Health Scores and Segmentation, Without the Astrology

Every Customer Success team eventually builds a health score. It is usually green, yellow and red, it sits at the top of the account list, and it gets quoted in meetings with real confidence. Ask how it is calculated and the room goes quiet. Someone knows there is a usage component. Someone remembers a survey feeding into it. Nobody can say what would have to change for a red account to turn green.

That gap is the whole problem. A health score is a model, and a model nobody can explain is astrology with a dashboard.

So start with what the thing actually is. A customer health score is a single number built from several weaker signals, each put on a common scale and combined with weights you choose on purpose. The formula is a weighted sum: take each signal, multiply it by how much you have decided it counts, add them up. That is it. There is no statistics degree hiding in there, which is exactly why the discipline has to come from somewhere else, the same way it has to come from writing your metric definitions down.

A score answers one question, not five

The most common design mistake is asking one number to predict churn, expansion, advocacy, implementation failure and support escalation at the same time. Those five things have different causes and different early signals. A customer can be deliriously happy and still fail to renew because the budget holder left. An account can be growing and quietly furious.

Pick the decision the score exists to serve. Which customers need intervention this week? Which are ready for a value review? Which look like expansion candidates? One score per decision is not extra work. It is the difference between a signal and a mood ring.

What goes into it

An illustrative starting point, and not a benchmark, might weight outcome attainment at 0.35, adoption at 0.30, relationship strength at 0.20, support experience at 0.10 and commercial stability at 0.05, with each of those normalized to a 0 to 100 scale. Outcome attainment means progress against what the customer bought you to achieve, which is the top of the chain that runs from promise to renewal. Adoption means habitual use of the workflows that produce that outcome. Relationship strength means the right people are still engaged and still confident. Support experience captures friction. Commercial stability covers contract, payment and budget behavior.

The order of those weights is the opinion I will defend. Outcome goes first, adoption second, and everything else is supporting evidence. Most health scores in the wild are usage scores in a costume: they weight product telemetry heavily because product telemetry is the easiest data to get, then treat the resulting number as a statement about value. Ease of collection is not evidence of importance.

Weights are not universal, and they are not permanent

The same five inputs should carry different weights at different points in the relationship. During implementation, onboarding milestones and technical readiness deserve most of the weight, because nothing else is observable yet. Near a renewal, verified value and stakeholder confidence matter more than last month's login count. For a consumption platform, growth in valuable workloads tells you far more than seat utilization, which may be structurally meaningless.

Segment changes the weighting too. Planhat's own health guidance makes the same point: build the score around your specific business and connect it to lifecycle playbooks, rather than adopting a template because it came with the software.

The part teams skip

Launching a health score is the easy half. The rigorous part comes afterward, when you backtest it against what actually happened: renewals, churn, contraction and expansion in the accounts you already scored.

Five questions do most of the work. Do red accounts churn more often than green ones, at a rate you can see? Does the score turn red early enough that someone could still act on it? Does one component dominate the total, making the other four decorative? Do the manual inputs a CSM types in simply encode that CSM's renewal forecast, so the score is a forecast with extra steps? And does missing product data systematically penalize one segment, so your worst-instrumented customers always look sickest?

A score that cannot survive those questions is not a predictive model. It is a workflow convenience, which is a perfectly respectable thing to be. Use it to sort a queue. Just stop citing it as evidence in a board meeting, because at that point it is decorative arithmetic.

Segmentation is the other half of the same problem

Health scoring tells you when to act. Segmentation decides how much service an account gets in the first place, and most teams decide it with one variable: annual recurring revenue.

ARR is a reasonable proxy for what you lose if the customer leaves. It says nothing about what it costs to make that customer successful. A thirty thousand dollar customer with complex integrations, regulated workflows and no internal expertise can consume more Customer Success capacity than a three hundred thousand dollar customer with a mature internal center of excellence. Segment on revenue alone and you will systematically underserve the accounts that need the most help while over-serving the ones that need the least.

A better model treats the service tier as a function of several things at once: economic value, growth potential, complexity, the effort success will actually require, lifecycle stage, and how the customer prefers to be served.

What comes out the other side is a set of service designs rather than a ranking. Strategic accounts get a dedicated senior CSM with specialist access, and the risk to manage is endless bespoke work. Enterprise accounts get a dedicated CSM, a structured cadence and digital support, with the risk that relationship activity quietly replaces outcome work. Mid-market runs on pooled or higher-ratio coverage plus digital programs, and the risk is too little personalization. The long tail runs digital-first, one-to-many and community, where the risk is treating automation as a substitute for service rather than a form of it. Then there is the case that breaks the grid: moderate economic value, high deployment risk, which needs temporary specialist coverage and a date on which that coverage ends.

GitLab is a useful public example here, because its success offerings are documented rather than described in a webinar. They range from on-demand pooled resources to designated support structures, and a separate Digital Success team owns scalable adoption and outcome programs for the long tail. You can disagree with the specific design and still learn from the fact that it is written down.

Why these two belong together

Health scoring and segmentation run on the same signals and answer adjacent questions. Segmentation sets how much attention an account is entitled to. Health scoring sets when that attention gets spent, and on what. Get the first one wrong and your capacity goes to the wrong accounts. Get the second one wrong and it arrives too late to matter.

Neither of them creates value. The chain does that: a promise made, adoption of the workflows that deliver it, an outcome the customer can verify, value they can point to, a renewal that follows from it, and growth after that. A health score's only job is to tell you which link is breaking, in which account, while there is still time. A segmentation model's only job is to make sure someone is watching. That is aiming attention, not measuring success, and aiming attention is most of what a Customer Success operating model is for.