Guides

Risk Grades

Risk grades translate numeric credit scores into human-readable letter classifications. They simplify decision-making by grouping applicants into risk tiers with clear approve/decline/review boundaries.

What Are Risk Grades?

A risk grade is a letter (or short label) assigned to an applicant based on where their total score falls within a set of predefined ranges. For example, an applicant with a score of 710 might receive a grade of “B” (Low Risk), while a score of 480 would map to “E” (Very High Risk).

Risk grades serve several purposes in production credit systems:

  • Simplified decisioning — Loan officers and automated rules can act on grades instead of raw scores
  • Policy alignment — Credit policy can define approval criteria per grade (e.g., “auto-approve A and B, manual review C, decline D and E”)
  • Monitoring — Track grade distribution over time to detect population drift
  • Reporting — Communicate risk levels to stakeholders without exposing model internals

Configuring Risk Grades

Risk grades are configured in the Calibr desktop app using the Cutoff Simulator, found in the Scorecard tab. The simulator shows the score distribution of your development sample and lets you drag cutoff boundaries to define grade ranges.

Each grade requires:

  • Grade label — A short identifier (e.g., A, B, C, D, E)
  • Min score — The lowest score included in this grade (inclusive)
  • Max score — The highest score included in this grade (inclusive)
  • Display label — A human-readable description (e.g., “Very Low Risk”)

Grade ranges must be contiguous and non-overlapping. Together they should cover the full possible score range.

Example Grade Mapping

GradeScore RangeLabelTypical Action
A720 – 900Very Low RiskAuto-approve
B660 – 719Low RiskAuto-approve
C600 – 659Medium RiskManual review
D500 – 599High RiskDecline or enhanced review
E0 – 499Very High RiskDecline

Risk Grades in the Scorecard Spec

When you deploy a scorecard with risk grades configured, they are embedded in the CalibrScorecardSpec under the risk_grades field:

json
"risk_grades": [ { "grade": "A", "min_score": 720, "max_score": 900, "label": "Very Low Risk" }, { "grade": "B", "min_score": 660, "max_score": 719, "label": "Low Risk" }, { "grade": "C", "min_score": 600, "max_score": 659, "label": "Medium Risk" }, { "grade": "D", "min_score": 500, "max_score": 599, "label": "High Risk" }, { "grade": "E", "min_score": 0, "max_score": 499, "label": "Very High Risk" } ]

Risk Grades in Scoring Responses

When risk grades are configured, every scoring response includes the matched grade:

json
{ "score": 687, "pd": 0.034, "risk_grade": "B", "risk_grade_label": "Low Risk", "details": { ... } }

If risk grades are not configured for a scorecard, the risk_grade and risk_grade_label fields are omitted from the response. The numeric score and PD are always returned regardless.

Best Practices

  • Use 4-6 grades. Fewer than 4 provides insufficient granularity; more than 6 adds complexity without meaningful differentiation.
  • Align with business policy. Set cutoff boundaries that match your institution's approval, review, and decline thresholds.
  • Monitor grade distribution. A significant shift in the proportion of applicants per grade over time can indicate population drift or data quality issues.
  • Document grade definitions. Ensure all stakeholders understand what each grade means and what action it triggers.