A customer health score built on behavioral signals, and acted on weekly, is the closest thing to a churn prediction system most SaaS founders can realistically build.
The idea behind a SaaS customer health score is simple enough: instead of waiting for a customer to cancel, you build a number that tells you they're about to. Simple in concept, hard to get right in practice. Most founders build a health score, watch it for a few weeks, and quietly stop looking at it when it fails to predict anything useful. The problem is usually not the score itself. It's that the wrong signals went in, or that nobody decided what to do when the score turned red.
The instinct is to lean on NPS surveys. They feel like customer data. They're not, really. They're customer opinion, collected monthly if you're lucky, and they tell you how someone felt on the day they filled it out. Behavior is harder to fake and harder to ignore. The signals that actually predict churn at scale are the ones Intercom and HubSpot have both confirmed through their own customer success work: login frequency, feature adoption depth, and whether the customer is actually consuming what they paid for.
At Intercom, the internal framing centers on "time to value," specifically how quickly a new account reaches the moment where the product is genuinely doing something for them. Accounts that hit that moment within the first two weeks are dramatically less likely to churn at the 90-day mark. That pattern is concrete enough that their CS team built a formal threshold around it. If a new customer hasn't sent their first message via Intercom within 14 days of signup, the account gets flagged. One behavioral trigger replaced a lot of manual gut-checking.
HubSpot's approach, documented through their CS training materials and their INBOUND conference sessions, weights feature adoption differently depending on where a customer sits in their contract cycle. In month one, breadth matters most: are they exploring the platform? In month eight, heading toward renewal, depth matters more: are they embedded across teams? A customer using five tools shallowly is less healthy than one using two tools deeply, even if the first looks more active on a surface-level usage metric.
Building the score: signals and weights
Start with three to five signals. More than that and the score becomes a noise machine that tells you everything and predicts nothing. A sensible starting stack for most B2B SaaS products: weekly active usage, the number of core features adopted, support ticket volume (consistently high volume from a single account is a warning sign, not a sign of engagement), and contract utilization, meaning whether the customer is actually using the seats, API calls, or storage they paid for. If your product has a clear activation event, whether that's a first integration set up, a first report generated, or a first campaign sent, add it as a binary signal and weight it heavily in the first 60 days.
Weight each signal on a 0-to-100 scale, then apply multipliers that reflect how predictive each one is for your own churn data. If you don't have churn data to calibrate against yet, start with equal weighting and adjust as patterns emerge. Document why you picked each weight. You'll want to revisit them in 90 days when you have actual renewal outcomes to compare against.
A composite score under 40 should trigger a CS touchpoint within the week. Between 40 and 60, the account goes on a watch list. Above 70, it's healthy and probably doesn't need proactive attention beyond normal check-ins. Calibrate these thresholds against your own data over time. Starting somewhere specific is better than a vague sense that something seems off.
The part most teams skip
Building the score is the easy part. Making sure someone actually acts on it is harder.
A health score sitting in a dashboard nobody checks is a more sophisticated way of not knowing your customers are about to leave. The CS team at Intercom runs a "red account review" weekly: every account below their health threshold gets a named owner and a specific next action before the meeting ends. Not a follow-up email to consider sending. A next action, meaning a call booked, a feature walkthrough scheduled, a conversation with the account executive about whether there's a commercial issue underneath the usage drop.
For lean teams, a founder running CS themselves or a single CSM covering 80 accounts, that weekly review can't be that formal. But the principle holds. Pick a threshold. Commit to a response. When an account crosses below it, something happens that week, not eventually.
A health score will occasionally flag an account as at risk when the customer has no intention of churning. That's fine. The cost of a proactive check-in on a healthy account is low. The cost of missing an account that's quietly heading for the door because it looked fine on a vanity metric is not.
What good looks like at scale
Totango, the customer success platform, has published research showing that companies with a formalized health score process see net revenue retention roughly 15 percentage points higher than those without one. For a SaaS business doing $2 million in ARR, that's potentially $300,000 in retained revenue annually. That gap doesn't close by accident.
The more useful number, though, is the one your own data will eventually give you: the correlation between a specific score threshold and actual churn at 90 days. Once you have that, you have something to act on with real precision. Until then, you're working from educated proxies. That's still worth doing. An educated guess acted on beats a perfect model nobody uses.
Gainsight, which powers CS operations at companies including Salesforce and Box, incorporates what it calls "sponsor tracking" alongside standard behavioral metrics. The logic is straightforward: monitor whether the internal champion at a customer account is still active in the platform. It matters because one of the most common causes of churn isn't bad product experience. It's the power user who bought your product leaving the company, and the new stakeholder not knowing why they're paying for it. If your product has a defined power user who isn't the billing owner, that relationship signal belongs in your score, and most lean teams never think to include it.
Don't wait until the model is perfect before you start using it. Build something honest, calibrate it against real outcomes, and make sure it drives action. The companies that predict churn well aren't the ones with the most sophisticated models. They're the ones where a named person looks at the score every week and picks up the phone.
Also read: How a SaaS Onboarding Email Sequence Turns Trial Users Into Paying Customers • How to Build a SaaS Affiliate Program That Actually Compounds Revenue • The SaaS Free Trial Conversion Playbook Most Founders Ignore