03 / Interactive explanation · synthetic data
More workers. Same bottleneck?
Change the capacity of each stage. Watch what happens to completed tickets and the work waiting behind them. These are idealised rates, not operational forecasts.
Set the system
Capacity is only useful downstream
Human review limits throughput.
Predict before you click. In the starting scenario, will doubling AI workers increase completed tickets per hour?
Still 20/hour. Four workers can draft 80/hour, but only 40 arrive and one reviewer can approve 20. The remaining 20 tickets/hour join the backlog.
The arithmetic and its limits
agent capacity = workers × 20
review capacity = reviewers × 60 ÷ review minutes
completed = min(arrivals, agent capacity, review capacity)
backlog growth = max(0, arrivals − completed)
We assume continuous, steady flow; every ticket needs review; no starting backlog, breaks, rework or handoff delay; and independent workers and reviewers. The bars share a scale that expands to the largest rate. Fractional outputs are average rates. Existing backlogs and waiting-time variation require a richer model. These values are not measurements.