What goes into a health score

What the number is doing

A health score compresses several signals about an account (engagement, relationship breadth, product usage, support history) into one figure so that a book of hundreds of customers can be ordered by where attention is most likely to pay off. Ordering a large book is genuinely hard without something like it, and that is the job it does well.

It follows that the score is a ranking instrument rather than a prediction about any particular account. The Health surface reflects this by binning customers into Healthy, Neutral and At-Risk rather than presenting the score as a precise quantity: the bin is what changes your behavior this week, and arguing about whether an account is a 61 or a 64 is arguing about a rounding decision inside a summary.

The signal underneath worth knowing by name

One input deserves separate attention because it behaves differently from the others: multi-threading, the number of active contacts on an account. Accounts with three or more renew more reliably, and the reason is structural rather than sentimental: a relationship resting on one person is one job change away from being no relationship at all.

That makes it the most useful cross-check on the surface. When at-risk counts rise and multi-threading falls at the same time, the honest diagnosis is that the company has lost contact breadth, and a save motion built on demonstrating product value will underperform one built on rebuilding relationships. The two responses are different work, and the metrics tell you which you are in.

• Three or more active contacts is the threshold that correlates with reliable renewal
• Multi-threading falling alongside rising at-risk means breadth, not value, is the problem
• A single-threaded account is one departure from being an unowned account

What the score cannot tell you

Two things reliably escape it. The first is anything the system cannot see: a change of sponsor discussed in a meeting nobody logged, a budget cycle nobody recorded, a competitor conversation happening entirely off your instrumentation. The second is causation. A score falls because inputs moved, and the inputs never say why.

So the working posture is to let the score choose where to look and never to let it conclude. It ranks the book; you open the accounts at the top of that ranking and find out what is actually happening. Teams that treat the number as the finding end up managing the metric, which is a well-documented way to make a metric stop being informative.

A worked example: two accounts, one score

Two accounts both land at the bottom of the Neutral band, close enough that the score cannot separate them. Opening them takes five minutes and finds two unrelated situations.

The first has strong usage, a champion who replies quickly, and a single support escalation eight weeks ago that dragged. Its score is depressed by one input that has already resolved, and the honest read is that this account is fine and the number is lagging. Nothing to do beyond noting why it looks worse than it is.

The second has flat usage, one contact, and no meaningful engagement for a quarter. Its score is depressed because everything that feeds the score is genuinely weak, and the single contact means there is nobody to ask about it. That account needs work this week, and it needs relationship work before anything else.

Same band, same approximate number, entirely different weeks of work. This is what people mean when they say a score ranks rather than diagnoses, and it is why the discipline is to let the score choose where to look and never to let it conclude.