Deviation Intelligence
The core product innovation — classifying overrides in real time
The system classifies overrides in near real time and applies selective intervention — preserve agent autonomy on corrective deviations, add friction only for cost-risk and policy-risk subsets, and continuously improve from outcomes. This is the product.
Deviation Taxonomy
Five types of override, each with a distinct intervention strategy.
Corrective Override
Tier LAgent corrects a weak copilot suggestion — wrong policy cited, missing context, or tone mismatch.
Empathy-Within-Policy
Tier LAgent uses judgment while staying inside policy and cost guardrails — softer tone, extra explanation.
Cost-Risk Override
Tier M–HCompensation likely above tolerance or precedent band for this case type.
Policy-Risk Override
Tier H–CPossible breach of mandatory process, safety, or compliance boundary.
Quality-Risk Closure
Post-auditFast closure with weak resolution quality — short interaction, no confirmation, unresolved signals.
Risk Scoring Model
Weighted multi-factor score, range 0–21. Computed in <500ms.
From policy tier: 0 = no match, 1 = advisory, 3 = mandatory, 5 = zero-tolerance
Normalized: (proposed_compensation − market_median) / threshold_band
From upstream fraud detection: 0 = no signal, 5 = confirmed indicators
Agent's high-risk deviation count in trailing 30 days
|copilot_confidence − agent_action_distance|
Intervention Tiers
Risk score maps to four intervention levels. Friction is concentrated on the ~10% that matters.
Learning Loop
Five feedback signals that continuously improve the system.
Calibration Cadence
High-risk deviation review; threshold adjustment proposals; FP/FN rates
Suggestion quality review; approved override batch promotion
Full threshold and policy mapping update with compliance sign-off
Model retrain evaluation; self-serve graduation review; cross-market comparison
Classifier Error Handling
The classifier will make mistakes. Each error type has a UX recovery path and systemic correction.
Design principle: Err toward Tier M (reason tag) rather than Tier H (block) when uncertain. A reason tag costs seconds; a false block costs minutes and trust.
Agent frustrated; resolution delayed; supervisor queue noise
Agent sees risk rationale inline; can submit structured appeal; appeal auto-logged for calibration
Weekly calibration: FP rate tracked per rule; rules with FP >20% flagged for threshold adjustment
Risk leaks through; caught only at post-audit
Post-resolution audit catches pattern; agent receives coaching feedback
Audit-flagged misses fed into calibration cycle; persistent misses trigger rule additions
Uncertain intervention — defaults conservatively
Defaults to Tier M (reason tag required); does not block agent
Ambiguous cases reviewed weekly; labeled and added to training set