The Three Dashboards That Lie

Your Customer Success dashboard has three panels. None of them predict who stays.

Satisfaction scores. Usage charts. Health scores. Those are the instruments the whole profession steers by, and in the real behavior of 1.3 million customers, all three predict almost nothing. I first wrote it down during a churn spike in 2018, staring at a wall of green accounts that were about to cancel: “Health scores don’t predict.” And underneath it: “Data is lying to us.”

I’ve made the argument before. This issue is the operational one: which panels to turn off, and exactly what to put up in their place.

Turn off the satisfaction panel. Test it the honest way — each customer’s score against that customer’s lifespan — and there is no statistically significant difference between promoters, passives, and detractors: the whole gap amounts to a few days of customer life over four years. If anything, the direction runs against the promoters. The “proof” that NPS predicts retention always compares aggregate score to aggregate performance; at the individual level, where the renewal decision actually gets made, the relationship vanishes. The whole industry ran this experiment for you: over five years, satisfaction went up 13% while retention went down 28%. Happier, and leaving faster.

The reason is simple. A satisfaction survey literally asks customers how they feel about you. But customers don’t stay because of how they feel about you — companies that are good at producing high satisfaction tend to also be good at delivering results, and it is the results, not the satisfaction, that drive retention. The correlation you think you see is borrowed from somewhere else.

Promoters, passives, detractors — no statistically significant difference in how long they stay. A satisfaction survey asks customers how they feel about you — and staying was never about how they feel about you.

Turn off the usage panel. This is the most seductive one, because activity feels like progress. It isn’t: at one company in the study, the heaviest users churned as fast as the lightest. I wrote the correction back in 2015: usage is a trailing indicator of problems, not a leading one. By the time usage drops meaningfully, the customer is already in trouble — the warning flag goes up too late. And it can’t warn you early even in principle: the early usage history of cancelled clients looks virtually identical to that of the ones who ultimately succeed. Logging in, clicking around, adopting features — activity was never the goal. The result is. And the moment you gate anything on adoption metrics, they get gamed inevitably.

Turn off the health-score panel. At a company that scored account health, the accounts its team rated “Great” churned no better than the ones rated “Poor.” A health score measures how a success team feels about an account — not whether the account got a result. And it degrades from there, because a health score is both your measurement and your target, and when we use a metric as the measurement and the target of our efforts, it defeats both purposes. In practice the score gets ticked: find the lowest-scoring driver, do the thing that turns it green, move on. Even a health score that accurately predicts churn is useless for prevention — accurate churn predictions are only useful for financial planning, and one way to make a score more accurate is to not address the risks. As long as you rely on a health score, your process will always be entirely reactive: by the time it indicates a problem, it’s actually too late. A health score is not a methodology.

A health score measures how your success team feels about an account, not whether the account got a result. Accounts rated “Great” churn no better than accounts rated “Poor” — and the score can’t be fixed, because any score used as a target gets gamed.

So what goes up instead? Three panels. Here is each one, and how to build it.

Panel one: Fit — read it at the point of sale. Fit is decided before a customer ever logs in: are they the kind of buyer the product actually serves? In the study, right-fit customers bond almost completely — a bonded core around 92% — while wrong-fit customers collapse, most gone within three years, no core at all. Same product, same team, opposite outcomes, set before anyone used anything. The panel is a fit score on every account at signature, built from the factors that survive the data. The strongest is commitment: put company size and deal size head to head and the commitment signal survives while raw size washes out — serious customers stay longer, not bigger ones. Even pricing reads as fit: a discount used to close a wavering buyer tracks with shorter loyalty, while a discount traded for a longer commitment tracks with longer. And you don’t have to brainstorm what “right fit” means: your ideal customer profile is usually not a hypothesis — it’s a fact sitting in your own retention data.

Panel two: Results — a measured record, per account. Not a gut call. A record of whether each customer reached the result the product was hired to deliver. Nothing in the study divides customers more: the ones with a measured result formed a ~99% bonded core and essentially stopped leaving; the ones without had no core at all. It even works on the wrong-fit customers — among accounts that weren’t ideal-fit, the ones who still reached a result retained 99% versus 70%. In my own benchmark, customers who achieve measurable results stay six times longer. The habit that runs this panel is the keystone habit of client success: measure and materialize — always be measuring the client’s own success metrics and proactively reporting the results back. There is a simple litmus test for whether a company actually operates on success rather than happiness: does it measure customer results at all? Most don’t. That’s the panel.

Panel three: The bonded fraction — the scoreboard. Follow your customers from signup and plot the share still with you. In a healthy base the curve falls, then flattens onto a floor. The height of that floor is the share of customers who are bonded — the customers who have actually stopped leaving. The median company bonds about 1 in 14 customers for good — a bonded fraction of 7%. The top-quartile company holds 65%. And it’s the one retention number that can’t be faked by a good quarter or a long contract. This panel is the scoreboard the other two exist to move: fit and results are the levers, the bonded fraction is where you watch them land.

One thing to square, because careful readers will have noticed: I’ve said the median company bonds about 7% of its customers, and I’ve also said the study’s pooled retention curve levels off at roughly a third. Both are true. Pool all 1.3 million customers onto one curve and a few very large customer bases carry it — and those bases have high floors. The market’s aggregate is carried by its winners. That gap is the whole reason this panel exists: no pooled number, mine included, tells you the height of your own floor.

Replace the dashboard with three panels: fit, read at the point of sale; results, measured on every account; and the bonded fraction — the share of your customers who have actually stopped leaving. It’s the one retention number that can’t be faked.

One rule for running the new panels: when a signal fires, it’s red — not yellow — from day one, reviewed on a fixed weekly cadence and tracked through to completion. Yellow is how a dashboard learns to lie again.

Customer Success built its instruments on the three signals that don’t predict loyalty and looked past the three that do. The fix is not a better survey or a smarter health formula. Turn the panels off. Score fit at the sale. Measure the result on every account. Watch the bonded fraction. That’s the whole dashboard.