Health Scores and Other Useful Fictions

The customer health score promises something every account team wants: a number that turns green, yellow, or red before the customer turns unhappy. Built from usage, engagement, support history, and commercial signals, blended into a single score, it looks like an instrument panel — and that is exactly its danger. A health score is a heuristic: a deliberate simplification that trades accuracy for attention. Used as a tripwire, it is genuinely valuable. Used as a verdict, it becomes a fiction with decimal places, and teams start managing the color instead of the customer. The discipline is not in building the score. Anyone can blend metrics. The discipline is in remembering, permanently and on purpose, what a composite number cannot know.

What a score can honestly warn about

A health score earns its keep in exactly one job: directing scarce human attention. It can notice, across hundreds of accounts, the usage that quietly halved after a champion left, the logins that stopped the month a reorganization was announced, the support tickets clustering around one broken workflow. No account manager watching a spreadsheet of raw signals catches those patterns as reliably; that is a real service, and worth having. But notice what the score detected — a change worth a conversation. It did not detect that the customer is leaving, or why, or what would change their mind. Silence from a customer can mean dissatisfaction, or a busy quarter, or a product so embedded nobody thinks about it anymore. The score cannot tell these apart. Only a person, asking, can. The healthy operating rule follows directly: every color change generates a question, never a conclusion — and the system is judged not by how red it paints accounts, but by how early it starts the conversations that matter.

The false precision, and how it sets in

The decay begins innocently: the blended score becomes the reported score, the reported score becomes the target, and soon a team is celebrating a portfolio that is 4 points greener while the customers underneath it have not changed at all. Weightings chosen in a workshop acquire the authority of physics. A score of 72 is treated as a fact about a customer rather than an artifact of somebody’s spreadsheet opinions — including opinions about which signals matter that were formed two product versions and one market ago. The maintenance nobody budgets for is the audit: which signals actually preceded the churns you have already lived through, and which are decorative? Scores that are never checked against outcomes are not instruments; they are horoscopes with better formatting. Keep the score, by all means — but keep it on a leash: recalibrated against real results, displayed next to the last human conversation rather than instead of it, and introduced in every meeting with the same sentence. This number is where we look first. It is never what we know.

One more question belongs in the audit, and it is the uncomfortable one: whose signals got chosen? A score built from product usage flatters products that are used daily and undervalues the ones that matter quarterly — the compliance tool opened four times a year, each time at the moment of maximum consequence. A score built from engagement with your emails measures your marketing, not your customer’s health. Every composite quietly encodes a theory of what a good customer looks like, and the theory usually describes the customers the vendor finds easiest to serve. Naming that bias does not disqualify the score; it disciplines it. The accounts your model cannot see clearly — the seasonal, the quiet, the deliberately low-touch — deserve a named human owner precisely because the instrument is blind there. The score should route attention. It should never be allowed to decide where attention is unnecessary.

← All Insights

© 2026 Cognosco Learning. All rights reserved.