Back to playbooks

Check the Facts Before a Denial

Use this playbook when a flag, score, mismatch, or anomaly could trigger denial, closure, disenrollment, or another materially consequential adverse action.

The point is simple: one automated signal should not become a real-world consequence by itself.

What problem this solves

The fastest harms often happen before appeals ever start. Action lands first.

If a single mismatch, risk score, fraud flag, or incomplete-data signal can trigger a high-stakes adverse action, the system has set the threshold too low.

What counts as corroboration

Corroboration means an independent second basis that is reviewable, documented, and not merely a restatement of the first signal.

Operating rule

Require corroboration before any materially consequential adverse action unless a narrow emergency exception is defined in advance.

Examples:

• a utilization alert should not stand in for a clinical review by itself
• a data mismatch should not stand in for substantive ineligibility by itself
• a fraud flag should not stand in for a final closure decision by itself

Emergency exception

Emergency-first action should be rare, narrow, and reviewable.

If action happens before corroboration:

• record the emergency basis
• set a fast post-hoc review clock
• require prompt corroboration
• reverse the action if corroboration fails

Metrics and tripwires

Track:

• share of adverse actions taken with corroboration
• emergency exception rate
• post-hoc reversal rate
• reviewer time per flagged case

Tripwires:

• adverse actions routinely proceed on one signal alone
• emergency exception use becomes routine
• reviewers cannot explain what counted as corroboration
• reversal rates show the threshold is too low

Bridge language

“Require corroboration before harm.”

“If one automated flag can still trigger the denial, the safeguard is cosmetic.”

Back to playbooks