Standing up Unified Health Scoring is a four-phase project. The phases are: design the scoring formula with input signals and weights, implement the formula in Data Cloud Calculated Insights (or Apex rollup), build threshold-driven workflows that turn the score into action, and validate the formula against actual outcomes over time. Each phase has a different stakeholder profile: clinical or success leaders define the signals, admins implement the configuration, operations owns the workflow, analytics owns the validation. Plan the project with shared milestones across all four groups.
- Design the scoring formula with input signals and weights
Work with clinical or customer-success leaders to identify the input signals that predict the outcome the score targets. For Health Cloud, the signals might be A1C trend, ED visits in 90 days, missed appointment count, social determinants of health, and active diagnoses. For Customer Success, login frequency, feature adoption breadth, support volume, NPS, days to renewal. Assign each signal a weight from 0 to 1 reflecting its predictive value; the weights should sum to 1. Document the rationale for each weight. Get sign-off from the business leader who owns the outcome (the chief medical officer for Health, the VP of Customer Success for the customer use case).
- Implement the formula in Data Cloud or Apex
For orgs running Data Cloud, build a Calculated Insight that computes the score as a SQL aggregation over the Unified Data Model. Schedule the insight to refresh daily (or more frequently for operationally-critical use cases). For orgs without Data Cloud, implement the formula as an Apex rollup with scheduled execution, or as a Custom Metadata-driven formula field that recalculates on record save. Test the implementation against a known sample (a patient or account where you know what the score should be) and confirm the formula produces that value. Iterate until the implementation matches the design.
- Build threshold-driven workflows
Configure thresholds on the score (Red, Yellow, Green or whatever rating system fits the use case). For each threshold, define the downstream action: assign a Care Manager Task, send an executive notification, trigger a customer success play, schedule a phone outreach, route a Case. Implement the actions through Flow, Apex, or your customer success automation tool. Test each action by setting a sample record to the relevant score and confirming the action fires. Document the threshold-to-action mapping in the operational runbook so the team understands why a given record gets a given workflow.
- Validate the formula against actual outcomes
After three to six months of scoring, validate the formula by comparing predicted outcomes against actual outcomes. For Health Cloud, run a cohort analysis: patients flagged Red, did they actually have worse 90-day outcomes (more ED visits, more hospitalizations)? For Customer Success, did at-risk accounts actually churn at higher rates? If the formula does not predict the outcome, refine the input weights or add new signals. Schedule a quarterly validation review. Document the validation results and any formula adjustments. Without the validation loop, the score drifts from reality over time and stops being useful.
- Initial scoring weights are hypotheses, not ground truth. Run quarterly validation against actual outcomes and refine weights based on what the data shows; without validation, the score drifts from reality.
- Health Cloud scoring touches PHI and is subject to HIPAA. Access logging, retention policy, and patient consent all apply. Build audit trails at the same time as the formula.
- A score without threshold-driven workflows is just a dashboard widget. The action layer (Care Manager Tasks, Customer Success plays) is what turns scoring into business outcomes.
- Data Cloud Calculated Insights run on a schedule, not real time. For operationally-critical use cases (a deteriorating patient, an escalating customer), the latency may be too long; consider supplementary real-time signals.
- Each scoring use case (Health Cloud vs Customer Success) has its own design and validation cycle. Reusing a Health Cloud score for Customer Success without rethinking the inputs almost never works.