.
Where Segmentvue fits this metric: we compute NRR per acquisition segment, not just in aggregate. A team running this on their own event data found that their self-serve signups had an NRR of 87%, dragging down an aggregate NRR that looked healthy at 103%. The high-touch accounts were expanding; the self-serve segment was quietly contracting. Aggregate NRR had masked a real problem for six months.
Metric 2: Segment-Level Churn Rate
Aggregate monthly churn rate is almost useless as a management number. A 3% monthly churn rate looks the same whether 3% of every segment is churning uniformly, or whether one segment is churning at 12% and three others are churning at 1%. Those situations require completely different responses.
Segment-level churn rate means computing churn separately for each behavioral or demographic segment you care about: acquisition source, plan tier, usage intensity bucket, industry vertical, or onboarding cohort. The goal is to find the segments with anomalously high churn so you can understand why before it compounds.
The practical challenge here is that useful segments are behavioral, not just demographic. Customers who logged in fewer than three times in their first 14 days have a materially different churn rate than customers who ran their first workflow in the first 48 hours. That difference is buried if you only slice by plan tier or company size.
We built Segmentvue's segmentation layer specifically to make behavioral slicing easy without requiring a data analyst to write a new SQL query for each hypothesis. The segments update daily as behavior changes, so when a cohort's engagement pattern shifts, the churn rate for that behavioral segment updates automatically rather than waiting for the next monthly report.
One thing to be clear about: segment-level churn rate is a diagnostic metric, not a target. You should not be optimizing segment-level churn directly. You should be using it to find which segments need intervention, then designing the intervention based on what you know about why those customers are disengaging.
Metric 3: LTV-to-CAC by Acquisition Cohort
LTV-to-CAC is the most misused metric in growth because it is almost always computed in aggregate, which makes it nearly meaningless for decisions. A company-wide LTV-to-CAC ratio of 4:1 tells you almost nothing about whether your paid social acquisition from last quarter is contributing positive unit economics, or whether the channel partners you onboarded in Q2 will pay back their acquisition cost.
Computing LTV-to-CAC by acquisition cohort changes the calculation entirely. You are now asking: for the customers who came in via channel X in month Y, what is their predicted lifetime value, and what did we spend to acquire them? That is a number you can act on. If paid search cohorts from Q3 show an LTV-to-CAC of 2.1 while organic cohorts show 6.8, you have a signal about channel allocation that aggregate numbers would never surface.
The LTV side of this requires prediction, not just historical revenue. If you are measuring LTV as actual cumulative revenue, you have to wait until the cohort has churned to know the number, which makes it useless for near-term decisions. Predicted 12-month LTV, updated as behavior evolves, gives you a number that is actionable while the cohort is still young enough to act on.
This is the harder part of Segmentvue's prediction layer to build: LTV prediction that stays accurate as a cohort's behavioral profile changes over time. A customer who started on a basic plan and upgraded twice has a different predicted trajectory than someone who started on the same plan and never expanded. The cohort-level LTV needs to reflect the current behavioral state of the cohort, not just its starting condition.
What These Three Have in Common
NRR, segment-level churn, and LTV-to-CAC by cohort are all computed on segments defined by something meaningful: behavior, acquisition source, tenure, or value tier. They are all forward-looking in some sense: NRR tells you the trajectory of the current base; segment churn tells you which parts of the base are accelerating toward exit; LTV-to-CAC tells you whether the customers you are acquiring will pay back their cost.
More importantly, all three require the same infrastructure to compute correctly. You need event-level behavioral data unified at the customer level. You need those customers tagged by acquisition source and cohort date. And you need a prediction layer that can estimate future value from current behavioral signals. Without that foundation, you can compute proxy versions of these metrics, but they will be lagging indicators rather than forward signals.
The Problem With Vanity Retention Metrics
The metrics that get reported most often in growth reviews are the ones that feel good to track: DAU/MAU ratio, feature adoption percentage, email engagement rate. These are not bad metrics. But they describe activity, not economics. A customer can be highly active in your product and still churn when their contract comes up for renewal if the product stopped solving the problem that motivated the original purchase.
We have seen this pattern in our own analysis work: accounts with above-average login frequency and feature breadth churning at renewal because the product's core value proposition shifted away from the problem the buyer cared about. Activity metrics said these accounts were healthy. NRR and predicted LTV said they were not. The economic metrics were right.
The discipline is to treat activity metrics as hypotheses about why economic metrics are moving, not as substitutes for them. High DAU/MAU is a candidate explanation for improving NRR. It is not a proxy for NRR itself.
Getting Started Without a Data Science Team
These three metrics require prediction and behavioral segmentation, which sounds like it needs a data science function to build from scratch. For a team of four or five people without dedicated analyst headcount, that is a real constraint.
The practical starting point is to get the event data piped somewhere where you can run queries against it: a basic warehouse, a CDP, whatever your stack includes. Then compute NRR and segment churn from historical data before adding the prediction layer. Those two metrics alone, computed on real behavioral segments, will surface problems that aggregate metrics have been hiding.
Once you have historical visibility, the prediction layer on top of it becomes a question of which customers' trajectories most need forecasting. For most early-stage teams, that means predicted churn risk and predicted 12-month LTV for the top 20% of accounts by revenue. That is where the unit-economics signal matters most, and it is where the cost of being wrong is highest.
Getting these three numbers right does not require running a full analytics platform on day one. It requires deciding that economic retention metrics matter more than activity metrics, and building the data infrastructure to compute them accurately before adding more metrics on top.