Every Customer Success team eventually builds a health score. Far fewer build one that changes what anybody does on a Monday morning. The difference is not the formula — it is whether the number is trusted enough to move budget, headcount and attention.
If you are the person who has to defend the retention number, here is what to demand from a health score before you let it into your reporting.
Most scores fail for the same three reasons
They live where the team does not work, so nobody sees them until a QBR. They cannot explain themselves, so the first CSM who disagrees quietly ignores them. And they are the same for every customer, so a 40-seat SMB and a 2,000-seat enterprise get judged by the same bar and neither reading is right.
None of that is a data problem. It is a design problem, and it is yours to set the standard for.
A score earns its place when it survives a follow-up question
The test is simple. When you present a portfolio view and someone asks "why is this account amber?", the answer has to be one sentence, and it has to be the same sentence the account owner would give.
If your team cannot explain a score in one sentence without opening a spreadsheet, the score is not ready to appear in a board deck.
Build it from what you already have
Health score projects stall when they turn into data projects. The wish list arrives — product telemetry nobody pipes into the CRM, survey data nobody collects — and six months later there is still no score.
Start with what is already on the record and reliable: how recently anyone talked to the customer, what they actually use, what their contract says, and what support looks like. An imperfect signal that covers every account ranks your portfolio correctly. A perfect signal that covers a third of it does not.
One standard, or you cannot manage the team
A score's real work is not telling you about one customer. It is making two accounts, two segments and two CSMs comparable. Without a shared definition of healthy, you cannot tell whether a book of business is in trouble or whether its owner is simply more pessimistic than the person next to them.
That is also the fastest coaching tool you will get: when the standard is explicit, the conversation moves from "how do you feel about this account" to "this account has been amber for five weeks — what has been tried".
Segment before you score
Enterprise customers go quiet between quarterly reviews and are perfectly healthy. SMB customers who go quiet for three weeks are usually gone. The same model cannot serve both, and forcing it to is how teams learn to distrust the number.
The score has to start work, not a meeting
A number that only gets read is overhead. A number that opens a task, notifies an owner or triggers an executive check-in is a system. The moment health crosses a line, something should happen without anyone scheduling a discussion about whether something should happen.
Five questions to ask before you approve the model
- Which decision will this number change, and who makes it?
- Can the account owner explain any score in one sentence?
- Does it apply the same standard across every CSM's book?
- Does a segment get judged by a bar that fits it?
- What happens automatically when an account crosses a threshold?
If the answers are clear, the score will show up in your forecast within a quarter. If they are not, you will have another dashboard nobody opens.