Loading patterns…
Automation handles the routine work, but pauses for a human to review, validate, or make a judgment call at critical decision points. The best of both worlds.
Rendering diagram…
Use human-in-the-loop when:
This pattern is perfect for the transition from manual to automated. Start with human review on 100% of items, then gradually reduce to exception-only review as confidence builds.
| Phase | Actor | Action |
|---|---|---|
| 1. Ingest | Automation | Receives input (document, request, data) |
| 2. Process | Automation | Applies rules, AI, calculations |
| 3. Classify | Automation | Determines: auto-approve or needs review? |
| 4. Review | Human | Reviews flagged items, corrects if needed |
| 5. Act | Automation | Executes the final action based on human-validated decision |
Determine which items need human review:
| Criterion | Auto-Process | Flag for Review |
|---|---|---|
| AI confidence score | > 95% | ≤ 95% |
| Dollar amount | < $1,000 | ≥ $1,000 |
| Customer type | Existing customer | New customer |
| Exception detected | No exceptions | Any exception |
| Random sample | — | 5% random for quality audit |
Give the human reviewer everything they need:
Based on the reviewer's decision:
If you're using AI/ML:
Periodically review who has access to what, confirm it's still needed, and revoke what isn't. Kills the slow-drip accumulation of stale permissions that turn into audit findings.
Record every significant action to an append-only log that can be inspected but not altered. Essential for compliance, incident response, and answering 'who changed this and when?'
Externalize complex branching logic — pricing tiers, eligibility rules, routing decisions — into a readable table that business users can maintain without a developer in the loop.