Customer health scores in the CRM: predicting churn before renewal

Most churn is visible ninety days before it happens — in support tickets nobody escalated, in a champion who changed job, in a quarterly review that was never booked. A health score turns those weak signals into a number account managers can act on.

Choose signals you actually have

A useful score uses four to six inputs, not twenty. Common ones that exist in almost every CRM:

• Engagement: days since last meaningful interaction with a decision-maker
• Support: open tickets, reopened tickets, average resolution time for this account
• Commercial: invoice delays, discount pressure, scope reductions
• Relationship: number of engaged contacts (single-threaded accounts are fragile)
• Product or service usage: logins, orders, tickets raised, shipments — whatever your delivery signal is

Scoring without pretending it is science

Assign weights, keep them visible, and validate against history. Take last year's churned accounts and check whether your formula would have flagged them. If it would not, the weights are wrong — not the concept.

A simple starting model: 100 points, minus 25 for no decision-maker contact in 60 days, minus 20 for two or more reopened tickets, minus 15 for single-threaded relationship, minus 15 for payment delays over 30 days, minus 10 for missed quarterly review.

Make the score operational

A score nobody acts on is decoration. Wire it to work:

1. Score drops below 70 → task for the account owner within 48 hours
2. Below 50 → alert to the account director and a mandatory save plan
3. Renewal within 120 days and score under 80 → renewal deal created early with risk reason
4. Score above 90 with growing usage → upsell or referral play

Reporting leadership will use

Show revenue by health band, not account count — 12 red accounts matter differently at 5k and 500k. Track band movement month over month, save rate for accounts that entered red, and churn reasons captured as structured fields rather than free text.

Common mistakes

• Scoring on activity volume instead of activity quality
• Hiding the score from the customer-facing team
• Recalculating quarterly, so the signal arrives after the decision
• No owner: a score with no accountable person changes nothing

FAQ

Can HubSpot calculate this natively?

Yes, with calculated properties and workflows, plus scoring properties for more complex weighting. Usage data comes in through integrations or a scheduled sync.

What if we do not have product usage data?

Use delivery and support signals instead: tickets, delivery milestones, response times, meeting cadence. They predict renewal risk well in services businesses.

Renewals surprising you? Book a free audit and we will build a health score from the data you already have.