Customer health, beyond the vanity dashboard

Most customer health dashboards are a comfort blanket with a number on it. Big green tiles, climbing line charts, an aggregate 'health score' that sits in the eighties and reassures everyone right up until the customer's cancellation email lands and someone asks, reasonably, how an account scored 84/100 the week it churned. The answer is always the same: the dashboard was measuring the past, and the past looked fine because the customer hadn't left yet.

A health score worth the screen space does something harder. It predicts. And prediction is precisely the thing a backward-looking dashboard cannot do, because every metric on it describes a decision the customer has already made. The difference between a vanity dashboard and a useful one isn't polish or the number of tiles. It's whether the signals point forward or back.

Lagging indicators feel safe and tell you nothing

The metrics that fill most dashboards are lagging indicators — they report outcomes after they've occurred. Monthly revenue from the account: that's the result of past decisions, still flowing while the customer quietly disengages. Total logins to date: a cumulative number that only ever goes up, hiding the fact that it stopped growing two months ago. Tickets closed: a measure of your activity, not their satisfaction. Each is real, each is easy to chart, and each describes a customer who has already decided how they feel, whether or not they've told you yet.

The cruel property of a lagging indicator is that it turns red only after the outcome it measures has happened. Revenue drops when the customer churns — not before, when you could still save them. By the time your health score reflects the problem, the problem is no longer a risk you can manage; it's an event you can only mourn. A dashboard built entirely from lagging indicators isn't an early warning system. It's an obituary that updates in real time, and its reassuring green is the most dangerous thing about it.

Leading indicators move before the decision

A predictive health score is built from leading indicators — the small behavioral shifts that precede a churn decision rather than confirm it. The account adopted you for one core workflow, and usage of that specific workflow is sliding even while total logins hold steady. The champion who ran the rollout has stopped showing up in the logs, and nobody's replaced them. New seats are taking longer to reach their first real outcome than the seats that onboarded six months ago. The tone of support conversations has cooled from collaborative to transactional. None of these dents revenue this month. All of them predict whether revenue survives next quarter.

What makes these signals powerful is also what makes them hard: they're early, noisy, and individually ambiguous. A dip in feature usage might be a holiday or might be the first sign of a switch to a competitor. A quiet champion might be on parental leave or might have left the company. No single leading indicator is conclusive, which is exactly why dashboards avoid them — they're harder to chart and harder to defend than a clean revenue line. But the conclusive metrics are the lagging ones, and being conclusive too late is the whole problem you're trying to solve.

A score is only useful if it triggers action in time

The purpose of a health score is not to assign a grade. A perfectly calibrated number that nobody acts on is just a more accurate obituary. The purpose is to trigger an intervention while the outcome is still changeable — to put a human in front of a wobbling account a quarter before the renewal, not the week of it. That timing requirement is the real spec. A score that updates the moment renewal approaches has failed at its only job, because the window in which a conversation could have changed the customer's mind has already closed.

This reframes what 'good' means for a health score. It's not the one with the most inputs or the prettiest dashboard. It's the one whose red flags arrive early enough that someone can still do something — reach out, fix the blocker, re-engage the champion, accelerate the stalled rollout. The score's value is measured entirely in saved accounts that the lagging dashboard would have surrendered, which means a slightly-less-precise score that fires early beats a perfectly-precise one that fires too late, every single time.

Health scoring is a triage problem

Step back and the shape of the problem is familiar: you have far more accounts than anyone can watch closely, a flood of weak and noisy signals, and a need to surface the few that genuinely deserve a human's attention right now. That's triage. It's the same structure as an inbound request queue — too much to inspect by hand, too important to ignore, and too judgment-laden for a rigid rule that says 'flag any account whose usage drops 10%' and drowns you in false alarms while missing the quiet ones that matter.

This is exactly the kind of work a model does well, because the judgment is pattern-matching over many weak signals rather than a single hard threshold. A model can weigh declining core-feature usage against a fading champion against cooling support sentiment and surface the handful of accounts where the pattern actually rhymes with past churn — not to decide their fate, but to route them to a person while there's still time to act. The machine reads the volume; the human handles the relationship. Same division of labor that makes any triage worth doing.

The signals are already in the work

The reason most teams settle for the vanity dashboard isn't that they prefer lagging indicators — it's that the leading ones are scattered across tools that don't talk to each other. Feature usage in one system, support sentiment in another, champion activity nowhere in particular, renewal dates in a spreadsheet. Assembling a predictive score from that mess is a quarter-long data-plumbing project, so teams reach for the metrics that are easy to get, which are precisely the ones that arrive too late.

When the work itself lives on one schema — the requests, the activity, the completion events, the support threads, all in one record with an audit trail — the leading indicators stop being a plumbing project and become a query. Planoda's insights draw from the same events the work already generates, and the same triage discipline that routes inbound requests can be pointed at accounts: surface the few showing the early pattern, route them to a human, log what was done. A health score built on the live record of the work, firing early enough to matter, is the version that turns the comforting green dashboard into something that actually saves the customer.