ForgeSDLC
Navigate
Home
Discover ForgeSDLC (101)
Practice (201)
Master (301)

This page is part of the ForgeSDLC knowledge base — an AI-assisted, human-directed methodology for taking product work from concept to production. For the core operating model and vocabulary, see Forge SDLC overview and What is ForgeSDLC?.

Customer health scoring and risk detection

Overview

Customer health scoring turns scattered signals — usage, support, sentiment, billing, and outcomes — into a prioritized view of risk and opportunity. It supports proactive customer management: who needs attention now, with what playbook, and whether interventions work. This guide covers components, methods, tiers, and implementation — without prescribing a single vendor or model.


Health score components

Component Typical signals
Product engagement DAU / WAU / MAU, depth (breadth of features, key events), recency and frequency, seat utilization
Support health Ticket volume and trend, severity / priority mix, reopen rate, CSAT / CES after resolution
Relationship health NPS or pulse surveys, executive sponsor engagement, QBR attendance, stakeholder coverage
Financial health Payment status, expansion vs. contraction signals, contract timeline and renewal window
Outcome achievement Success plan milestones, implementation checklist completion, stated goal progress

Weight components by what predicts your outcomes (retention, expansion, advocacy) — not by what is easiest to measure.


Health score architecture

flowchart TB subgraph sources["Data sources"] P[Product analytics] S[Support / CRM] B[Billing] V[VoC / surveys] O[Outcomes / success plans] end subgraph engine["Scoring engine"] N[Normalize & join] R[Rules / weights / model] SC[Account score] end subgraph output["Outputs"] T[Risk tiers] A[Alerts & playbooks] D[Dashboards & reviews] end sources --> N --> R --> SC SC --> T T --> A SC --> D

Scoring methodology comparison

Method Pros Cons Data requirements
Weighted average Transparent; easy to explain to CSMs and leadership May miss nonlinear interactions Clean metrics per component; calibration against outcomes
Machine learning Can capture complex patterns “Black box” risk; governance harder Historical churn/expand labels; feature store discipline
Rule-based Fast to ship; aligns to known failure modes Brittle if product/market shifts Documented business rules; regular review
Hybrid Rules for known risks + model for edge cases More operational complexity Both rule definitions and labeled outcomes

Start simple where culture and data maturity are low; add sophistication when actions and measurement keep pace.


Weight calibration

  1. Hypothesize drivers from churn interviews, support themes, and product analytics.
  2. Correlate component metrics with churn, downgrade, or negative expansion (and with positive outcomes you want to reinforce).
  3. Iterate weights quarterly or when major product changes alter behavior.
  4. Validate with CSMs: false positives erode trust; false negatives miss saves.

Document lineage: which fields feed which sub-score, refresh cadence, and known gaps.


Risk tier framework

Tier Score range (example) Characteristics Automated actions CSM actions Escalation
Healthy (green) Upper band Strong usage, low support distress, positive or neutral sentiment Positive triggers (expansion cues, advocacy asks) Proactive QBRs; growth plays As needed for strategic accounts
At-risk (yellow) Middle band Mixed signals; usage dip or support spike Alerts to owner; suggested playbooks Outreach, success plan refresh, training Manager visibility on aged yellow
Critical (red) Lower band Severe usage collapse, exec churn, payment risk, or explicit risk statements High-priority routing; exec notification rules Rescue plan, exec sponsor, commercial levers per policy VP / cross-functional war room where warranted

Thresholds should be calibrated to your base rates — a “red” that fires on 40% of accounts is not operational.


Risk detection through intervention and learning

flowchart TD D[Detect signal change] --> T[Assign / update tier] T --> I[Intervention: playbook + owner] I --> M[Monitor response] M --> L[Learn: playbook effectiveness] L --> D

Close the loop: track whether tier changes and plays correlate with improved usage, renewal, or save rate.


Leading vs. lagging indicators

Type Examples Use
Leading Login frequency drop, key feature abandonment, support escalation pattern, champion departure, stalled onboarding milestones Early warning; trigger playbooks before renewal crisis
Lagging Cancellation request, payment failure, signed non-renewal, contract expiry without engagement Confirms outcome; feeds model training and post-mortems

Health scores should emphasize leading signals for actionability; lagging signals validate and tune the model.


Health score by business model

Model Emphasis
SaaS (seat / usage) Product depth, seat activation, admin health
Marketplace Supply and demand balance, liquidity, quality / dispute signals
API platform Call volume, error rates, latency SLO adherence, key integration health
Enterprise Stakeholder map coverage, security / procurement milestones, executive engagement

One global score rarely fits; consider sub-scores by motion (e.g. product vs. relationship) with a composite for prioritization.


Implementation roadmap

Phase Focus
Phase 1 — Manual scoring Spreadsheet or CRM fields; CSM judgment + weekly review; document definitions
Phase 2 — Automated data collection Pipeline from product, support, billing; consistent account IDs; data quality checks
Phase 3 — Predictive model Labeled outcomes; validate lift over rules; governance and explainability policy
Phase 4 — Prescriptive actions Recommended next best action; experiment framework for plays

Skipping Phase 2 quality usually wastes model effort.


Technology stack

Layer Examples
CS platforms Gainsight, ChurnZero, Totango, Vitally — health scores, plays, dashboards
Custom Warehouse + dbt + reverse ETL or in-app flags; full control, higher build cost
Integration Event pipelines, CRM as system of record for account, ticketing sync
Dashboards Executive rollup, CSM workbench, manager queue — different detail density

Design dashboards for decisions, not vanity: “what do I do Monday morning?”


Anti-patterns

Anti-pattern Effect
Too many components Noise; unstable scores; CSM distrust
Equal weighting Ignores true drivers of churn
No action triggers Score is analytics theater
No benchmarks Cannot tell normal seasonality from crisis
Gaming the score Incentives drive theater metrics instead of customer outcomes

External references

  • Gainsight — Customer Success resources and health score playbooks.
  • Customer Success (Mehta et al.) — Foundational framing for CS operations and metrics.
  • ChurnZero — Blog and materials on engagement scoring and plays.
  • TSIA — Frameworks for service and success economics and maturity.

Keep project-specific customer success documentation in docs/product/customer-success/ and support playbooks in docs/operations/, not in this file.