This is a research-evidence companion to our guides on product-qualified signals and expansion and customer value realization. Start there for the operating playbook; this piece checks what the evidence actually says about which usage metric predicts renewal.
SaaS companies track dozens of product usage metrics, login frequency, feature depth, seat utilization, time-to-value, to predict which accounts will renew. The most detailed available research on this exact question found that none of them, tested individually, reliably predicted renewal once relationship and pricing signals were in the model. The academic literature on this specific question is thinner than the confidence of vendor benchmarks suggests, and the most common practical failure isn’t picking the wrong usage metric, it’s that most companies’ usage data isn’t clean enough to use in the first place.
The Vendor Claims vs. What’s Actually Been Tested
Search for which usage metric predicts SaaS renewal and product-analytics vendors answer confidently: accounts that adopt five or more features show 60 to 80 percent lower churn; features with repeat use inside the first seven days show over 3x higher 90-day retention. These numbers circulate widely across product-analytics and customer-success vendor blogs, without a published methodology, sample size, or independent study behind them, the same pattern this series has already found behind the “208% more revenue” sales-marketing alignment stat and Forrester’s “107% ROI” customer success claim.
The underlying academic literature is real but genuinely thinner here than for most of the topics in this series. Product usage metrics vary by platform, licensing model, and industry in a way that resists standardization, which is exactly why a systematic literature review, not a quantitative meta-analysis, is the realistic academic format for this question: there isn’t yet a large enough base of comparable, published effect sizes to pool.
Renewal Rate, Retention Rate, and NRR: What Usage Metrics Are Actually Trying to Predict
Before asking which usage metric predicts renewal best, it helps to be precise about what “renewal” actually means as a target. SaaS companies track several related but distinct rate measures, and usage-metric research often blurs them together.
Renewal rate, sometimes called logo renewal rate for a saas renewal rate measured on account count rather than dollars, tracks the percentage of customers who renew their contract, a simple count of accounts regardless of contract size. Retention rate is closely related but is sometimes calculated on a dollar basis instead, tracking how much revenue survives the renewal cycle rather than how many accounts do. Churn rate is the inverse of retention rate: the percentage of customers, or revenue, lost. Net revenue retention (NRR) goes a step further, netting expansion revenue (upsells, seat growth) against contraction and churn, so a company can, in principle, post 100%+ NRR even while losing some accounts, if the customers who stay expand enough to offset the ones who leave.
This distinction matters for usage-metric research specifically, because a usage signal that predicts logo-level churn well might say nothing useful about NRR, and vice versa. A customer expanding seats, a usage signal that looks healthy, can still be a churn risk on the underlying use case if the expansion is driven by one enthusiastic team while the rest of the account disengages, exactly the kind of nuance a single usage number collapses away. The Lime Technologies case study below measured a straightforward binary renewal outcome, not NRR, a reasonable scope for a case study, but it means the findings say more about who stays than about how much revenue a “stay” is actually worth.
For a SaaS business trying to build a repeatable, automated renewal process, the practical implication is to pick the target metric first, logo renewal rate for volume-driven businesses, NRR for expansion-driven ones, and then evaluate any usage metric against that specific target. A high renewal rate paired with flat or declining NRR is a common trap this distinction helps catch early.
The Most Detailed Real-Company Test of This Question
The clearest available answer comes from a 2025 Aalto University master’s thesis built on real, anonymized CRM data from Lime Technologies Finland, a B2B CRM software vendor, covering 337 customers and data from 2019 to 2024. The study compared five machine learning models (Logistic Regression, SVM, Decision Tree, Random Forest, and XGBoost) on their ability to predict churn, then used Permutation Feature Importance and SHAP analysis to identify which individual features actually drove the predictions, not just which features the models happened to use most often internally.
XGBoost was the strongest model, catching every true churner in the test set (recall of 1.0) with an AUC-ROC of 0.988. But the more useful finding for this question is what happened when usage-specific features were tested individually. The dataset’s own “usage” metric (login activity relative to licenses purchased) ranked last among the top 10 features by internal model importance, and showed no statistically significant contribution in the permutation analysis (p ≈ 0.39). The broader Customer Health Score total, which blends usage with several other signals, ranked moderately (3rd by internal importance) but also wasn’t statistically significant on its own once the strongest predictor was accounted for.
That strongest predictor was relation_strength, a composite relationship metric built from support-interaction frequency and stakeholder engagement, not usage. It dominated every measure the study ran: highest internal importance by a wide margin, the only feature with a strongly statistically significant permutation score (p ≈ 0.002), and the clearest directional SHAP pattern. The second-most-important feature in the permutation analysis wasn’t a usage metric either, it was average license rate, a pricing signal, which the study’s authors note is notable precisely because it hadn’t even ranked in the model’s own top 10 by internal importance, showing up only when tested for its real contribution to prediction accuracy on unseen data.
Perhaps the most practically useful finding didn’t come from the modeling at all: two raw usage variables the researchers originally wanted to test, unique logins and the number of licenses actively in use, had to be dropped from the dataset entirely before modeling even began, because too much of the data was missing. This is a data-quality problem, not a methodology problem, and it’s a strong candidate for the actual reason “which usage metric predicts renewal” is hard to answer in practice: many companies’ own usage telemetry isn’t reliable enough to test.
The Academic Literature Confirms the Question Is Still Open
A 2024 paper in Industrial Marketing Management by Sanchez Ramirez, Coussement, De Caigny, Benoit, and Guliyev, “Incorporating usage data for B2B churn prediction modeling,” tackles this exact question directly, structuring usage data from roughly 3,959 B2B software subscriptions into features across three dimensions (timing, granularity, and domain-specific derived measures) to test how much usage data actually improves churn-prediction performance. The paper is closed access, so this article can confirm its existence, authors, and research framework, but can’t independently verify its specific effect-size findings beyond what’s summarized in its abstract, so treat any exact numbers you see cited elsewhere from it as something to confirm at the source before repeating.
What both papers agree on structurally is the framing: usage data is worth structuring carefully and testing rigorously, not assumed to work by default. Neither treats “install a usage-tracking tool” as sufficient on its own.
Why “Which Usage Metric” Might Be the Wrong Question
The proprietary-metric problem is real and structural, not just a data-quality inconvenience. A “feature” in a CRM, an “active seat” in an HR platform, and a “session” in an analytics tool aren’t measuring the same underlying behavior, which is exactly why usage-metric research resists the kind of standardized meta-analysis that switching costs or sales-marketing alignment research supports. Every company’s usage telemetry is a custom instrument, calibrated to its own product, which limits how far a benchmark from one vendor’s blog post can travel to a different company’s product.
That’s also consistent with what our customer value realization research already argues from the practitioner side: high product adoption doesn’t guarantee a renewal, because usage alone doesn’t confirm the customer has connected that usage to a business outcome they’d pay to keep. The Lime Technologies case study’s own finding, that a relationship-strength metric beat every usage signal, and that pricing beat usage too, is the quantitative version of the same point.
Building Usage Patterns Into a Renewal Process, Not a Standalone Score
Even with the caveats above, usage patterns aren’t worthless, they’re just insufficient alone. The practical best practices that hold up against the evidence look less like “track this one saas metric” and more like building usage into a broader renewal process that also weighs relationship and financial signals.
For an existing customer, the useful question usage data can answer isn’t “will this account renew” in isolation, it’s “is this account’s usage pattern consistent with the outcome they bought the saas product for.” A sudden drop in a previously steady engagement metric is a legitimate trigger to track renewal risk, not because the drop itself proves churn risk, but because it’s a cheap, automatable signal that something changed and a human should look. That’s a different, more defensible use of usage patterns than feeding a single number into a scoring model and trusting the output as one of your key metrics on its own.
For enterprise SaaS companies with dedicated customer success teams, this usually means routing usage-pattern alerts to a CSM for a manual check, not automating a retention action off usage data alone. For smaller SaaS businesses without that capacity, combining a light usage-pattern trigger with a billing and financial-metrics check, is the account’s recurring revenue and license rate stable, gets closer to what the evidence above actually supports: usage plus a pricing or relationship signal, not usage in isolation.
None of this replaces having actual renewal strategies and retention strategies built around real customer retention, not just usage monitoring. Usage data can flag an account worth a conversation; it can’t have the conversation, and it can’t substitute for the value-realization work our customer value realization research covers. The saas companies that improve renewal outcomes over time tend to be the ones that treat a usage-pattern alert as the start of a retention motion, not the end of one, checking whether the account’s revenue growth or conversion rate on new seats reflects genuine expansion or a false positive from an unusually active single user.
What This Means for Tracking Renewal Risk
- Audit your own usage-data quality before trusting it. If your platform can’t reliably tell you who logged in last week, a usage-based renewal model built on that data is unreliable by construction, not by choice of metric.
- Don’t adopt a vendor’s usage threshold uncritically. “5+ features = lower churn” is a claim, not evidence, until you’ve tested it against your own churned-account history the way the case study above tested its own metrics.
- Treat usage as one input, not the model. The evidence available says a well-built HubSpot health score blending relationship, pricing, and usage signals will outperform any single usage number used alone; see our broader research on what a health score should actually measure.
- Weight relationship and contract signals at least as heavily as product usage. Support-interaction frequency, stakeholder breadth, and pricing/license terms carried more verified predictive weight in the available research than any individual usage metric did.
- Use product-qualified signals for expansion timing, not as a renewal guarantee. Usage-based expansion triggers are a different question than renewal risk, don’t let a PQL score substitute for actually checking whether the account is at risk.
- Validate any model against your own churned and renewed accounts. The strongest finding across every study cited in this series is the same one: internal validation against your own data beats trusting a published benchmark that was never run on a company like yours.
- Track logo renewal rate and dollar retention separately, not just one blended number. The distinction above between renewal rate and NRR isn’t academic, it changes which usage signals are even worth testing against your specific target.
Frequently Asked Questions
Which product usage metric best predicts B2B SaaS renewal? The most detailed real-company test of this question found none of the individual usage metrics tested (login-based usage, feature stickiness, composite health-score total) were statistically significant predictors once relationship-strength was in the model. A pricing/contract signal, average license rate, outranked every usage-only metric.
Is the “5+ features adopted = lower churn” statistic real? It is a widely repeated vendor benchmark (Gainsight, ChurnZero, and similar product-analytics vendors), not a cited, peer-reviewed finding. No public methodology, sample, or study backs the specific percentages in circulation, the same pattern as other repeated B2B stats this series has checked.
Why is usage data so hard to use for churn prediction in practice? Because it’s often unreliable at the source. In the most detailed available case study, two raw usage variables (unique logins and licenses in active use) had to be dropped entirely from the dataset due to a high proportion of missing values, before any modeling even started.
Does this mean product usage data doesn’t matter for renewal? No. It means no single usage metric, tested in isolation, has been shown to reliably predict renewal on its own. Usage data still has a role as one input in a broader model, alongside relationship and pricing signals, not as a standalone renewal predictor.
What should a SaaS company track instead of chasing a single usage metric? Relationship-strength signals (support contact frequency, stakeholder breadth, recency of interaction) and pricing/contract signals showed real predictive power in the available research. Treat usage telemetry as a supporting input once its own data quality is verified, not the primary signal.
This piece pairs with our guides to product-qualified signals and expansion, customer value realization, and our research on customer health scores and churn.
Most product analytics tools sell certainty about which usage metric matters. SWOTBee builds renewal and health-score systems on what the evidence and your own account history actually show, not a vendor’s unverified benchmark.
