Protocol
Reduce batch size when feedback delay dominates
Smaller batches can trade setup overhead for earlier information.
When it fits
- When large batches create long intervals before errors or preference mismatches are discovered.
When to avoid it
- Queueing equations have specific assumptions. Use them for directional reasoning and measurement design, not as exact forecasts for arbitrary multi-stage knowledge work.
Why it matters
Reduce batch size experimentally where the cost of late learning is high, then measure setup cost, queue time, defect detection and coordination overhead.
Steps
- Name the late-feedback cost.
- Choose a smaller reversible batch.
- Measure setup and coordination overhead.
- Compare time-to-feedback and rework.
An example
Instead of migrating 5,000 records before reconciliation, process 500-record batches so mapping defects surface earlier.
Check your result
The smaller batch improves feedback timing enough to justify its extra overhead.
Keep this limit in mind
- Queueing equations have specific assumptions. Use them for directional reasoning and measurement design, not as exact forecasts for arbitrary multi-stage knowledge work.
Connected ideas
Use beforeBuffer variability where it is cheaper to absorb
Evidence and sources
Supports
Little's Law relates long-run average work in a stable system to arrival/throughput rate and average time in system as L = λW under stated mathematical conditions.
The relation does not identify root cause, queue discipline or the correct intervention; workplace applications must respect the assumptions.
A Proof for the Queuing Formula: L = λW · Abstract