Human-in-the-Loop Cost
Human-in-the-loop cost is the approval labor added to an AI workflow when some runs pause for a person to review evidence and allow, edit, or reject an action.
Question
How much does human approval add to each AI workflow run?
Quick answer
Formula: approval_cost_per_run = approval_rate * approvals_per_reviewed_run * review_minutes_per_approval / 60 * reviewer_cost_per_hour
Formula: all_in_cost_per_run = base_ai_cost_per_run + approval_cost_per_run
- Assumption: approval rate is the share of all runs that reach at least one human gate.
- Assumption: loaded hourly cost includes the reviewer labor cost relevant to the decision.
- Assumption: the AI-assisted and fully human baselines complete the same unit of work.
Example: reviewing 25% of runs once for three minutes at $65 per hour adds about $0.81 to every run. A $0.25 pre-review AI workflow therefore costs about $1.06 all-in.
Approval Pattern Comparison
| Pattern | Input shape | Economic effect |
|---|---|---|
| Always review | Approval rate is 100% | Every minute of review is added to every run |
| Sample review | A fixed share of runs reaches one gate | Labor falls in direct proportion to the sample rate |
| Exception review | Only flagged runs reach review | Low approval rate can still be costly when review time or specialist labor is high |
| Multiple gates | Approvals per reviewed run is greater than one | Decision count compounds review labor on each reviewed run |
Measure the Inputs
Use workflow traces to count runs reaching approval and separate that rate from the number of decisions on a reviewed run. Measure hands-on review time rather than elapsed queue time. Use loaded reviewer cost so the result can be compared with the fully human baseline on the same basis.
Where This Fits
Start with Agent Run Cost or Tool Use Cost for the automated layer. Add approval labor with the Human-in-the-Loop Cost Calculator, then use Break-even Price for the commercial threshold.
Implementation References
Approval and pause-resume workflows are documented in the OpenAI Agents SDK, Microsoft Agent Framework, Amazon Bedrock Agents, and LangChain.
Related reads
Open companion tools: Human-in-the-Loop Cost, Agent Run Cost, Tool Use Cost
Price the full workflow: Break-even Price
Related reads: Agent Run Cost, Tool Use Cost, How Much Does an AI Agent Cost?
Run the Human-in-the-Loop Cost Calculator
Model approval labor, reviewer capacity, and all-in workflow cost from explicit inputs.
Open Human-in-the-Loop Cost