Human-in-the-Loop

Escalate the right document decisions to people without turning every workflow into manual work.

What This Means in M3 Forge

Human-in-the-Loop in M3 Forge is a first-class workflow primitive, not an afterthought. Reviewers can:

Approve a gated business decision
Correct extracted or generated data
Route a case to the right downstream path

These actions happen inside a tracked request lifecycle with priorities, SLAs, and auditability.

Review Patterns

Pattern	Reviewer action	Typical use case
Approval	Approve or reject continuation	Payment release, policy exception, contract decision
Correction	Fix extracted or generated values	Low-confidence OCR or field extraction
Routing	Choose the next workflow path	Escalate to legal, underwriting, or fraud review

Example: Confidence-Based Extraction Review

In this pattern, the team keeps the fast path automatic while preserving human control over the exceptions that matter.

What Reviewers Get

Shared review queue with status and priority
Review panel for approval, routing, or field correction
SLA tracking and timeout behavior
Audit history for what changed, who changed it, and when

Example Operating Model

For a payer-ops or claims team:

The extractor processes inbound forms automatically.
Missing, ambiguous, or low-confidence fields become HITL correction requests.
High-value claims trigger an approval request before release.
Escalated cases route to a specialist queue.

This keeps the humans focused on ambiguity, policy, and risk instead of rekeying everything.

Why This Matters

Improves trust in automation for regulated workflows
Gives operations teams explicit control over exceptions
Feeds correction data back into future model improvement
Supports selective, measurable oversight instead of blanket manual review

The strongest HITL systems are selective. If everything goes to review, you do not have automation. If nothing can be reviewed, you do not have operational trust.