The first version of our own customer health score was wrong. It ranked one of our most at-risk accounts as an 82 the week before they gave notice. We had over-weighted support-ticket volume — low tickets meant "healthy," it turned out, only if you ignored the fact that nobody senior had emailed them in two months. Lesson learned, model rewritten.
A customer health score is a single number — usually 0 to 100 — that summarises how well a client relationship is going. It exists so leaders can look at a portfolio and instantly see which accounts are drifting, which are thriving and which need a phone call today. That's the theory. In practice most health scores are colouring exercises no-one trusts. This piece is about the difference.
What it's actually for
A useful score does three jobs at once:
- Rank a portfolio so leaders can focus attention.
- Predict churn and expansion before they show up in revenue.
- Explain itself, so the account owner knows what to fix.
A score that ranks but doesn't explain is a mystery. A score that explains but doesn't predict is a lagging report. All three, or don't bother.
Why the common designs fail
They only score the product
Software vendors count feature usage. Services firms don't have that lever. Even for SaaS, product usage misses the whole layer of executive sponsorship, contract friction and competitive threat that actually drives renewal decisions.
They rely on manual "green / amber / red" flags
A colour set by the account manager whose bonus depends on it will be green. Every time. If the signal requires the person being measured to input it, the signal is gone before it starts.
They mash everything into one weighted sum
A single weighted sum hides tradeoffs. An account can score 80 with a dying relationship and a happy support ticket, and the number reveals nothing. Break the score into components and show every one.
A design that works for services firms
The model we ended up with — after that first embarrassing miss and a few smaller ones — has five components, each 0–100, then aggregated:
- Relationship health — strength and recency of communications with the buying committee, weighted by seniority.
- Stakeholder coverage — how many roles on the buying committee have an active relationship on your side.
- Delivery signal — support ticket volume, escalations, CSAT, project milestones hit or missed.
- Commercial signal — outstanding invoices, contract renewal proximity, scope changes.
- Strategic signal — champion job changes, competitor mentions in meeting transcripts, tech-stack changes.
Show all five. Roll them up with weights, but never only show the roll-up. That was our specific mistake for the first few months — hiding the components because "leaders want one number." Leaders want one number until it's wrong once, and then they want the five.
Where the data comes from
Almost all of it exists already in systems you own: Microsoft 365 or Google Workspace for relationships, your ticketing system for delivery, your invoicing platform for commercial, your meeting notes for strategic. You do not need surveys. Surveys tell you about clients who reply to surveys.
How to roll it out
- Compute silently for a month. Do not share the scores yet. Look at the extremes and check that the low-scoring accounts really are the ones you'd worry about, and the high-scoring ones really are your reference clients. If they're not, the model is broken; fix it before anyone else sees it.
- Share within the leadership team. Use it as an agenda for weekly portfolio reviews.
- Share with account owners. Only once trust is established — and always with the components visible, never just the roll-up.
- Never tie it to compensation. The moment you do, the inputs get gamed and the score dies.
A working definition
A customer health score is a leading indicator built from data your team doesn't have to enter, weighted across relationship, delivery, commercial and strategic signals, always broken down into its components, and used to focus human attention — not to replace it.