An enterprise implements a “human-in-the-loop” system where:
- AI handles 90% of decisions autonomously
- Humans review the remaining 10% of complex cases
- But humans can only effectively review 8% of all instances per day due to capacity
If the AI flags 12% of cases as “complex” (requiring human review), what happens to system performance over time?
Answer: The system creates a growing backlog. The AI identifies 120 cases per 1,000 that require human review, but humans can only handle 80 cases per 1,000 per day. This 40-case daily deficit will accumulate, eventually forcing either system redesign or acceptance of unreviewed complex decisions.