The short version: a health score is a leading indicator, not a report card. It exists so a CSM knows which five of thirty accounts are drifting before renewal day, not after.
A customer health score is a composite indicator that predicts whether an existing customer is likely to churn, stay, or expand. It blends usage data, executive sponsorship, support ticket volume, NPS, payment status, and product-feature adoption into a single 0 to 100 number, then bands that number into red, yellow, and green so customer success teams know who to call first. The best scores are validated against actual churn, not vibes.
Six things to understand before building or buying a health score. Each one is the difference between a model that predicts renewal and a model that becomes a dashboard nobody opens.
A health score rolls six to ten raw signals into one scalar between 0 and 100. The point is compression: a CSM cannot watch twelve dials per account across a book of eighty, so the score becomes the dial they watch. The individual signals are still visible underneath, but the top-level number is what triggers the next action.
The job of the score is to answer two questions early. Will this account renew. Will this account expand. A high score plus heavy feature adoption is an upsell conversation. A low score plus falling logins is a save play. The score is only useful if those actions change based on the number.
The common inputs cluster into five families. Usage (logins, active users, feature depth). Relationship (executive sponsor present, number of champions, admin on the account). Support (ticket volume, severity, resolution time). Sentiment (NPS, CSAT, survey responses). Commercial (invoices paid on time, contract value, product-feature adoption against entitlements).
A score of 73 means nothing to a CSM on a Monday morning. A yellow account means something. Bands turn the number into a decision: red means intervene now, yellow means check in this week, green means log the win and look for expansion. The bands are where the model stops being a report and starts being a playbook.
The uncomfortable test: run the score backward against last year's churned accounts. Were they red before they left? If yes, the model is doing its job. If churned accounts were green until the cancel call, the model is theater. Most first drafts fail this test, which is why health scoring is iterative, not one-and-done.
A score is only as useful as the automation that reacts to it. When green flips to yellow, the CRM creates a task, pings the CSM, and queues a check-in sequence. When yellow flips to red, it escalates to the manager and starts a save play. Without the automation layer, the score is a vanity dashboard.
A health score is a weighted combination of raw account data. The combinations differ by product, segment, and contract type, but the underlying signals almost always come from the same five families below. Start with the signals your product already generates, prove the model works at a small scale, and only then expand the input set. The biggest mistake is launching with twenty inputs and no way to tell which one is actually predictive.
How often people in the account open the product, how many distinct users are active per week, and how many of the core features they actually touch. A deal that sold ten seats and has three weekly actives is a different account than a deal that sold ten and has nine. Usage is the single strongest predictor across most SaaS models.
The deal was sold to a specific human. If that human has left, moved teams, or stopped responding, the renewal is in trouble regardless of what the usage data says. Track sponsor presence as its own signal, with a hard drop in the score when the champion leaves the company or changes role.
Open tickets are a signal, but the shape of the ticket history matters more than the count. A quiet account is not necessarily a happy account, and a loud account is not necessarily unhappy. Weight recent severity-one tickets, slow resolutions, and repeat issues on the same feature higher than raw ticket count.
Direct sentiment data from Net Promoter Score surveys, post-support CSAT, and in-app pulse surveys. Weight this input lower than people expect, because response rates are low and respondents skew. Still, a detractor on the primary admin seat is a stronger churn signal than almost any other single input.
Finance is a leading indicator people underuse. An account that has started paying late, requested payment-term changes, or asked finance questions about the contract is signaling churn risk before the CSM hears about it. Pull invoice-aging data into the score so the commercial trajectory is visible alongside the product trajectory.
The deal was sold on a set of outcomes tied to specific features. If the account is only using one of five entitled features, the ROI story is weak and the renewal conversation gets hard. Track adoption against the sold value, not just the technical login count, so the score reflects what was promised at purchase.
Once the inputs are chosen, each one needs a weight. Weighting is where most health score projects stall, because teams either make the weights up (the model is a guess) or try to let a machine learn them without enough data (the model overfits). Below are three strategies that work at different company sizes, from a ten-account book to a thousand-account book. Start simple, prove the model, then graduate.
The simplest workable model. Choose five inputs you trust, give each one 20 percent of the score, and ship. The strength is explainability. A CSM can look at a yellow account and see exactly which signal dragged it down. Flat weights are usually right for the first ninety days of health scoring at any company.
The best CSM on the team already has a mental health score, they just have not written it down. Interview them, extract the weights (usage matters more than NPS, sponsor presence matters most, late payments trump support tickets), and codify the result. This is faster than training a model and more accurate than flat weights.
Once there are at least a hundred renewal outcomes on record, a model can learn which signals actually predicted churn. The output is a set of weights grounded in evidence instead of intuition. The risk is overfitting on a small sample, so the learned weights should be reviewed by humans before replacing the expert weights.
Not every signal is additive. A late invoice, a departed champion, or a severity-one ticket in the last thirty days should pull the score down, not just fail to pull it up. Build negative weights into the model from day one, so bad news lowers the score the way good news raises it.
A login from eight months ago is not the same as a login from Tuesday. Apply time decay so recent signals dominate the score. The decay curve depends on the product cycle, but a thirty-day half-life is a reasonable starting point for most B2B SaaS. Without decay, long-tenured accounts look healthier than they are.
The score needs to translate into action. Bands are the handoff from analytics to playbook. Three bands are enough for almost every team, and the thresholds should be chosen so each band contains a workable share of the book, not a mathematically pretty cutoff. The playbooks below are the ones most customer success teams converge on once the scoring is in place.
The account is using the product, the champion is engaged, invoices are paid, and the recent trend is up or flat. The CSM playbook is advocacy and expansion: ask for the case study, introduce product-marketing to the champion, test the upsell conversation on the next business review. Green accounts fund the growth motion.
Something is drifting. Usage is slipping, a ticket volume is up, the champion has gone quiet, or a payment came in late. The playbook is a proactive check-in, a usage review with the admin, and an executive-level touch if the sponsor is involved. Yellow is where health scoring earns its keep, because the save is still cheap.
The account is actively on the churn path. The CSM, the account executive, and often the VP of CS or the sales leader are involved. The playbook is a direct, named conversation with the sponsor, an audit of value delivered against the contract, and a renegotiation path if the fit is wrong. Not every red saves, but every red deserves a real attempt.
A green account trending down is more interesting than a yellow account trending up. Show the thirty-day trend next to the current score, and route alerts on direction, not just band. A seven-point drop inside green is worth a check-in. A ten-point rise inside red is a save that is already working.
A health score that nobody validates and nothing automates is a dashboard, not a system. Validation closes the loop by checking whether the score actually predicted the renewal outcome. Automation closes the loop by making sure a band change becomes a task, a notification, or an escalation inside the CRM instead of a line on a weekly report. Both are required for the score to earn its slot in the stack.
The simplest validation. Pull every account that churned in the last twelve months, compute what their score would have been thirty, sixty, and ninety days before cancel, and check the band distribution. A healthy model shows a clear drift from green to red over the ninety-day window. A broken model shows no signal until the cancellation lands.
Red accounts should churn at a dramatically higher rate than green accounts. If reds churn at 35 percent and greens churn at 32 percent, the score is not separating risk. If reds churn at 40 percent and greens churn at 3 percent, the model has signal. The ratio between band churn rates is the single clearest validation metric.
When an account flips from green to yellow, the CRM creates a task on the CSM, logs the trigger, and optionally queues a nurture sequence. When yellow flips to red, the manager is pulled in and an escalation playbook starts. The automation layer is where health scoring stops being analytics and starts being customer success work.
Leadership wants to know whether the book is getting healthier or sicker. The reporting view is the band mix on a thirty-day rolling basis, split by segment and by CSM. A rising share of green is a leading indicator for renewal performance. A rising share of red is a hiring, product, or pricing signal, depending on where the reds concentrate.
Strkr AI reads the account record, the usage data, and the ticket history, suggests weighting changes when churned-account patterns shift, and flags accounts whose score looks fine but whose qualitative signals (support thread tone, meeting cadence, sponsor replies) say otherwise. The score stays yours. The AI acts as a second reader, not the owner.
Products change, segments shift, and the signals that predicted churn last year may not predict it this year. Review the inputs and weights every quarter against the latest renewal data, and resist the urge to add inputs faster than you can validate them. A small, tested model beats a sprawling untested one every time.
Strkr pulls usage, support, finance, and survey signals into one editable health score, bands it into red, yellow, and green, and triggers CRM playbooks on every band change. Pricing is published. The platform tour shows exactly what ships today.
A customer health score is a composite indicator between 0 and 100 that combines several raw signals (usage, executive sponsor, support tickets, NPS, payment status, feature adoption) into one number designed to predict churn risk and expansion potential. Customer success teams use the score to prioritize which accounts to work on, with red, yellow, and green bands triggering specific playbooks.
The common inputs cluster into five families: usage (logins, active users, feature depth), relationship (executive sponsor presence, champion count), support (ticket volume, severity, resolution time), sentiment (NPS, CSAT, survey responses), and commercial (invoices paid on time, feature adoption against entitlements). Most teams start with four to six inputs and expand only after validating the first version.
Start by choosing five to six inputs you trust and giving each one a flat weight (20 percent each if five, 16 percent each if six). Add negative weights for late payments and departed champions so bad news lowers the score. Apply recency decay so last week counts more than last year. Band the result into red (0 to 39), yellow (40 to 69), and green (70 to 100). Validate against last year's churn before relying on it.
There is no single right formula, because the signals that predict churn vary by product and segment. The best formula is the one that passes a backtest: churned accounts should have shown as red or yellow at least thirty days before cancellation, and the churn rate inside red should be dramatically higher than inside green. Start simple with flat weights, prove the model, then graduate to learned weights once there are a hundred renewal outcomes on record.
Run the score backward against last year's churned accounts and check whether they trended red before leaving. Compare the churn rate inside each band: reds should churn at many times the rate of greens. If the band churn rates look similar, the model has no signal and the inputs or weights need to change. Backtesting on real outcomes is the only honest check on whether the score is predictive or decorative.
NPS is one input, a health score is a composite. NPS tells you what a single respondent said on a single survey. A health score blends the survey signal with usage, support, sponsorship, and commercial data, so it keeps working between surveys and reflects the full account state. Most mature customer success teams treat NPS as one of five to ten inputs feeding the score, not as the score itself.
Bands turn the numeric score into a decision. Green (70 to 100) is a healthy account, often an expansion candidate. Yellow (40 to 69) is drifting and needs a proactive check-in this week. Red (0 to 39) is on the churn path and needs an active save play. The thresholds are tunable so each band contains a workable share of the book, and the band change (not the point change) is what triggers CRM automation.
Strkr ships a health score object that pulls usage, support, finance, and survey signals from the records already in the CRM. Weights are editable per segment, bands trigger tasks and sequences on change, and Strkr AI suggests weighting adjustments when churned-account patterns shift. The score is auditable, the playbook is automated, and the model is validated against real renewal outcomes instead of guessed into a dashboard.
No sales call, no migration consultant, no four-month implementation. Enter your card, get 14 days of the full Pro tier, cancel any time before day 14 with zero charge. Spin up a workspace, import your CSV, and have something useful before lunch.
