An account director at a general contractor is preparing for a quarterly review with a hospital owner. The open-issues log sits in Procore, two invoices are past terms in the billing system and the notes in the customer relationship management (CRM) system show the owner's facilities lead stopped joining calls last month. Nobody has lined those facts up, so the review opens with a friendly recap while the owner shortlists another builder for the next wing.
A customer health score is a composite metric that predicts whether a client account is likely to renew, bring you repeat work, expand or leave. It rolls adoption, engagement, support activity, feedback, payment history and relationship signals into one number, so account owners know which relationships need attention before the next renewal or pursuit.
Moments like the project handover often decide whether an owner calls you again, and a stale score misses them. This guide gives you a sample scoring model you can adapt, a calibration method and a look at how AI agents keep the score current across connected systems.
What Is a Customer Health Score?
A customer health score combines several account signals into one value that predicts renewal, expansion or churn. No single signal tells the full story, so the score weighs them together and points the account owner toward the accounts worth investigating first.
The score earns its place in three ways:
Early warning: Slower payments or falling owner participation show up months before a renewal conversation, while there is still time to act.
Prioritized attention: The account whose score dropped sharply gets reviewed before a stable account with a louder sponsor.
Repeat-work signals: A healthy account with active future plans is a candidate for the next pursuit or service expansion.
Keep relationship health separate from the performance of a single project. One difficult job needs context before it changes how you rate the whole account. Property teams apply the same thinking when they track tenant warning signs ahead of a lease renewal.
A Sample Customer Health Score Model for Built-World Accounts
Normalize each input to a 0-100 scale, with higher values meaning healthier conditions, then multiply each one by its weight and add the results. Treat the model below as an example to adapt. It is not a universal standard or a validated benchmark for the built world.
Input | Signals to collect | Example weight | Why it is included |
|---|---|---|---|
Adoption | Use of the service or platform, or progress toward agreed outcomes | 30% | Shows whether the account is getting value |
Engagement | Response rates, meeting attendance, logins where relevant | 20% | Reveals changes in participation |
Support activity | Issue volume, severity, resolution time and tone | 15% | Captures unresolved friction |
Feedback | Net Promoter Score (NPS), customer satisfaction (CSAT) scores and direct owner feedback | 15% | Adds the account's own view |
Commercial signals | Payment history, contract changes, time to renewal | 15% | Connects health to money and contract terms |
Relationship signals | Champion changes, executive sponsor activity | 5% | Flags dependence on one contact |
Customer Health Score = (Adoption × 0.30) + (Engagement × 0.20) + (Support × 0.15) + (Feedback × 0.15) + (Commercial × 0.15) + (Relationship × 0.05)
For a project-based client, swap in measurable inputs such as payment days against agreed terms, overdue owner decisions, recurring issues and completed commitments. Payment timeliness ties directly to construction cost tracking, so pull it from the same records your finance team already trusts. Count each behavior once, in adoption or engagement but never both.
Example band | Score | Action it triggers |
|---|---|---|
Healthy | 80-100 | Confirm the value delivered and look for repeat-work opportunities |
At risk | 60-79 | Account owner reviews the changing signals and plans targeted outreach |
Critical | 0-59 | Escalate for a relationship review and agree a recovery plan |
A critical issue still needs review when the weighted average looks healthy. Show missing or stale inputs as gaps, since treating them as positive signals inflates the score.
How to Build and Calibrate Customer Health Scoring
Customer health scoring works when the inputs are trustworthy and the weights reflect your own outcomes. Build and calibrate the model in this order:
Inventory the signals: List each input, the system it comes from, who owns it and how often it refreshes.
Define healthy and unhealthy values: Set thresholds per account segment, since an owner on a single project behaves differently from a portfolio client.
Normalize the inputs: Convert every metric to the same 0-100 scale before applying weights.
Pull commitments from the source: Confirm contract terms, amendments and promised review cadences in the governing documents, using a contract review process where terms are complex.
Validate against past outcomes: Compare the scores of retained and lost accounts, then inspect false alarms and missed risks. A 2026 churn-prediction study found contract type and tenure among the strongest drivers of predicted churn in its telecom dataset, which shows why weights should come from your own history.
Review on a schedule: Retest the weights at least quarterly and after any major change in how you serve clients.
Keep the example weights only if your results justify them.
How AI Agents Keep an AI Customer Health Score Current
Manual calculation goes stale between reviews. An AI customer health score stays current when AI agents handle the gathering and comparison work on a defined schedule:
Assemble signals: Agents can pull the selected inputs from connected systems such as Procore project records, HubSpot CRM data, billing and project files.
Recalculate on new data: Agents can apply the agreed scoring rules whenever source records update and compare the result with the previous score.
Explain the change: On a material drop or band change, agents can list which inputs moved and by how much.
Route the next step: Agents can send an alert or review task to the account owner with the evidence attached.
The Technology & Services Industry Association (TSIA) describes AI that can monitor customer health continuously and trigger outreach, and the same pattern applies to built-world accounts. Recurring project issues are often the earliest signal. Datagrid's Change Analyser Agent can analyze requests for information (RFIs), nonconformance reports (NCRs) and field changes for patterns, root causes and cumulative impact across an owner's projects, which pairs well with disciplined RFI management.
Agents depend on clean, permissioned connections, and some systems sync on a delay. Keep missing records and late updates visible, define which record governs each input, and test the calculation and alert routing before relying on them. The account owner still decides whether a pattern reflects relationship risk or an expected project phase.
What Tools Automatically Track Customer Health Scores
Three categories of tools track customer health scores automatically, and each suits a different setup:
Customer success platforms: Best when you want dedicated scorecards, playbooks and account workflows in one place.
CRM add-ons: Best when account records and tasks already live in your CRM and you want scores beside them.
AI agents across existing systems: Best when signals sit in project, billing, support and relationship systems that no single platform covers.
Every approach needs consistent account identifiers and agreed scoring rules. Pick the one that matches where your records already live.
Keep Account Health Current With Datagrid
Datagrid's AI agents can keep the inputs behind your health score current, so account reviews start from evidence:
Signal assembly: Agents can gather project, billing, CRM and support signals into one review.
Score recalculation: Agents can apply your scoring rules when source records change and flag band changes.
Change explanations: Agents can show which inputs moved and link to the records behind them.
Recurring issue analysis: The Change Analyser Agent can surface repeated RFIs, NCRs and field changes across an owner's projects.
Conflict flags: Agents can flag disagreements between project records, billing and CRM notes for the account owner.
Account owners keep the relationship judgment, the recovery plan and every conversation with the client. Get started with Datagrid by connecting the systems behind one at-risk account and comparing the score with your last manual review.
Frequently Asked Questions About Customer Health Scores
What Is a Good Customer Health Score?
In the example model, 80-100 is healthy, 60-79 is at risk and 0-59 is critical. Set your own bands by checking which scores preceded renewals, repeat work and lost accounts in your history.
How Often Should a Customer Health Score Update?
Update it whenever a source record changes, or at least as often as your systems sync. Quarterly manual updates miss slow declines that build over several months.
Can a Healthy Score Hide a Critical Issue?
Yes. A weighted average can mask one severe problem, such as an unresolved safety or payment dispute. Review critical issues directly, whatever the overall score says.
